MOX-Report No.

38/2016

The Cardiovascular System: Mathematical Modeling,

Numerical Algorithms, Clinical Applications

Quarteroni, A.; Manzoni, A.; Vergara, C.

MOX, Dipartimento di Matematica

Politecnico di Milano, Via Bonardi 9 - 20133 Milano (Italy)

mox-dmat@[Link] [Link]
 The Cardiovascular System: Mathematical Modeling,
Numerical Algorithms, Clinical Applications
A. Quarteroni1 , A. Manzoni1 , C. Vergara2

October 29, 2016

1
Chair of Modelling and Scientific Computing, École Polytechnique Fédérale de Lausanne,
Switzerland, {[Link],[Link]}@[Link]
2
MOX, Dipartimento di Matematica, Politecnico di Milano, Italy,
[Link]@[Link]

Keywords: Blood flow simulation, fluid-structure interaction, heart function, parameter

estimation, uncertainty

Abstract
Mathematical and numerical modeling of the cardiovascular system is a research
topic that has attracted a remarkable interest from the mathematical community be-
cause of the intrinsic mathematical difficulty and due to the increasing impact of
cardiovascular diseases worldwide. In this review article, we will address the two prin-
ciple components of the cardiovascular system, the arterial circulation and the heart
function. We systematically go through the complete pipeline from data imaging ac-
quisition, setting the basic physical principles, analyzing the associated mathematical
models that comprise PDEs and ODEs systems, proposing sound and efficient nu-
merical methods for their approximation, simulating both benchmark problems and
clinically inspired (driven) problems. Mathematical modeling itself features tremen-
dous challenges, due to the amazing complexity of the cardiocirculatory system, the
multiscale nature of the involved physiological processes, and the need of devising com-
putational methods that are stable, reliable, and efficient. A critical issue is about
filtering the data, identifying the parameters of mathematical models, devising opti-
mal treatments, accounting for uncertainties. For this reason, we will devote the last
part of the paper to control and inverse problems, including parameter estimation,
uncertainty quantification and the development of reduced order models that are of
paramount importance when solving problems with high complexity, that would be
out of reach otherwise.

Contents
1 Introduction 4

1
I THE ARTERIAL CIRCULATION 5

2 Basic facts on quantitative physiology 6

3 Landscape on data 10
3.1 Geometric vascular data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.1.1 Acquisition of clinical images . . . . . . . . . . . . . . . . . . . . . . 11
3.1.2 Image enhancement . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.1.3 Image segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1.4 Building the computational mesh . . . . . . . . . . . . . . . . . . . . 15
3.2 Boundary vascular data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Biological vascular data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4 Modeling blood flow 20

4.1 The fluid problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.2 Mechanical wall models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2.1 Modeling the structure as a 2D membrane . . . . . . . . . . . . . . . 24
4.3 The coupled fluid-structure interaction problem . . . . . . . . . . . . . . . . 25
4.4 The boundary issue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.4.1 Conjecturing velocity and pressure profiles . . . . . . . . . . . . . . . 27
4.4.2 Augmented formulation . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.4.3 A control-based approach . . . . . . . . . . . . . . . . . . . . . . . . 29
4.5 Geometric reduced models and multiscale approach . . . . . . . . . . . . . . 30
4.5.1 The 1D and 0D models . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.5.2 The geometric multiscale coupling . . . . . . . . . . . . . . . . . . . 34
4.6 Numerical strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.6.1 Numerical methods for the fluid problem . . . . . . . . . . . . . . . 37
4.6.2 Numerical methods for the vessel wall problem . . . . . . . . . . . . 41
4.6.3 Numerical methods for the fluid-structure interaction problem . . . 43
4.6.4 Numerical methods for defective boundary problems . . . . . . . . . 54
4.6.5 Numerical methods for the geometric reduced models and multiscale
approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

II THE HEART FUNCTION 58

5 Basic facts on quantitative physiology 58

5.1 Basic anatomy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.2 The cardiac cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.3 Electric propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.4 Mechanisms of contraction and cardiac blood fluid-dynamics . . . . . . . . 63
5.5 A brief summary of heart diseases . . . . . . . . . . . . . . . . . . . . . . . 64

2
6 Landscape on data 65
6.1 Cardiac geometric data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.2 Cardiac boundary data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.3 Cardiac biological data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

7 Modeling the heart 71

7.1 Cardiac electric activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
7.1.1 The bidomain model . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
7.1.2 Cardiac cell models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
7.1.3 Reduced continuous models: the monodomain and Eikonal equations 76
7.1.4 Boundary conditions and Purkinje network models . . . . . . . . . . 78
7.1.5 Computing the surface electrocardiogram signals . . . . . . . . . . . 79
7.1.6 Numerical discretization . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.2 Cardiac mechanics and electro-mechanics coupling . . . . . . . . . . . . . . 84
7.2.1 The continuous mechanics problems . . . . . . . . . . . . . . . . . . 84
7.2.2 The coupled electro-mechanics problem . . . . . . . . . . . . . . . . 87
7.2.3 The issue of boundary conditions for the mechanics problem . . . . 89
7.2.4 Numerical approximation . . . . . . . . . . . . . . . . . . . . . . . . 90
7.3 The ventricular fluid-dynamics . . . . . . . . . . . . . . . . . . . . . . . . . 93
7.4 Valves modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
7.4.1 Mechanical models of the leaflets . . . . . . . . . . . . . . . . . . . . 96
7.4.2 Reduced models for fluid-valve interaction . . . . . . . . . . . . . . . 97
7.4.3 Interaction between blood and leaflets . . . . . . . . . . . . . . . . . 100
7.5 The total heart integration modeling . . . . . . . . . . . . . . . . . . . . . . 103

III OPTIMIZING, CONTROLLING, ADDRESSING UNCERTAINTY,

REDUCING COMPLEXITY 105

8 Going beyond direct simulation 105

9 Control and Optimization 107

9.1 Optimality conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
9.1.1 Sensitivity approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
9.1.2 Adjoint approach and Lagrange multipliers . . . . . . . . . . . . . . 109
9.2 Numerical approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
9.2.1 Iterative methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
9.2.2 All-at-once methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
9.3 Examples of applications in cardiovascular modeling . . . . . . . . . . . . . 115
9.3.1 Optimal design of bypass grafts . . . . . . . . . . . . . . . . . . . . . 115
9.3.2 Optimal control of electrical defibrillation . . . . . . . . . . . . . . . 118

3
10 Parameter estimation from clinical data 122
10.1 Variational approach: PDE-constrained optimization . . . . . . . . . . . . . 123
10.2 Sequential approach: Kalman filter and extensions . . . . . . . . . . . . . . 125
10.2.1 The Kalman filter algorithm . . . . . . . . . . . . . . . . . . . . . . 126
10.2.2 Extended Kalman Filter . . . . . . . . . . . . . . . . . . . . . . . . . 130
10.2.3 Unscented Kalman filter . . . . . . . . . . . . . . . . . . . . . . . . . 130
10.3 Examples of applications in cardiovascular modeling . . . . . . . . . . . . . 132
10.3.1 A variational approach for estimating the tissue compliance . . . . . 135
10.3.2 A Kalman filter approach for estimating material parameters in car-
diac mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

11 Including uncertainty 141

11.1 Forward Uncertainty Quantification . . . . . . . . . . . . . . . . . . . . . . 142
11.2 Inverse Uncertainty Quantification . . . . . . . . . . . . . . . . . . . . . . . 144
11.2.1 Static problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
11.2.2 Dynamical problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
11.3 Applications to Cardiovascular Modeling . . . . . . . . . . . . . . . . . . . . 151
11.3.1 Backward uncertainty propagation in a simplified blood flow model . 154
11.3.2 Backward uncertainty propagation in cardiac electrophysiology . . . 154

12 Reduced order modeling 157

1 Introduction
This is a review paper on the mathematical and numerical modeling of the cardiovascular
system (CS), a research topic that has attracted remarkable interest from both the math-
ematical and bioengineering communities over the past 25 years. The driving motivation
for such interest is the increasing impact of cardiovascular diseases (CVD) in our lives.
According to [384], CVD are the major cause of death worldwide, leading to more than
17.3 million deaths per year, a number that is expected to grow to more than 23.6 million
by 2030. In Europe, nowadays they correspond to nearly half of all deaths (47%).
In this paper we focus on the two principal components of the CS, the arterial cir-
culation and the heart function with its electrical and mechanical activities. Geometric
complexity, lack of data to feed the mathematical models, multiphysics and multiscale na-
ture of the processes at hand, represent major challenges when trying to reproduce both
function and malfunction.
Due to its composite nature, the CS is first modeled by means of stand-alone core
components describing a single functionality, like e.g. the artery fluid-dynamics, the heart
electrical activity, the fluid dynamics in the left ventricle, etc. Each core model needs
to be properly analyzed mathematically and efficiently approximated numerically, often
by specifically devised methods. The next step is the integration of the core models into
global, coupled integrated models apt at describing a meaningful and coherent part of the
(or even the entire) CS system. This step requires the introduction of suitable coupling

4
conditions and of (novel) numerical strategies for a stable, reliable, and computationally
effective solution of the global problem.
Clinical data play a decisive role for CS models and, at the same time, they represent
a formidable challenge. Clinical radiological images (such as Computer Tomography and
Magnetic Resonance Imaging) are necessary to construct the computational domains. The
procedure of geometric reconstruction is difficult and, especially for the heart, requires ad-
vanced mathematical and numerical tools. Standard radiological images can sometimes
be useless: some cardiovascular components may have a size smaller than the spatial res-
olution featured by the imaging device (this is e.g. the case of the Purkinje network); in
other cases the elevated brightness gap between fluid and vessel wall, makes the detection
of the latter very hard. Boundary data are also difficult to obtain. When the computa-
tional domain results from an artificial truncation, specific physical quantities (e.g. fluid
velocity or pressure) should be provided at those locations of the arterial tree correspond-
ing to the artificial boundaries. However, this would require invasive measurements that
cannot be easily carried out. Finally, the huge inter- and intra-patient data variability and
uncertainty represent further sources of concern toward model calibration and validation.
In spite of all these difficulties, a wealth of models has already been successfully applied
to address both physiological and pathological instances. The aim is from one side a
better understanding of the physical and quantitative processes governing the CS, and
on the other side the opening of new frontiers in therapeutic planning and the design of
implantable devices (such as e.g. medical stents and cardiac defibrillators).
The literature about the mathematical and numerical modeling of CS is huge (as
the reader would realize by browsing this paper’s references, a tiny subset of the overall
existing ones). In the forthcoming sections we will try to provide an outlook to the main
contributions in this field. Here, among the several books, monographes, and review
papers published so far, we mention [188, 540, 482] for the circulatory system and [448,
521, 118, 468, 470] for the heart.
This review paper consists of three main parts, i.e. i) modeling the arterial circulation
(Sects. 2, 3 and 4), ii) modeling the heart function (Sects. 5, 6 and 7), and iii) solving
inverse problems and including uncertainty (Sects. 8, 9, 10 and 11). Both parts 1 and 2 are
composed by an introductory section on physiology (Sects. 2 and 5), a section describing
the available data and their use (Sects. 3 and 6), and a final section on the mathematical
and numerical modeling (Sects. 4 and 7). Regarding the third part, in an introductory
section we underline the need of going beyond a single (forward) simulation in some
applications (Sect. 8). This represents the common denominator of three topics recently
applied to cardiovascular mathematics: control and optimization (Sect. 9), parameter
estimation (Sect. 10), and uncertainty quantification (Sect. 11).
When appropriate (in particular in Sects. 4, 7, 9, 10 and 11), we report some numerical
results to highlight the effectiveness of the numerical strategies here presented. Unless
otherwise specified, all our numerical results have been obtained using the Finite Element
library LifeV, see [Link] for more details.

5
Part I
THE ARTERIAL CIRCULATION

2 Basic facts on quantitative physiology

The cardiovascular system is a close circuit that carries oxygenated blood to all the tissues
and organs of the body. Functionally, it can be regarded as made by three compartments:
the heart, the systemic and pulmonary circulations, and the microvasculature. In this
section we will recall the most important features about the physiology of the systemic
circulation characterizing the mathematical models that will be introduced later on. We
will also highlight the main peculiarities of the pulmonary circulation. Heart physiology
will be addressed in Section 5.
The systemic circulation is composed by the arteries, that carry the oxygenated blood
ejected by the left heart to the living tissues, and the veins that allow the non-oxygenated
blood to returning to the right heart. The exchange of oxygen between blood and the body
tissues occurs in the microvasculature, which in fact separates the systemic arterial tree
from the venous systems. In the pulmonary circulation, non-oxygenated blood ejected
by the right heart flows in the pulmonary arteries towards the lungs where it becomes
oxygenated and goes back to the left heart through the pulmonary veins.
Blood is composed by plasma (about 55% of its total volume) which consists of water
(about 92% of plasma volume), proteins and ions. The remaining part of blood corresponds
to the blood cells, whose 97% of volume is occupied by erythrocytes (red blood cells) that
carry the oxygen in oxygenated blood. The other cells are leukocytes (white blood cells)
and platelets. The diameter of blood cells is approximately 10−3 cm, whereas that of the
smallest arteries/veins is about 10−1 cm. This is the reason why blood in the systemic
and pulmonary circulations is often considered as Newtonian, i.e. characterized by a
linear relationship between internal forces and velocity gradients [443, 188] However, in the
smallest arteries, such as coronaries (the arteries perfusing the heart and the corresponding
veins, see Figure 1, right), or in presence of a vessel narrowing (stenosis), a non-Newtonian
blood rheology is more appropriately assumed, see, e.g., [93] and references therein.
Thanks to the heart contraction, the blood flow is pulsatile and blood is pumped into
the two circulations by means of discrete pulses with a pressure usually varying during an
heartbeat in the ranges 70−130 mmHg and 20−30 mmHg for the systemic and pulmonary
networks, respectively (1 mmHg ≃ 133.3 P a = 1333 g/(cm s2 )).
In the systemic circulation, blood first enters the aorta (the largest artery with diameter
equal to about 2.5 cm in adults, see Figure 1, left) and then flows through a network of
hundreds of branching arteries of decreasing size, reaching all the regions of the body.
Dimensions and numbers of veins are comparable with those of arteries. The waveform
of the flow rate as a function of time is characterized by different peak values when
moving downstream towards the smallest arteries. In particular, the flow rate peak value
is about 200 cm3 /s in aorta, 80 cm3 /s in the abdominal aorta, 15 cm3 /s in the carotids
(the arteries supplying blood to the brain, see Figure 1, middle), and 1 cm3 /s in coronaries

6
(corresponding to a maximum blood velocity of about 150 cm/s in aorta, 100 cm/s in the
abdominal aorta, 80 cm/s in the carotids, and 40 cm/s in coronaries). Also the shape of
the waveforms changes while moving downstream, see Figure 2, left. In particular, in the
ascending aorta, after the systolic peak the flow rate decelerates assuming null or even
negative values, whereas in the abdominal aorta and in carotids is more spread out and
always positive. In any case, we can distinguish the systolic phase, i.e. the interval of
acceleration and deceleration of blood flow, and the diastolic phase, i.e. the interval of
almost constant or negative flow 1 . A different situation occurs in coronaries, where the
peak flow rate is reached during diastole, see Figure 2, right. Coronaries are not directly
fed by the heart; indeed, blood in the proximal part of the aorta (the sinuses of Valsalva
from which coronaries originate) during diastole is allowed to enter the coronaries thanks
to the elastic response of the aorta (see below for more details).

Figure 1: Visualization of the aorta (left), carotids (middle), and coronaries (right)

In the pulmonary circulation blood first enters the pulmonary artery (diameter equal
to about 3.0 cm in adults) and then flows into another network of branching arteries of
decreasing size reaching the lungs. The waveforms and peak intensities are similar to those
of systemic arteries.
The different characteristics of blood flow in the arteries of the systemic circulation
ρ DU
result in different values of the Reynolds number Re = f µ (ρf being the blood density,
D and U characteristic vessel dimension and blood velocity, respectively, and µ the fluid
viscosity), a dimensionless quantity which quantifies the importance of the inertial terms
over the viscous ones. In particular, Re ≃ 4000 in the aorta and Re ≃ 400 in coronaries,
1
The previous definition of systole and diastole is formulated from the point of view of the arteries. An
almost equivalent definition could be given from the point of view of the heart, see Section 5.

7
Figure 2: Typical flow rate waveforms in ascending aorta, abdominal aorta and carotids
(left), and in coronaries (right)

with intermediate values when moving downstream the aorta. Thus, blood covers a range
of Reynolds numbers where both the inertial and the viscous components of the flow are
relevant. Although in the aorta Re is higher than the critical value of 2000 above which the
flow would not be laminar any longer in a straight pipe, the pulsatile nature of blood flow
does not allow fully transition to turbulence to develop. It is debated whether in aorta at
least transitional-to-turbulence effects may occur. In this respect, some authors speculate
that the helicoidal velocity pattern in aorta induced by the torsion of the heart contraction
inhibits any transition to turbulence, thus supporting the thesis that in healthy conditions
turbulence is never observed in the cardiovascular system [383]. This is not necessary the
case for some pathological conditions, such as carotid stenosis, yielding a narrowing of the
vessel lumen and an increased complexity of the geometry together with higherqReynolds
2Af
numbers, see, e.g., [2, 326, 288, 316]. As for the Womersley number W = µ (A
and f being characteristic cross-section vessel area and time frequency of the flow rate
signal, respectively), a dimensionless number quantifying the pulsatility of flow, we have
decreasing values in the systemic circulation moving downstream (W ≃ 10 in aorta, W ≃ 3
in carotids). Similar values of Re and W are found in the pulmonary arteries.
As for the veins of the systemic circulation, comparable values with respect to arteries
of flow rate, Reynolds and Womersley numbers are found, the only difference being that
the blood flow waveform is more spread out with respect to the corresponding arteries.
Another major difference is given by blood pressure values. In the arteries the range of
pressure is almost the same, independently of the location in the tree (70 − 130 mmHg),
whereas in the veins it falls down assuming a mean value of about 10 mmHg. This is due
to the big resistances that blood flow experiences at the microvasculature. The latter is
composed by thousands of arterioles and venules and billions of capillaries. The blood
velocity and vessel dimensions are here greatly reduced (about 10−1 cm/s the first one
and 10−2 cm the second ones). This means that Re is very small in comparison with the
systemic circulation so that viscous forces completely dominate the inertial ones. As a
result, the highest resistance to flow is found in microvasculature, thus provoking a big

8
decrease in the blood pressure. Since the typical dimension of capillaries is comparable
to that of erythrocites, a multiphase model looks appropriate for their mathematical de-
scriptions [156]. Finally, we observe that, unlike arteries, veins can be supplied by valves
that prevent backflow of blood and venous flow highly depends on the muscle contraction
and respiratory effects.
As observed, blood pressure assumes the same range of values (70 − 130 mmHg) along
the entire systemic arterial tree. More precisely, a negligible dissipation is experienced by
the pressure signal in large and medium sized vessels before reaching small vessels and
microvasculature. Of course, at a given instant the pressure is not constant-in-space along
the tree. Indeed, a time shift characterizes the pressure waveforms at different locations
which generate gradient pressures between proximal and distal regions facilitating the
blood movement. These spatial gradients are due to the propagating nature of the pres-
sure which is in fact a wave traveling along the arterial network. The wave speed ranges
from about 500 cm/s in the aorta to 1200 cm/s in coronaries. The presence of bifurca-
tions or high resistance regions (such as the microvasculature) produces wave reflections
propagating backward to the heart.
The propagation of a pressure wave along the vascular tree is due to the vessels com-
pliance, i.e. the ability of the vessel to distend under the forces exerted by the blood
pressure. Vessel wall displacements are quite large, reaching up the value of 10% of the
lumen diameter. This is possible thanks to the structure of the vessel walls: their to-
tal thickness is about 10% of the lumen diameter and are composed by three layers, the
intima, the media, and the adventitia. The inner part of the intima is the endothelium
(facing the blood), whereas the remaining part is composed by connective tissue. The me-
dia and the adventitia mainly have a major role in characterizing the mechanical response
of the vessel wall from the mechanical point of view. Their main structural components
are elastin and collagen. The media is also formed by smooth muscle cells which provide
the tone to the vessel wall. Elastin forms complex networks that are very distensible, pro-
viding the elasticity of the vessel wall at small strain. Instead, collagen forms stiff fibers
oriented in a helical form that provide tensile strenght at large strain. Thus, the artery
vessel wall is characterized by highly non-linear elastic properties. The quantity of elastin
and collagen decreases going downstream along the arterial network, whereas the quantity
of smooth muscle cells increases. This allows the arteries more proximal to the heart, in
particular the aorta, to be very extendible and, thanks to the high peripheral resistances
due to the elevated tone of the distal arteries and to the microvasculature, to store during
systole about 50% of the entering blood. This blood reserve is then discharged during
diastole owing to the vessel wall elastic response (windkessel effect). This is responsible
for the smoothing of the blood flow waveform discussed above going downstream along
the arterial network, which guarantees a nearly continuous peripheral blood flow and thus
an almost continuous exchange of oxygen with the tissues. Also pulmonary arteries walls
are extendible (with muscular tone increasing going downstream), their thickness being
however only about 1% of the lumen diameter.
As already observed, there is mutual exchange of energy between blood and extendible
vessel walls: the latter accumulate elastic potential energy under the forces exerted by
the blood pressure, which is then transfered to the blood as kinetic energy. From the

9
mechanical point of view, this gives rise to a fluid-structure interaction problem. This
process occurs at short time scales, proportional to the duration of a heartbeat (∼ 1 s).
Other interaction mechanisms may take place at larger time scales yielding a wall mod-
ification of vessel properties. This occurs in the case of several arterial diseases, such as
atherosclerosis and aneurysm formation. In the first case, an increased permeability of
vessel wall to lipoprotein provokes a cascade of events at the cellular level which leads to
the accumulation of fatty material in the intima, just below the endothelium, and then to
plaque formation in the media. Preferential sites of atherosclerotic plaque formation are
the carotids and the coronaries. The main complications are the partial occlusion of the
lumen with consequent (cerebral or heart) ischemia, or even the total occlusion resulting
in (cerebral or cardiac) infarction. An aneurysm consists in the dilatation of the vessel
wall with formation of a (possibly huge) bulge, mainly in aorta and cerebral arteries, due
to a loss of elastin and to the consequent remodeling of collagen, resulting in a weakening
of the arterial wall. 80-90% of ruptured abdominal aortic aneurysms and 45% of ruptured
cerebral aneurysms result in death. The role of blood fluid-dynamics has been recognized
to be crucial for the development of both these diseases [215, 27]. In particular, wall shear
stresses, i.e. the viscous/friction forces exerted by the blood on the endothelium, although
about 100 times smaller in magnitude than pressure, regulate the permeability of the wall
to lipopotrein and the loss of elastin, thus playing an important role in atherosclerosis and
aneurysm development. For both these arterial diseases, this supplementary interaction
between fluid and structure occurs at time scales of several years.
More on the physiology of the systemic and pulmonary circulations and microvascula-
ture in view of mathematical modeling is available in, e.g., [401, 478, 188].

3 Landscape on data
The ultimate and ambitious goal of mathematical models in medicine is to provide quan-
titative results for enhancing the understanding of biophysical processes and provide clin-
icians with useful indications in support of their diagnostic and therapeutic procedures.
To these aims, it is mandatory to consider data that are patient-specific (according to
the bioengineering jargon), i.e. related to real patients. Obtaining and processing patient-
specific data is a major issue which alone deserves specific review papers. Here, we provide
a brief overview of the most common techniques for acquisition and analysis of “clinical”
data. This data preprocessing is essential prior to the set up of a numerical simulation.
In this section, we address the case of data related to the arterial (or venous) circula-
tion, whereas in Section 6 we will discuss cardiac data. In arteries we have two processes
interplaying with one another, the blood flow in the vessel lumen (the region occupied by
the blood, which is referred to as the fluid domain) and the displacement of the vessel wall
(referred to as structure). We need geometric, boundary, and biological data, which are
discussed in what follows.

10
3.1 Geometric vascular data
Geometric data are necessary to build the geometry of the computational domains wherein
the differential problems are numerically solved. At the end of the geometric preprocessing
step, we should obtain the fluid computational domain for the blood fluid-dynamics prob-
lem, and the structure computational domain, for the vessel wall displacement problem.
The processing of geometric data for blood flow simulations is a major task since vessels
exhibit high morphological variability due, e.g., to the noticeable vessel tortuosity and the
presence of several bifurcations. Moreover, in unhealthy cases, this variability is further
emphasized, because of the possible presence of calcifications, stenoses, aneurysms, or even
prostheses (such as stents).
The geometric preprocessing consists of the following steps, which are usually per-
formed in sequence [11, 10]: acquisition of clinical images, image enhancement, image
segmentation, and generation of the computational mesh. These items are addressed be-
low.

3.1.1 Acquisition of clinical images

Angiography is an imaging technique successfully used to “identify” the vessel lumen.
It exploits the property that a liquid inside the vessel appears brighter than the vessel
wall and the surrounding tissue. Angiographies are usually acquired as two dimensional
(2D) images, corresponding to different slices of the domain of interest, but also three
dimensional (3D) acquisitions of volumes are possible.
One of the most common techniques to obtain an angiography is X-ray imaging, based
on the projection of X-ray beams through the body onto suitable screens, and on the
contrast produced on the 2D image by the different absorption properties of the body
structures. To highlight the vessel lumen, a radio-opaque dye is inserted into the blood
stream through the arterial system. To reconstruct tortuous geometries, a rotational
angiography (RA) is performed, where X-ray sources and detectors are rapidly rotated
around the patient, allowing one to acquire many projections within few seconds. The
excellent spatial resolution of projection angiography (about 0.2 mm, 0.4 mm for RA)
makes this technique the gold standard for most vascular imaging applications. Another X-
ray angiography technique, widely used for blood flow simulation, is based on the computed
tomography (CT) technology, where multiple X-ray sources and detectors are rotated
rapidly around the patient, allowing one to acquire 3D images with an excellent spatial
resolution (less than 1 mm, computed tomography angiography, CTA). Unlike projection
angiography, another advantage of CTA is the possibility of using intravenous instead of
arterial injections. Recently, temporally-resolved CTA imaging (4D-CTA) became feasible.
This allows one to obtain several (15-20) 3D images during an heartbeat.
Difficulties may arise for the presence of metal artifacts due to metallic prostheses such
as peacemakers, which results in streaks on the images obscuring anatomical details, see
e.g. [493, 167] for possible mathematical treatments.
Another, widely used technique to obtain angiographies is Magnetic Resonance (MR),
based on the different decay rate featured by body structures after the application of a

11
radiofrequency (RF) energy. This is called Magnetic Resonance Angiography (MRA). The
generated contrast in the images could be tuned by selecting different RF stimuli. This
allows MRA to be suitably indicated to detect soft tissues. Another advantage of MRA is
that angiography could be generated without using exogenous agents. However, usually
an intravenous injection of a paramagnetic contrast agent is used to improve the blood
signal and reduce the acquisition time (Contrast-Enhanced (CE)-MRA).
Finally, we mention Ultrasound (US) imaging, based on the reflections of high fre-
quency sound waves (few M Hz) transmitted into the body. US is the least expensive and
invasive among the techniques here discussed, and allows one for real-time acquisition of
2D images. In contrast, the spatial resolution is the poorest one. Recently, also 3D images
could be acquired by means of 3D US, by reconstructing a 3D volume from 2D slices.
On the other side, only a few techniques allow one to obtain vessel wall images nowa-
days. Among them, we cite Black Blood (BB)-MRI, by which the vessel wall and the
surrounding tissue can be viewed as well, and Intravascular Ultrasound (IVUS), which
is however very invasive since the transducer is placed directly in the artery (typically a
coronary) via a catheter.
No matter which technique is being used, from a mathematical standpoint we can
assume that at the end of the acquisition step we obtain a vector I clin , whose j − th
component, Ijclin , corresponds to the intensity of the image at the point xj in a gray-
scale representing the contrast generated by the imaging technique. The collection of the
points xj , j = 1, . . . , N clin , assembles the lattice Lclin , N clin being the total numbers of
acquisition points (in practice, either pixels or voxels) where the image contrast has been
evaluated. Here and in what follows, a lattice is a simple collection of points determined by
the points coordinates. It may be useful to associate to the image intensity vector I clin a
corresponding image intensity (scalar) function, that is typically obtained by interpolation.
We will denote it by I clin (x).

3.1.2 Image enhancement

Medical images are often affected by noise and artifacts that may interfere with the quality
of the final results of the preprocessing step. Thus, prior to the reconstruction of the 3D
geometry, an imaging enhancement is usually performed.
A popular enhancement technique is resampling, consisting in suitably changing the
resolution of the images in one or more directions. In practice, an interpolation of im-
age intensity values I clin onto a different (more refined) lattice is performed. The most
commonly used methods are constant interpolation, first order composite Lagrangian in-
terpolation, B-spline [555], and windowed sinc interpolation.
The noise in the medical images may be due to thermal effects in the signal processing
electronics or to other undesired sources. Reduction of noise could be obtained by means
of a smoothing filter which does not require any prior information about the nature of the
noise and has a regularizing effect on the image. This technique is the most commonly
used both for CT and MR images. A very popular filter is the Gaussian one, consisting
in performing a discrete convolution with a Gaussian kernel over the lattice Lclin of the
image intensity I clin . Unfortunately, together with the noise, smoothing could filter also

12
significant high frequency image contents. Moreover, since the image is separated by the
background by sharp boundaries, characterized by high frequency content, the smoothing
filtering could blur and move the boundaries. To prevent this, anisotropic diffusion filtering
has been introduced [445]: the heat equation is solved for a new image intensity function,
with diffusion coefficient decreasing for increasing values of the gradient magnitude of
intensity. By so doing, the filtering is not performed at the boundaries where the gradient
is large.
Another technique, called multiscale vessel enhancement [196], exploits the specific
tubular shape of vascular geometries and, accordingly, assumes that the smallest modulus
of the eigenvalues of the Hessian matrix of the image intensity function I clin is small, while
the other two are large and of equal sign.
At the end of this substep, we obtain a new image intensity vector I en whose j − th
component, Ijen , represents the intensity of the enhanced image in a gray-scale at the point
xj , 1, . . . , N en , belonging to the lattice Len (and, correspondingly an associated enhanced
image intensity function I en (x) through interpolation). Here, N en is the total number of
points where the enhanced image intensity vector is available. Usually, N en > N clin .

3.1.3 Image segmentation

Image segmentation is the cornerstone of the preprocessing step. It consists in the con-
struction of the shape of a vascular district from the image obtained after the enhancement
substep. In particular, the segmentation allows one to detect those points of the lattice
Len which are supposed to belong to the boundary of the vessel lumen. The precise def-
inition of the boundary of the lumen is a challenging task which requires, in general, a
great user’s experience.
The first technique we describe is thresholding, consisting in selecting a threshold k to
identify the points xj ∈ Len such that Ijen > k. This is motivated by the assumption that
k separates different anatomical structures, in our case the vessel lumen (characterized by
intensity values larger than k) and the background, obtained by the collection of points
for which Ijen ≤ k. The value of k is determined either manually or through a suitable
algorithm. In the latter case, one commonly used strategy is the full width at half max-
imum (FWHM) consisting in setting the threshold halfway between the peak intensity
within the lumen and the intensity of the background. For the segmentation of special
structures, such as calcifications or stents, higher-bound thresholds are used [69].
A more sophisticated class of segmentation methods than thresholding is given by front
propagation methods, where the propagation of a suitable wavefront is tracked. The speed
of the wave is small in regions where I en changes rapidly and high for the other regions, so
that the wavefront is supposed to slow down when approaching the boundary. The most
popular front propagation method is the fast marching method which provides an efficient
solution to the Eikonal problem
1
k∇T (x)k = , x ∈ Den ,
V (I en (x))

where Den ⊂ R3 is a region that contains all xj ∈ Len , and where suitable boundary

13
conditions on a selected boundary where the propagation starts are prescribed [612]. In
the previous equation, V is the speed of the wavefront and T (x) the first arrival time at
point x. In fact, T are iso-contours, representing a collection of surfaces describing the
shape of the waveform. The vessel boundary is then represented by the points xj ∈ Len
such that T (xj ) = T b (up to a given tolerance), where T b is a suitable value selected by
the user.
Another class of segmentation methods is given by deformable models, where a suitable
energy is minimized, allowing the deformation of the body (in our case the boundary of
the vessel lumen) to reach a final state with smallest energy, accounting for external terms
derived from the image and internal terms constraining the boundary to be regular. The
most used class of deformable models is the level set method, where a deformable surface is
represented implicitly as the zero-level of a higher dimensional embedding function [517].
Deformable models, e.g. based on cylindrically parametrized surface meshes, incorporate
anatomical knowledge of the vessel shape [196, 607].
As for the segmentation of the vessel wall, in [525], starting from BB-MRI images, the
vessel wall outer boundary was segmented using the same deformable model used for the
vessel lumen segmentation. Usually, BB-MRI or other images detecting the vessel wall
are not routinely acquired in the clinical practice. In this case, a reasonable approach to
obtain the vessel wall is to extrude the reconstructed boundary lumen along the outward
unit vector by using a suitable function specifying the vessel wall thickness in the different
regions of the district of interest.
In those cases where the image intensity vectors I clin and I en refer to 2D slices, the ap-
plication of the previous segmentation strategies leads to identify several vessel boundaries
(contours), one for each slice, which now needs to be connected to obtain the 3D boundary
surface. This operation is called surface reconstruction. A simple procedure consists in
connecting successive contours by straight lines defining surface triangle edges. This strat-
egy is not suited in presence of changes of shape such as in bifurcations. A better surface
reconstruction is provided by least-square fitting of polynomial surfaces to the contour set
[586]. This strategy is suitable to manage bifurcations whose branches are fitted separately
with a successive extension into the parent vessel. A variant of this approach has been
proposed in [204], where contours are first filled with triangles which are then connected
to the triangles of the adjacent contours by means of tethraedra. The final lumen surface
is then represented by the boundary of this tetrahedral mesh (formed by triangles). We
mention also shape based interpolation where, for each contour, a characteristic function
with positive (respectively, negative) values for points located inside (respectively, outside)
the contour is built. The final lumen boundary surface is then represented by the zero
level-set of the interpolation of all these characteristic functions [486]. Finally, we briefly
describe interpolation by means of radial basis functions (RBF), that provide a flexible
way of interpolating data in multi-dimensional spaces, even for unstructured data where
interpolation nodes are scattered and/or do not form a regular grid, and for which it is
often impossible to apply polynomial or spline interpolation [84, 193]. The coefficients in
the linear combination with respect to the RBF basis are determined by solving a suitable
linear system, which is invertible under very mild conditions [438].
A special mention is deserved by centerline reconstruction. The centerline is a one-

14
dimensional curve centered inside the vessel lumen. Many segmentation tools use the
centerline as starting point making the assumption that the shape of the section is known
around each centerline location [419]. Centerline reconstruction allows a complete recon-
struction of the computational domain when using one-dimensional modeling of blood
flow, see Section 4.5.1.
In any case, at the end of the segmentation step we obtain the lattice Lsurf which
collects the points xj , 1, . . . , N surf , classified as belonging to the lumen vessel surface or
to the outer wall, N surf being the total numbers of points of the surface lattice.

3.1.4 Building the computational mesh

Once the final boundary lattice Lsurf (made of points standing on the lumen boundary) is
made available, we are ready to build the volumetric mesh T vol in the lumen. This mesh
usually consists of unstructured tetrahedra, because of their flexibility to fill volumes of
complex shape.
Unstructured volumetric meshes are constructed starting from an analytical expres-
sion, say S(x), representing the surface associated to the boundary lattice Lsurf . This
expression can derive from an explicit representation, for instance a bivariate parametric
function built as a collection of adjacent polygons. The latter are typically triangles, gener-
ated by Lagrangian shape functions, or patches, generated by high degree polynomial such
as NURBS [532]. Alternatively, the surface is represented implicitly as the iso-surface of
an embedding function. Notice that some of the segmentation strategies described above,
such as deformable models and those used for the surface reconstruction, provide directly
an analytical expression S(x) of the lumen boundary surface.
For the construction of unstructured volumetric meshes T vol , we mention here two
possible approaches. In the first one, a boundary surface mesh T surf is first generated.
To this aim, we start from a lattice Lesurf (in principle different than Lsurf ) composed by
points of S. Then, the Voronoi diagram for Lesurf is constructed. This is a partition of S
into non-overlapping regions, each one containing exactly one point (node) of Lesurf and
being composed by all the points of S that are closer to that node than to any other node.
Starting from the Voronoi diagram, it is possible to obtain a mesh T surf by joining with
straight lines nodes in regions sharing an edge (Delaunay mesh generation). We underline
that the vertices of the mesh T surf do not necessarily coincide with the points of the
lattice Lesurf . Popular algorithms to generate a Delaunay mesh have been proposed, e.g.,
in [590, 591]. Once a surface mesh T surf is made available, the volumetric mesh T vol is
generated. The latter could be obtained by, e.g., advancing front methods, where, starting
from the triangles of the surface mesh, a front composed by internal nodes is generated.
These new nodes allow one to identify tetrahedra, whose validity is verified by checking
that they do not intersect the front [46].
The second approach relies on directly generating the volumetric mesh T vol , e.g., by
means of a Delaunay 3D mesh generation, where a starting volumetric lattice Levol is ob-
tained by locating the nodes in the volume V(x) contained in S(x). One of the main
problems related to this approach is that boundary meshing is often difficult, since the
related surface triangulation could not be of Delaunay type. An alternative approach is

15
given by octree mesh generation, where V(x) is embedded in a box and successive subdi-
visions are performed until the smallest cells permit to accurately describe the boundary.
Although being faster, this strategy generates meshes with poor quality near the boundary.
When a volumetric mesh T vol is obtained, a further step (mesh optimization) prior to
the generation of the final mesh could be introduced, so as to improve its quality. This
prevents mesh distortion, e.g. the presence of very small angles, which could deteriorate the
convergence of algorithms for the solution of the PDE of interest and thus their accuracy.
Mesh optimization leads to an optimal mesh, which produces the best accuracy for a
given number of nodes. This is obtained by selecting (a priori or a posteriori through
error estimates) a lower bound for the mesh size (in case assuming different values in
different regions of V(x)). Mesh optimization is incorporated in the strategies described
above. For example, in octree mesh generation, this is obtained by splitting a cell only if
its children have a size larger than the lower bound.
A mesh is deemed valid for blood flow simulations if it allows to recover outputs of
physical interest. In arteries, the mesh should be fine enough to capture Wall Shear Stresses
(WSS) [86], and, to this aim, the construction of a boundary layer mesh is essential, even
at low Reynolds numbers [53]. WSS expresses the magnitude of tangential viscous forces
exerted by the fluid on the lumen boundary Σt , defined by
v
u 2
uX ¡ ¢2
W SS = µt ∇v n) · τ (j) on Σt ,
j=1

where v is the fluid velocity, n the outward unit vector, and τ (j) the tangential unit vectors,
j = 1, 2. Note that W SS is a scalar function of x ∈ Σt and t > 0. In componentwise
notation,
  2 1/2
X2 X 3 µ ¶
∂vi (j)
W SS = µ   n k τi   on Σt .
∂xk
j=1 i,k=1

As for the structure domain, hexahedral meshing is preferable so as to prevent the

locking phenomenon, whereas tetrahedral meshes are used when conforming meshes at
the boundary lumen interface are needed in view of fluid-structure interaction problems
(see Section 4.3). Usually, three or four layers of elements are enough to obtain an accurate
result [609].

For recent reviews on geometric reconstruction for blood flow simulation, see [10, 510, 332].

3.2 Boundary vascular data

The differential problems we will treat in the next sections need appropriate boundary
conditions. For our problems (incompressible Navier-Stokes equations for the fluid and
finite elasticity for the structure), we anticipate the kind of boundary conditions that
should ideally be prescribed:
¡ ¢
v = g f on ΓD,tf , −pn + µ ∇v + (∇v)T n = hf on ΓN,t f ,

16
for the fluid problem and

d = gs on ΓD,t
s , T s n = hs on ΓN,t
s ,

for the structure problem. In the previous equations, the Dirichlet and Neumann bound-
aries, ΓD,t j and ΓN,t D,t
j , respectively, are such that Γj ∩Γj
N,t
= ∅, ΓD,t
j ∪Γj
N,t
= ∂Ωtj , j = f, s,
t t
Ωf and Ωs being the fluid and structure domains at time t, see Figure 3. Moreover, p
¡ ¢
denotes the fluid pressure, −pn + µ ∇v + (∇v)T n is the fluid normal Cauchy stress,
d the structure displacement, T s the Cauchy stress tensor of the wall material, and
g f , g s , hf , hs given data, see Section 4. In the previous definitions of domains and bound-
aries, the superscript t means that they possibly change in time.
As we will see in the following, the boundary of the computational domain (either the
fluid or the structure one) will be composed of two parts, namely the physical boundary and
the artificial boundary. On the physical boundary, often suitable conditions are suggested
by physical principles. For example, for the fluid problem, no-slip Dirichlet conditions
should be prescribed at the lumen boundary, since it is assumed that the fluid particles
perfectly adhere to the vessel wall. This leads to a homogeneous Dirichlet condition (v = 0)
in the case of rigid walls, and to a kinematic interface condition (v = ḋ) for fluid-structure
interaction problems (see Section 4.3). As for the structure problem, at the internal
physical boundary (that is at the lumen boundary) often the fluid pressure is prescribed.
This leads to a Neumann boundary condition (T s n = −P n, P being a measurement of
the fluid pressure) ¡for a pure structure
¢ problem, and to a dynamic interface condition
T
(T s n = −pn + µ ∇v + (∇v) n) for fluid-structure interaction. On the outer wall
boundary Γext , the interaction with the surrounding tissue should be considered. This is
often modeled by means of a Robin boundary condition of type

αST d + T s n = Pext n on Γext , (1)

which assimilates the surrounding tissue to a sequence of elastic springs with rigidity αST
and where Pext is the external pressure [380].
On a different side, the artificial sections are those introduced by the truncation of the
computational domains, see Figure 3. Truncation is done in order to focus on a specific
domain of interest. Ideally, the boundary conditions to be used on artificial sections should
derive from clinical measurements.
The technique mainly used to obtain boundary data on artificial boundaries is US.
This is because of its non-invasiveness and the fact that it is daily used in the clinical
practice. If the US beam is swept (unlike in geometric acquisitions where it is kept fixed)
through a plane or sector, it is possible to measure the blood velocity in a single point of
a cross-section Γt in the direction of the US beam by exploiting the Doppler effect. The
velocity is then converted into a flow rate measure Q across Γt for each time (this is the
principle of the so-called Echo-Color-Doppler technique [509]) and can then be used to
prescribe at each time a flow rate condition
Z
ρf v · n dγ = Q. (2)
Γt

17
Figure 3: Possible choices of the Dirichlet and Neumann boundaries (left) and physical
and artificial boundaries (right) for a carotid domain in the fluid stand-alone problem
(reconstructed from MRA images)

The previous condition is defective since it is not enough to ensure well-posedness of the
fluid problem. The treatment of defective conditions will be addressed in Section 4.4.
Another technique which allows one to measure cross-sectional flow rates is based on the
acquisition of thermal
R images [349]. Another quantity easily measurable by means of US
is the lumen area Γt dγ. This information could be used to prescribe a defective condition
for the vessel wall.
More sophisticated techniques could acquire velocity measures in several points on the
same cross-section, thus in principle leading to a Dirichlet boundary condition (possibly
after a suitable regularization). This is the case of Phase Contrast (PC)-MRA, where the
blood velocities are encoded into images for several instants of the heartbeat [383]. The
spatial resolution of modern PC-MRA are of the order of 1 − 2 mm on each cross section
and 4 mm along the longitudinal axis. The required mesh size is often less than 1 mm so
that interpolation is however needed to obtain a usable Dirichlet condition. The temporal
resolution is of about 0.03 s.
If the lumen artificial cross section is orthogonal to the axial direction, then the viscous
terms in the fluid normal Cauchy stress are very small [248]. In this case, a measure of the
pressure P (t) could be used to prescribe a Neumann boundary condition. This could be
obtained for the arterial system non-invasively by means of a sphygmomanometer, which
usually measures the pressure at the level of the radial aorta (remember that the pressure
peak could be considered constant along the arterial tree, at least until the capillary net,
see Section 2). To have a continuous monitoring of the pressure (for example during
hospital recovery) or to have a measure in the venous system (where the pressure falls
down) a catheter with a transducer could be placed in the district of interest. In any case,

18
the average pressure over the cross section is measured. This leads at each time to the
following defective boundary condition for the normal component of the normal Cauchy
stress of the fluid Z
1 ¡ ¡ T
¢ ¢
pn − µ ∇v + (∇v) n · n dγ = P. (3)
|Γt | Γt
Unfortunately, no measurement is available at the artificial sections of the structure,
so that “empirical” and “practical” choices are made (see Section 4.4).
Finally, we observe that measurements acquired at several instants during an heartbeat
could be used in principle also for physical boundaries. For example, from 4D-CTA the
boundary lumen displacement could be used as Dirichlet condition for the structure prob-
lem, whereas from PC-MRA the blood velocity at the boundary lumen could be used to
prescribe a Dirichlet condition for the fluid problem. Since at the physical boundaries the
physical principles are used to prescribe boundary conditions, these “extra” data could be
used in view of a validation of the numerical results or in a parameter estimation fashion,
as described in Section 10. PC-MRA allows one to acquire also internal measures of blood
velocity. This could be used again for a validation or in a parameter estimation context.

3.3 Biological vascular data

Finally, we need to know the values of physical parameters appearing in the differential
problems. For the fluid, two parameters characterize the problem, namely blood density
and blood viscosity. Although density is in principle easily measurable, no patient-specific
acquisition is in general made because its value does never significantly depart from an
average value of 1.06 g/cm3 . The range of variability of viscosity is instead larger. Indeed,
its value depends on the shear rate (non-Newtonian behavior) and on the physical state
of the patient. When the assumption of Newtonian fluid (holding for medium and large
healthy vessels) is made, typical values of the viscosity ranges in the interval (0.03 −
0.04 P oise(= 1 g/(cm s))). Again, no patient-specific viscosity measures are usually made
and, unless for pathological situations, a value in the previous range is selected.
The parameters characterizing the vessel wall depend (also in number) by the constitu-
tive law used to represent its behavior. More typically, they are the density, the elasticity
(or compliance) and the degree of incompressibility. For linear elasticity models, the latter
two are quantified by the Young modulus and Poisson modulus, respectively. The density
value is very similar to that of blood, the range most commonly used reaching values up to
1.2 g/cm3 . The compliance of the vessel could be qualitatively seen as the ratio between
the volume and pressure variations [401]. Patient-specific measures could be obtained by
the simultaneous (invasive) measures of pressure and cross-sectional area at different loca-
tions, by measuring the rate of propagation of flow waves [63], or by elastography where
the elasticity properties are analyzed by images before and after the application of a defor-
mation [418]. The range of variability of vessel compliance is quite wide (200 − 800 kP a in
normal conditions), with a great dependence on the location and on possible presence of
aneurysms or calcifications. As for the Poisson modulus, only ex-vivo measurements are
possible. An acceptable value used by the community is 0.49, meaning that vessels could
be considered as quasi-incompressible. Finally, let us mention that the coefficient αST in

19
(1) can be regarded as representative of the Young modulus of the surrounding tissue. As
such, it thus could be measured, even though this is a difficult endeavor. Estimates have
been provided, e.g., in [339].
When not available from measurements, patient-specific values of biological data could
be alternatively obtained by means of parameter estimation mathematical techniques.
This will be the topic of Section 10.

4 Modeling blood flow

4.1 The fluid problem
In large and medium sized arteries, those more typically affected by vascular diseases,
blood can be modeled by means of the Navier-Stokes (NS) equations for incompressible
homogeneous Newtonian fluids [444, 541, 542, 188]. For non-Newtonian rheological models
necessary to describe some specific flow processes, such as clotting or sickle cell diseases,
or more generally flow in capillaries, we refer e.g. to [492, 169].
For the mathematical formulation of the problem, we write the fluid equations with
respect to an Eulerian frame of reference, and we denote by Ωtf = Ωf (t) ⊂ R3 the time-
varying arterial lumen, at time t > 0 (see Figure 4, left).

Figure 4: Representation of the fluid domain on the left and structure domain on the right.
The fluid domain here illustrated is that of an abdominal aorta in presence of aneurysm,
reconstructed from CTA images. The structure domain has been obtained by extrusion
of the fluid one

20
 Then, at each time t > 0, we look for fluid velocity v and fluid pressure p such that
µ ¶
∂v
ρf + ρf (v · ∇)v − ∇ · T f (v, p) = 0 in Ωtf , (4a)
∂t
∇·v =0 in Ωtf . (4b)

Notice that volumetric forces acting in the fluid domain (e.g. due to gravity) were set to
zero since they are quite often neglected. Moreover,
¡ ¢
T f (v, p) = −pI + µ ∇v + (∇v)T (5)

is the fluid Cauchy stress tensor, µ being the blood viscosity. As we consider only Newto-
nian rheology here, µ is assumed to be constant.
Finally, problem (4) is completed by the initial condition

v|t=0 = v 0 in Ωf ,

where Ωf = Ω0f , and boundary conditions. The latter typically prescribe no-slip conditions
on the physical boundary Σt ,
v = φ on Σt , (6)
the upstream velocity on the proximal boundaries, say Γtin ,

v = v up on Γtin , (7)

and traction conditions on the distal boundaries, say Γtout ,

T f n = hf on Γtout . (8)

Here, v 0 , v up , φ and hf are suitable functions with the required regularity [479]. Notice
that the lumen boundary displacement φ at this level is a known function of space and
time. For rigid boundaries, we have φ = 0.
When patient-specific measures are available, other conditions might be prescribed.
However, measures seldom provide a complete data set to be used in the computation, see
our previous discussion in Section 3.2; this prompts the issue of solvability of Navier-Stokes
equations, that we address in Section 4.4.
For each t > 0 (a.e. t > 0), the weak form of (4) together with the boundary conditions
(6)-(7)-(8) reads: Find v = v(t) ∈ [H 1 (Ωtf )]3 , v = v up on Γtin , v = φ on Σt , v = v 0 for
t = 0 in Ωf , and p = p(t) ∈ L2 (Ωtf ) such that
Z Z
∂v t t
ρf · w dω + Af (v, v, w) + B (p, w) = hf · n dγ, (9a)
Ωtf ∂t Γtout

B t (q, v) = 0, (9b)

for all w ∈ V t = {[H 1 (Ωtf )]3 : w = 0 on ∂Ω \ Γtout } and q ∈ L2 (Ωtf ), and where we have
set Z Z ³ ´
Atf (z, v, w) = ρf (z · ∇)v · w dω + µ ∇v + (∇v)T : ∇w dω,
Ωtf Ωtf

21
and Z
t
B (q, w) = − q∇ · w dω.
Ωtf

Existence of a global in time weak solution of the previous problem has been proven in
[331] for the case Ωf = R3 and in [260] for the case of bounded domain. The uniqueness
has been proven only for the two-dimensional case [338]; for the three-dimensional case,
only locally in time uniqueness results are available [462].

4.2 Mechanical wall models

The problem that models the deformation of vessel walls is given by the elasto-dynamics
equation that is usually written in a reference domain Ωs = Ωs (0) ⊂ R3 using a Lagrangian
framework. For any t > 0, the material domain Ωts = Ωs (t) (depicted in Fig. 4, right) is
the image of Ωs by a proper Lagrangian map L : Ωs → Ωts . We use the abridged notation
gb = g ◦ L to denote in Ωs any function g defined in the current solid configuration Ωts .
For the sake of simplicity, we assume the arterial wall to obey a (possibly nonlinear)
finite elastic law relating stress to strain in the arterial tissue; for more complex behaviors
of arterial walls see, e.g., [257, 259]. In more realistic settings, stress is a function of the
strain but also of the past loading history [200].
The problem we consider reads: Find, at each time t > 0, the structure displacement
d such that
∂2db
ρs 2 − ∇ · T b = 0 in Ωs ,
b s (d) (10)
∂t
where ρs is the structure density. Notice that volumetric forces acting in the solid domain
(e.g. due to muscle forces) were set to zero since they are quite often neglected.
The previous equation has been written in terms of the first Piola-Kirchhoff tensor
T b which is related to the Cauchy tensor T s (d) thanks to the relation T
b s (d) b s = JT s F −T .
Here, F = ∇x is the deformation tensor, the gradient being taken with respect to the
reference space coordinates and x being the coordinates of points in the current configura-
tion. Correspondingly, J = det(F ) represents the change of volume between the reference
and the current configurations; notice that F (and thus J) depends on the current config-
uration Ωts .
For a hyperelastic material, the first Piola-Kirchhoff stress tensor is obtained by dif-
ferentiating a suitable Strain Energy Density Function Θ,

b s = ∂Θ .
T (11)
∂F
Several non-linear elastic energy functions have been proposed for arteries. For the Saint
Venant-Kirchhoff material,
µ µ ¶¶2 õ ¶2 !
Eν 1 T E 1 T
Θ(C) = tr (C − I) + tr (C − I) , (12)
2(1 + ν)(1 − 2ν) 2 2(1 + ν) 2

22
where C = F T F , E is the Young modulus and ν the Poisson modulus of the vessel wall.
More complex and accurate functions widely used for arteries are based on separating the
isotropic and elastic behaviour due to the elastin and the anisotropic one accounting for
the stiffening increment at large displacements due to the collagen,

Θ = Θiso + Θaniso . (13)

A common choice for the isotropic part is the Neo-Hookean law

G
Θiso (C) = (tr(C) − 3) , (14)
2
whereas for the anisotropic part, an exponential law is often considered:
k1 ³ (k2 (a·(Ca)−1)2 ) ´
Θaniso (C) = e −1 , (15)
2k2
where a is the unit vector identifying the preferred direction of the collagen fibers, G
the shear modulus, k1 , k2 material parameters where k2 characterizes the stiffness of the
material for large displacements [199, 257, 484]. More complete laws also account for the
symmetrical helical arrangement of the collagen fibers, introducing a second predominant
direction of fibers [259]. When not available from medical images, the fibers directions are
computed by means of suitable algorithms under the assumption that their orientation
is mainly governed by the principal stresses, see, e.g., [242, 307]. For distal arteries of
muscular type, viscoelastic and pseudoelastic terms are also considered [256].
Sometimes, the arterial tissue is considered as incompressible by enforcing to the strain
energy function the incompressibility constraint J = 1 by means of a Lagrange multipliers
approach [255]
Θinc = Θ + ps (J − 1). (16)
Here ps is the hydrostatic pressure related to the vessel wall displacement, which plays
the role of Lagrange multiplier for the incompressibility constraint. Correspondingly, the
Cauchy stress tensor is augmented as follows

T inc
s (d, ps ) = T s (d) + ps I,

where T s is the component arising from the energy Θ. However, experimental studies
showed that the arterial tissue has in fact a nearly incompressible behavior [83]. This
means that the strain energy function could be decomposed into two terms

Θ(C) = Θvol (J) + Θisoc (C̄),

where C̄ = J −2/3 C, detC̄ = 1. The isochoric part Θisoc is given by the general function
(13) provided that C is substituted by C̄ and characterizes the mechanical response of
the material to incompressible deformations. The volumetric part is usually given by
κ
Θvol (J) = (J − 1)2 ,
2

23
where κ (bulk modulus) is a penalty parameter to enforce the incompressibility constraint
J = 1 [336]. In this case, the related Cauchy stress tensor is the same of the compressible
case. For a review of arterial vessel wall models we refer the reader to [259].
Problem (10) has to be completed by the initial conditions

∂d ¯¯
d|t=0 = d0 , ¯ = d1 in Ωs ,
∂t t=0
and boundary conditions. The latter typically prescribe on the artificial sections Γtartif
either
d = 0 on Γtartif (17)
(fixed boundary) or d · n = 0 together with (T s n) · τ (j) = 0, j = 1, 2, τ (j) being the
unit tangential directions (displacement allowed in the tangential direction), whereas at
the internal physical boundary Σt they prescribe the solid traction

T s n = hs on Σt . (18)

In the previous conditions, d0 , d1 and hs are suitable given functions. When considering
the Fluid-Structure (FS) coupling, hs is of course provided by the normal Cauchy stress
from the fluid side, see Section 4.3. To account for the effect of the tissues surrounding
the artery, the algebraic law (1) is often prescribed at the external lateral surface Γext of
Ωs , to mimic the elastic behavior of this tissue [380].
For each time t > 0, the weak form of (10) together with the boundary conditions
(1)-(17)-(18), in the case of fixed boundaries, reads: Find d b = d(t)
b ∈ D, d b = 0 on
b = d0 and ∂ db = d1 for t = 0 in Ωs , such that
Γartif , d ∂t
Z b Z ³ ´ Z Z Z
∂2d b b b·b bs · b
ρs ·b
e dω + T s d : ∇b
e dω + αST d e dσ = b ·b
Pext n e dσ + h e dσ,
Ωs ∂t2 Ωs Γext Γext Σ
(19)
e ∈ D = D 0 , where D t = {e ∈ [H 1 (Ωts )]3 : e = 0 on Γtartif }.
for all b
The existence of strong (steady) solutions of the previous problems could be proven us-
ing the theory developed in [29]. For example, this is the case of the Saint Venant-Kirchhoff
constitutive law given in (12) [107]. The existence and uniqueness of weak solutions are
guaranteed by the coercivity and convexity of the energy Θ [106, 127]. However, a con-
vex strain energy function is not able in general to describe instabilities such as buckling
[28]. Thus, to avoid the use of convex functions, the property of polyconvexity has been
introduced [29]. Its fulfillment guarantees physically admissible solutions [127]. Both the
neo-Hookean law (14) and the exponential one (15) satisfy the polyconvexity property
[32].

4.2.1 Modeling the structure as a 2D membrane

In some circumstances, because of the small thickness of the vessel wall, a non-linear
shell model has been proposed, see, e.g., [333, 611]. In this case, the structure problem

24
is described by two-dimensional equations defined with respect to the middle surface,
consisting of the computation of the deformation of this surface.
A simpler equation may be obtained if the structure is modeled as a 2D membrane
whose position in space at any time exactly coincides with internal boundary Σt , yielding
the so-called generalized string model [478]

∂ 2 dbr
ρs H s − ∇ · (P ∇dbr ) + χHs dbr = fbs in Σ. (20)
∂t2
Here Σ represents the reference membrane configuration, dr is the radial displacement, Hs
the structure thickness; the tensor P accounts for shear deformations and, possibly, for
E 2
prestress, χ = 1−ν 2 (4ρ1 − 2(1 − ν)ρ2 ), where ρ1 (x) and ρ2 (x) are the mean and Gaussian
curvatures of Σ, respectively, [413], and fs the forcing term, given by a measure of the
fluid pressure. Equation (20) is derived from the Hooke law for linear elasticity under
the assumptions of small thickness, plane stresses, and negligible elastic bending terms
[613]. To account for the effect of the surrounding tissue, the term χ in (20) needs to be
augmented by the elastic coefficient of the tissue αST [189].
A further simplification arises when Σ represents the lateral surface of a cylinder. By
discarding any dependence on the circumferential coordinate, model (20) reduces to

∂ 2 dbr ∂ 2 dbr EHs

ρs H s − kGH s + Hs dbr = fbs in Σ, (21)
∂t 2 ∂z 2 (1 − ν 2 )R02
k being the Timoshenko correction factor, G the shear modulus, R0 the initial cylinder
radius, and z the axial coordinate. Often, in the latter case, also a visco-elastic term of
3d
the form γv ∂∂2 z∂t
r
is added, with γv representing a suitable visco-elastic parameter [478].

4.3 The coupled fluid-structure interaction problem

Blood flow in the vessel lumen and deformation of the vessel wall are intimately connected
through a fluid-solid interaction (FSI). In particular, fluid and structure interact through
the fluid-solid (FS) interface Σt which coincides with the physical fluid boundary and the
internal vessel wall boundary introduced in the previous sections. The coupled problem
is obtained by combining at each t > 0 (4) and (10) as follows:
µ ¶
∂v
ρf + (v · ∇)v − ∇ · T f (v, p) = 0 in Ωtf , (22a)
∂t
∇·v =0 in Ωtf , (22b)
∂d
v= at Σt , (22c)
∂t
T s (d) n = T f (v, p) n at Σt , (22d)
∂2db
ρs 2 − ∇ · T b =0
b s (d) in Ωs , (22e)
∂t
df = d at Σt , (22f)

25
together with the boundary conditions (1),(7),(8),(17), and where df is the displacement
of the fluid domain Ωtf at the FS interface and we have used the convenction that n is the
structure outward unit normal. The matching conditions (22c)-(22d) enforced at the fluid-
solid interface express the continuity of velocities (kinematic condition) and the continuity
of normal stresses (dynamic condition), respectively, whereas condition (22f) guarantees
the geometry adherence between the fluid and structure domains (geometric condition).
The well-posedness analysis of the coupled problem (22) (supplemented with the relevant
boundary conditions) has been carried out under several regularity assumptions. We refer
to, e.g., [44, 220, 62, 348], for a comprehensive description of this topic.
For each time t > 0, the weak formulation of the FSI problem (22) together with
its boundary conditions (for the sake of simplicity, we set v up = 0 in (7)) reads: Find
b
(v(t), d(t)) ∈ W t = {(w, b e) ∈ [H 1 (Ωtf )]3 ×[H 1 (Ωs )]3 : (w, b
e) = (0, 0) on Γtin ×Γartif and w =
b
b = d0 and ∂ d = d1 for t = 0 in Ωs , and p(t) ∈ L2 (Ωt ),
e on Σt }, v = v 0 for t = 0 in Ωf , d ∂t f
such that
Z Z b Z ³ ´
∂v ∂2d b : ∇b
bs d
ρf · w dω + Atf (v, v, w) + B t (p, w) + ρs 2
· b
e dω + T e dω
Ωtf ∂t Ωs ∂t Ωs
Z Z Z
+ b
αST d · b
e dσ = hf · n dγ + b ·b
Pext n e dσ,
Γext Γtout Γext
B t (q, v)
= 0,
df = d at Σt ,
(23)
e) ∈ W t and q ∈ L2 (Ωtf ). Notice that in the previous weak formulation, the
for all (w, b
two terms arising after integration by parts and involving the normal Cauchy stresses T f n
and T s n at the interface Σt cancel out, thanks to (22d) and to the special choice of the
test functions in W t .
After introducing the following total energy for the FSI problem
Z Z Z Z
ρf ρs ¯¯ḃ¯¯2 b
¯ ¯2
¯b¯
E3D (t) = |v|2 dω + d
¯ ¯ dω + Θ(d) dω + αST ¯d¯ dγ, (24)
2 Ωtf Ωs 2 Ωs Γext

the following conservation property holds true for the case of homogeneous boundary
conditions [183, 189]:
Z
d µ ¡ ¢2
E3D (t) + ∇v + (∇v)t dω = 0.
dt 2 Ωtf

When the membrane model (20) is used instead of (22e), the matching conditions
(22c)-(22d) are replaced by:
∂dr
v·n= at Σt ,
∂t
T f (v, p)n · n = −fs at Σt ,

where dr is the membrane displacement written in the current configuration and fs is the
forcing term of the membrane equation, see (20), acting only at the FS interface; in this

26
case it also plays the role of structure stress exerted at the FS interface. Owing to (20)
itself, we can rewrite the previous interface conditions as follows:
∂dr
v·n= at Σt , (26a)
∂t µ ¶
∂ 2 dr
T f (v, p)n · n = − ρs Hs 2 − ∇ · (P ∇dr ) + χHs dr at Σt . (26b)
∂t

Since the coupling only occurs in the radial direction, we have to complete the conditions
at Σt for the fluid problem in the tangential directions by prescribing further equations
on the fluid variables, e.g., homogeneous Dirichlet or Neumann conditions [409].
In [178] an effective formulation to solve the FSI problem with a membrane structure
is proposed, whereas in [113] the accuracy of the FSI-membrane problem is discussed in
comparison to a full 3D/3D simulation. In particular, for the Hooke law, the wall shear
stresses computed with these two FSI models are in good agreement for a distal arterial
tract such as a femoropopliteal bypass. Instead, when larger displacements are considered
such as in the ascending aorta, the discrepancies between the two FSI models increase.

4.4 The boundary issue

According to the mathematical theory of the incompressible Navier-Stokes equations, three
scalar conditions need to be prescribed at each point of the boundary. This however is
seldom realistic in clinical practice. For instance, PC-MRI provides velocity data, but this
technique is not routinely used and ad-hoc studies are mandatory [383]. Alternatively, the
flow rate Q = Q(t) can be obtained across a boundary cross section Γt , by proper post-
processing of data retrieved by Echo-Cholor-Doppler or by thermal images, see Section
3.2. This yields the flow rate condition (2). In other situations, at both the inlet and outlet
cross sections, pressure measurements P = P (t) may be considered as representative of
an average estimate, leading to condition (3).
From a mathematical perspective, (2) and (3) are defective conditions as they prescribe
only one scalar function over the entire section Γt [184]. Several strategies have been
proposed so far to supplement (2) or (3) with other conditions that allow to “close” the
system. For the sake of clarity, we can classify them according to three different strategies,
which are reported hereafter.

4.4.1 Conjecturing velocity and pressure profiles

A common trick to effectively prescribe the flow rate condition (2) consists in prescribing
a velocity profile
v(t) = g(t) on Γt , (27)
where g = g(t, x) is chosen in such a way to satisfy (2), that is
Z
ρf g(t) · n dγ = Q(t). (28)
Γt

27
The flow rate condition (2) is therefore replaced by the standard (vectorial) Dirichlet
condition (27). A classical choice for g is a parabolic profile (for example for flow simu-
lations in the carotids [82]), a constant profile (often used for the ascending aorta [380]),
or that obtained from the Womersley solution [245]. Both the parabolic and Womersley
profiles require a circular section to be prescribed on, while non-circular sections require
an appropriate morphing [245].
In spite of its straightforward implementation, this choice has a major impact on the
solution, in particular in the neighborhood of the section Γt and for elevated values of the
Reynolds number [562]. To reduce the sensitivity of the results on the arbitrary choice of
the profile, the computational domain can be artificially elongated by operating what is
called a flow extension [381].
A similar approach could be applied to the mean normal Cauchy stress condition (3)
as well. In the case at hand, we can postulate that the pressure on Γt is constant and that
the viscous stress can be neglected, that is we can prescribe
³ ´
pn − µ ∇v + (∇v)T n = P n on Γt . (29)

Notice that the previous condition in particular satisfies the defective condition (3). Con-
dition (29) is generally acceptable because the pressure changes in arteries mainly occur
along the axial direction and the viscous stresses are negligible on orthogonal cross-sections.
Since P n plays the role of boundary normal Cauchy stress when implemented in the
framework of finite element approximations, no further action than just assembling the
matrix for Neumann conditions is required. For this reason, this treatment has been given
the name of “do-nothing” approach [248]. As pointed out in [559, 558], this procedure in
fact is not completely “innocent”. The do-nothing approach corresponds to the following
weak formulation (for the sake of simplicity we assume homogeneous Dirichlet conditions,
t
v up = 0): Find for each t > 0, v ∈ Ve , v = v 0 for t = 0 in Ωf , and p ∈ L2 (Ωtf ) such that
Z Z Z
∂v
ρf · w dω + Atf (v, v, w) + B t (p, w) = hf · n dγ − P w · n dγ,
Ωtf ∂t Γtout Γt
B t (q, v) = 0,
t ¡ ¢
for all w ∈ Ve = {[H 1 (Ωtf )]3 : w = 0 on ∂Ω \ Γtout ∪ Γt } and q ∈ L2 (Ωtf ).
A do-nothing formulation for the flow rate conditions is possible too, see [248, 559].
Notice that, alternatively to (3), other defective conditions involving the fluid pressure
could be considered as well. This is the case, for example, of mean pressure conditions
ρf 2
[248] or conditions involving the total pressure [189], defined by ptot = p + |v| . For
2
comprehensive review of these conditions, we refer the interested reader to [482].

4.4.2 Augmented formulation

An alternative approach consists in regarding the flow rate boundary condition (2) as
a constraint for the solution of the fluid problem and then enforcing it by a Lagrange
multiplier approach. Being a scalar constraint, we need a scalar multiplier λ = λ(t) at

28
each time, resulting in the following weak formulation (we consider again the case of
t
homogeneous Dirichlet conditions): Find for each t > 0, v ∈ Ve , v = v 0 for t = 0 in
Ωf , p ∈ L2 (Ωtf ), and λ ∈ R such that
Z
R ∂v t t t
ρf Ωtf ∂t · w dω + Af (v, v, w) + B (p, w) + C (λ, w) = hf · w dγ,
Γtout
B t (q, v) = 0, (30)
Q
C t (ψ, v) = ψ ,
ρf
t
for all w ∈ Ve , q ∈ L2 (Ωtf ), and ψ ∈ R, and where we have set
Z
t
C (ψ, w) = ψ w · n dγ,
Γt

see [184, 561], where the well-posedness of this problem is also analyzed.
Besides prescribing the flow rate condition (2), the previous augmented formulation
enforces at each time a constant-in-space normal Cauchy stress on Γt aligned with its
normal direction, which precisely coincides with the Lagrange multiplier λ, that is
³ ´
T
−pn + µ ∇u + (∇v) n = λn on Γt .

This method is particularly suited when the artificial cross section is orthogonal to the
longitudinal axis, so that vector n is truly aligned along the axial direction.
Since the velocity spatial profile is not prescribed a priori, this technique has been
used to improve the parabolic-based law implemented in the Doppler technology for the
estimation of the flow rate starting from the peak velocity [457, 568, 458].
The extension of the augmented formulation to the case of compliant walls is addressed
in [191] and to the quasi-Newtonian case in [159].
An augmented formulation has been proposed in [184] to prescribe condition (3) as
well. However, as noticed in [184], in this case it yields at each time the condition

v = λn at Γt ,

where λ = λ(t) is again the constant-in-space Lagrange multiplier. This is a non-homogeneous

Dirichlet boundary condition for the fluid velocity, which is in general incompatible with
the no-slip condition v = φ prescribed at the physical boundary Σt . For this reason, this
approach is not followed any longer.

4.4.3 A control-based approach

A different strategy for the fulfillment of condition (2) is based on the minimization of the
mismatch functional µZ ¶2
1
J(v) = v · n dγ − Q , (31)
2 Γt

29
constrained by the fact that v satisfies the incompressible Navier-Stokes equations [190].
This PDE-constrained optimization – which can be regarded as the dual of the previous
augmented strategy – yields a system of optimality conditions (also referred to as Karush-
Kuhn-Tucker (KKT) system) to be fulfilled, see Sect. 9.1.2 for further details. In particular,
in [190] the normal component of the normal Cauchy stress on Γt is used as control variable
for the minimization of the mismatch functional. This approach has been considered for the
compliant case in [191], whereas in [324, 201, 202] the non-Newtonian, quasi-Newtonian,
and visco-elastic cases are addressed.
The same approach can also be used to fulfill the defective condition (3) provided that
a suitable functional to be minimized is introduced [190]. This allows to prescribe (3) on a
section oblique with respect to the longitudinal axis too. In this case the control variable is
the complete normal Cauchy stress vector, that is also the direction of the normal Cauchy
stress is a priori unknown.
Boundary data may be lacking for the cross sectional of the vessel wall as well. In this
case we end up with defective BC issues for the vessel wall, see, e.g. [482].

4.5 Geometric reduced models and multiscale approach

4.5.1 The 1D and 0D models
Numerical modeling of the entire cardiovascular system by means of 3D models is currently
not affordable because of the complexity of the computational domain, which is composed
by thousands of arteries and veins and billions of arterioles, capillaries, and venules [401].
In many applications, reduced dimensional models are used instead, either as stand-alone
models or coupled with the 3D ones.
The first one-dimensional (1D) model was introduced almost 250 years ago by L. Euler
[162]. More recently, this approach has been brought into the engineering environment
by [38, 265, 267]. These models allow the description of blood flow in a compliant vessel
where the only space coordinate is that of the vessel axis.
1D models may be derived from 3D models by making simplifying assumptions on
the behavior of the flow, the structure, and their interaction [469, 439]. The starting
fluid domain is represented by a truncated cone, see Figure 5. Referring to cylindrical
coordinates (r, ϕ, z), we make the following simplifying assumptions: (i) the axis of the
cylinder is fixed; (ii) for any z, the cross section S(t, z) is a circle with radius R(t, z); (iii)
the solution of both fluid and structure problems does not depend on ϕ; (iv) the pressure
is constant over each section S(t, z); (v) the axial fluid velocity vz dominates the other
velocity components; (vi) only radial displacements are allowed, so that the structure
deformation takes the form d = der , er being the unit vector in the radial direction and
d(t, z) = R(t, z) − R0 (z), where R0 (z) is the reference radius at the equilibrium; (vii)
the fluid is supposed to obey the Poiseuille law, so that the viscous effects are modeled
by a linear term proportional to the flow rate; (viii) the vessel structure is modeled as a
membrane with constant thickness.
We introduce the following
R quantities: A(t, z) = |S(t, z)| = πR(t, z)2 (lumen sec-
tion area), v̄(t, z) = A−1 S(t,z) vz (t, z)dS (mean velocity), s(r/R) such that vz (t, r, z) =

30
 Figure 5: Fluid domain for the derivation of the 1D model

R
v̄(t, z)s (r/R(t, z)) (velocity profile), Q(t, z) = ρf S(t,z) vz dS = ρf A(t, z)v̄(t, z) (flow rate),
R
P (t, z) = A−1 S(t,z) p(t, z) dS (mean pressure).
As for the structure and its interaction with the fluid, we need to introduce a membrane
law, which in fact prescribes a relation between the pressure and the lumen area (which
is determined by dr ) of the following form

P (t, z) = Pext + ψ(A(t, z), A0 (z), β(z)), (32)

∂ψ
where ψ is a given function satisfying ∂A > 0, ψ(A0 ) = 0. Here β is a vector of parameters
describing the mechanical properties of the membrane.
By integrating over the sections S the momentum fluid equation (22a) in the z−
direction and the mass conservation law (22b), we obtain the following system
∂U ∂U
+ H(U ) + B(U ) = 0 z ∈ (0, L), t > 0, (33)
∂t ∂z
Z R
u2z 1 1 2
S
where U = [A Q]T is the vector of the unknowns, α = = s (y)dy is the
Av̄ 2 qA 0
∂ψ
Coriolis coefficient, Kr = −2πµs′ (1) is the friction parameter, c1 = ρAf ∂A , while
" #
0 1
H(U ) = ³ ´2 , (34a)
Q
c21 − α A 2α Q
A
" #
0
B(U ) = (34b)
Kr Q
A +
A ∂ψ ∂A0
ρf ∂A0 ∂z + A ∂ψ ∂β
ρf ∂β ∂z

represent the flux matrix and the dissipation vector term, respectively. A complete deriva-
tion of the model can be found e.g. in [437, 265, 439]. Classical choices of the velocity
profile s are the flat one (α = 1) and the parabolic one (α = 4/3).
The term ∂A ∂z in B is typically non-positive, accounting for the vessel “tapering”, i.e.
0

the reduction of the area of the lumen when proceeding from proximal to distal arteries.
The term ∂β ∂z originates from possibly different mechanical properties along the vessel, to
describe, for example, the presence of atherosclerotic plaques or vascular prostheses.

31
 If A > 0, system (33) has two real distinct eigenvalues (see, e.g., [469])
q
λ1,2 = αv̄ ± c21 + v̄ 2 α(α − 1), (35)

hence it is strictly hyperbolic (see e.g. [334]). Under physiological conditions, c1 ≫ αv̄,
yielding λ1 > 0 and λ2 < 0, thus we have two waves traveling in opposite directions, that
are associated to corresponding characteristic variables. An explicit expression of these
variables as a function of the physical variables could be in general derived
Wi = ζi (A, Q), i = 1, 2. (36)
A simple membrane law (32) can be obtained by the following algebraic relation [185,
189], √ √ √
A − A0 πHs E
ψ(A, A0 , η) = η , with η = , (37)
A0 1 − ν2
where ν is the Poisson modulus
r of the membrane, E its Young modulus , and Hs its
√
η A
thickness, yielding c1 = 2ρf A0 . This simple law, stating that the membrane radial
displacement dr is linearly proportional to the fluid pressure, is successfully considered in
many applications, see, e.g., [524, 367, 223]. Other laws have been proposed to account for
additional features of arterial walls, such as visco-elasticity, wall-inertia, and longitudinal
pre-stress [478, 185].
Remark 1. 1D models do not allow one to describe secondary flows, such as vortices or
recirculation regions. However, they provide average quantities about the axial component
of the velocity, the radial vessel wall displacements, and the pressure of a complex network
at computational costs that are orders of magnitude lower than those of the corresponding
3D FSI models [306, 250, 60, 68, 351].
The accuracy of the solution provided by 1D models is addressed, e.g., in [20, 524,
490], where the numerical results of different networks are successfully compared with
clinical measurements, and in [367, 386], where a comparison with in vitro measurements
is performed for a complete network of the cardiovascular system.
A further geometrical reduction is represented by the so-called lumped parameters models,
which are zero-dimensional (0D) models obtained by integrating the 1D problem over the
axial direction. These are typically used to describe the peripheral part of the arterial and
venous tree, such as the capillaries and the arterioles.
In this case, only dependence on time is allowed and nominal values of the unknowns
are used as representative of the entire compartment. To this aim, we introduce the
average flow rate and pressure in the district at hand, defined respectively as
Zzd Zzd Z
b =1 ρf
Q(t) Q(t, z) dz = vz (t, z) dSdz,
l l
zp zp S(t,z)
Zzd Zzd Z
1 1
Pb(t) = P (t, z)dz = p(t, z) dSdz,
l V
zp zp S(t,z)

32
where zp and zd are the proximal and longitudinal abscissas of the segment, respectively, l
its length and V the volume. The convective term is dropped since in the peripheral sites
the velocity is small.
If we take the longitudinal average of the momentum equation given by the first of
(33) and we combine it with (37), we obtain the following ordinary differential equation
(ODE) [439]
b ρf KR l
ρf l d Q
+ b + Pbd − Pbp = 0,
Q (38)
A0 dt A20
where Pbd and Pbp are the distal and proximal pressure, respectively. When taking the
longitudinal average of the mass conservation law given by the second of (33) and using
(37), we obtain [439]:
√
A0 l dPb bd − Q
b p = 0,
+Q (39)
η dt
where Q b d and Q
b p are the distal and proximal flow rate, respectively.
The two ODE’s (38)-(39) can be regarded as the starting point toward a 0D description
dQb ρf l
of a compartment model of an arterial tract. In fact, the term L , with L = ,
dt A0
corresponds to the blood acceleration, RQ, b with R = ρf KR l , stems from the blood
A2
√ 0
dPb A0 l
resistance due to the viscosity, while C , with C = , is due to the compliance
dt η
of the vessel wall. Usually, an electric analogy is used to easily interpret 0D models. In
particular, the flow rate plays the role of the current, whereas the pressure is the potential.
Accordingly, the acceleration term is represented by an inductance, the viscosity term by
a resistance, and the compliance term by a capacitance.
To close the system (38)-(39) (featuring 4 unknowns) we need to include also the
boundary conditions originally prescribed to the 1D model. For instance, we can assume
a Dirichlet condition at the inlet and a Neumann condition at the outlet. Thus, we may
localize the unknown pressure Pb at the proximal section, (Pb ≈ Pbp ), assuming that the
distal pressure Pbd is given. Similarly we assume that the flow rate Q b is approximated by
b b
Qd and that the proximal flow rate Qp is given. Then, from (38)-(39), we obtain
b
dQ
Pb − L b = Pbd ,
− RQ
dt
(40)
dPb b=Q
bp ,
C +Q
dt
corresponding to the electrical circuit drawn in Fig. 6. Other sequences corresponding to
different boundary conditions and then to different state variables are possible too, see,
e.g. [482]. Even though these schemes are equivalent in terms of functionality, they play
a different role when coupled with higher dimensional models, see, e.g., [482].
For a description of more complex vascular districts, we may combine several 0D el-
ementary tracts, by gluing them owing to classical continuity arguments. However, the

33
 Figure 6: Example of lumped parameter scheme for an arterial tract

lumped parameters models have been mainly used to provide suitable boundary conditions
at the distal artificial sections of 3D and 1D models. In this case, one simple compart-
ment is enough to describe the entire arterial system downstream the region of interest.
Examples are provided by the windkessel model [594], featuring an average resistance and
capacitance, the 3-element windkessel [594], where a second resistance is added before
the windkessel compartment, and the 4-element windkessel model [526, 527], where an
inductance element is added to the 3-element windkessel model. A 0D model given simply
by a resistance is used to provide absorbing boundary conditions at the outlets of the
fluid domain in FSI simulations, see, e.g., [413]. Instead, more sophisticated approaches
account for the propagative dynamics associated with the peripheral circulation, such as
the structured tree model [420], which assumes an asymmetric self-similar structure for the
peripheral network.

4.5.2 The geometric multiscale coupling

The geometrical multiscale approach, first introduced in [480], consists in the coupling
among 3D, 1D, and 0D models. The idea is to use higher dimensional models in those
regions where a description with a great detail is required, and lower dimensional models
in the remaining part of the region of interest. This allows one to describe a wide portion
of the circulatory system.
As discussed earlier, 0D models are typically used to provide boundary conditions
for 3D and 1D models. For this reason, the coupling between 3D or 1D models with
an extended 0D network has been rarely considered in applications. Instead, the 3D-1D
coupling has received a great deal of attention. For this reason, we detail here only the
latter case, while referring to, e.g., [472, 481, 574, 297, 372, 239] for the 3D-0D coupling,
and [187, 175] for the 1D-0D coupling.
As reported in Figure 7, we consider a 3D/FSI problem (22) in a 3D cylindrical domain
together with initial conditions and boundary conditions at the proximal boundaries and
at the external structure.
At the distal boundaries, the 3D problem is coupled with the 1D model (33) written
in the domain z ∈ [0, L], together with initial conditions and a boundary condition at the
distal boundary. We name Γt = Γtf ∪ Γts the coupling interface from the 3D side, which
corresponds to the point z = 0 from the 1D side (see Fig. 7).

34
 Figure 7: Schematic representation of the reference 3D-1D coupled model

A major mathematical issue is how to couple 3D and 1D models at the common

interface. Several strategies can be pursued, yielding to many (alternative) sets of interface
conditions. For a rigorous derivation of the 3D-1D problem and a detailed discussion of
the interface conditions, we refer the interested reader to the recent review article [482].
Instead, here we follow the guiding principle described in [189], where suitable interface
conditions are derived from a global energy estimate. In particular, we introduce, together
with the 3D energy (24), the 1D energy [183]
Z Z L
ρf L
E1D (t) = A v̄ 2 dx + χ(A) dx,
2 0 0
Z A
where χ(A) = ψ(τ ) dτ , ψ being the vessel law (see (32)), Notice that the stand-alone
A0
1D problem satisfies bounds for this energy functional as proved in [183].
ρf ρf
Let Ptot = ψ(A) + v̄ 2 be the total pressure for the 1D model, and ptot = p + |v|2
2 2
that of the 3D model. Then, we have the following result [189].
Proposition 1. For the interface coupling conditions holding at Γt let us assume that the
following inequality holds
Z Z
T f (v, ptot )n · v dγ + T s (d)n · ḋ dγ + Q|z=0 Ptot |z=0 ≤ 0. (41)
Γtf Γts

Then, for all t > 0, the coupled 3D-1D problem (22)-(33) with homogeneous boundary
conditions satisfies the energy decay property
d ¡ 3D ¢
E (t) + E 1D (t) ≤ 0.
dt
The previous result provides an indication on how to find suitable interface conditions
for the 3D-1D coupled problem. In particular, for inequality (41) to be fulfilled it is
sufficient that the following interface conditions
Z
ρf v · n dγ = Q|z=0 , (42a)
Γtf

(T f (v, ptot )n) |Γt = −Ptot |z=0 n (42b)

f

35
hold for the fluid, together with
T s (d)n = 0 on Γts (43)
for the structure [186, 189]. Similarly, inequality (41) holds if relation (43) is replaced by
d·n=0 on Γts ,
(44)
(T s (d)n) × n = 0 on Γts .
The interface conditions (42) prescribe the continuity of the flow rate (kinematic condition)
and a dynamic condition involving the total pressures. Note that (43) and (44) are in
fact independent of the 1D model, resulting in boundary conditions for the 3D structure
problems only. This allows a discontinuity to manifest in the displacement between the
3D and the 1D model.
Usually, dynamic interface conditions involving the pressure (instead of the total pres-
sure) are considered in place of (42b), such as
Z
1
T f (v, p)n dγ = −ψ(A|z=0 )n. (45)
|Γtf | Γtf

However, the previous condition does not satisfy the compatibility condition (41). More
precisely, in this case we have
à Z !
d ¡ 3D 1D
¢ ρf (Q(t)|z=0 )3 2
E (t) + E (t) = − |v(t)| v(t) · n dγ .
dt 2 (A(t)|z=0 )2 Γtf

Even though the right hand side is not necessarily (always) negative, numerical evidence
indicates that condition (45) leads to stable results for hemodynamic applications, see
[351]. This interface condition is indeed the most commonly used among the dynamic
ones.

4.6 Numerical strategies

In this section, we comment on the numerical solution of the problems presented in the
previous subsections. We do not have the ambition to be exhaustive, due to the problem
complexity. Rather, we review some of the methods which are among the most suited
in hemodynamics. In particular, for the space discretization, we will focus on Galerkin-
type methods, such as Finite Elements, Spectral Elements, and Discontinuous Galerkin
methods.
We will use the following notation. Let ∆t and h be the time and space discretization
parameters. In our examples ∆t is assumed to be fixed, but adaptive strategies could
be considered as well, see, e.g., [564]. Correspondingly, the discrete time instants are
tn = n∆t. h is instead defined as usual as a representative value of the mesh size, for
example h = minK hK , where K are the tetrahedra of the mesh and hK the radius of
the sphere inscribed in K. Given the functions w(t) and z(x), we denote by wn the
approximation of w(tn ) and by zh (x) the Galerkin approximation of z(x).

36
4.6.1 Numerical methods for the fluid problem
We start to review some numerical methods for the fluid problem (4) together with its
initial and boundary conditions.
As for the time discretization, usually implicit methods with a semi-implicit treatment
of the convective term and (in case of moving domain) of the fluid domain are considered.
The problem is solved at the discrete time tn+1 in the domain Ω∗f and with convective
term ρf (v ∗ · ∇)v n+1 , where Ω∗f and v ∗ are suitable extrapolations of Ωn+1
f and v n+1 of
the same order of the time discretization. This choice introduces a CFL-like restriction on
the time step to preserve absolute stability [479]. However, this condition is very mild in
hemodynamic applications, since, for accuracy purposes, the pulsatility of the blood signal
and the quick dynamics around systole can only be accommodated by choosing a small
∆t. Usually, a second order approximation is considered a good choice in hemodynamics;
in this respect, BDF2 and Crank-Nicolson are the mostly used methods [475].
The first class of methods we present is based on a decomposition of the semi-discrete
problem at the spatial continuous level (differential splitting or projection methods). The
basic idea underlying these methods is to split the computation of velocity and pressure,
with a final step aiming at recovering the incompressibility constraint. We detail in what
follows the Chorin-Teman method [104, 543], originally proposed for homogeneous Dirich-
let conditions and fixed domain, which is the progenitor of these methods. We only detail
the case of the Backward Euler discretization.

Chorin Temam method. For n ≥ 0, at time tn+1

1. solve in Ωf the advection-reaction-diffusion problem with homogeneous Dirichlet

condition for the intermediate unknown velocity ven+1
µ ³ ´T ¶
en+1 − v n
v n+1 n+1
ρf − µ ∇e v + ∇ev v n+1 = 0;
+ ρf (v n · ∇)e
∆t

2. solve in Ωf the pressure problem with homogeneous Neumann conditions

ρf
△pn+1 = en+1 ;
∇·v
∆t

3. correct the velocity

∆t n+1
en+1 −
v n+1 = v ∇p .
ρf

This splitting method is based on the Ladhyzhenskaja theorem [214], stating that a vector
function belonging to [L2 (Ωf )]3 can be always decomposed as the sum of a solenoidal part
and of a gradient term. In fact, the correction step corresponds to project the intermediate
velocity onto H = {w ∈ [L2 (Ωf )]3 : ∇ · w = 0, w · n|∂Ωf = 0}. Thus, it is possible to
show that v n+1 and pn+1 are in fact solution of the original semi-discrete problem. The
Chorin-Temam method is very effective since it overcomes the saddle-point nature of the
problem and solve two standard uncoupled elliptic problems. However, it suffers from

37
inaccuracies at the boundary. In particular, the tangential velocity cannot be controlled
(see the definition of H) and spurious pressure values appear as a consequence of the
artificial Neumann condition for the pressure problem [485]. This has an effect on the
accuracy of the semi-discrete solution, in particular the following error estimate holds true
[485]: √
kv(tn , x) − v n (x)k[H 1 ]3 + kp(tn , x) − pn (x)kL2 . ∆t.
The use of higher order time approximations leads to the same accuracy.
An improvement of the previous method is given by the rotational incremental variant
of the Chorin-Temam scheme [545]. We detail in what follows the case of BDF2 and sec-
ond order extrapolation of the convective term, since the first order approximation does
not lead to any improvement.

Rotational-incremental Chorin-Temam method. For n ≥ 0, at time tn+1

1. solve in Ωf the advection-reaction-diffusion problem with homogeneous Dirichlet

condition in the intermediate unknown velocity ven+1 :
µ ³ ´T ¶
v n+1 − 4v n + v n−1
3e n+1 n+1
ρf −µ ∇ev + ∇ev v n+1 +∇pn = 0;
+ρf ((2v n −v n−1 )·∇)e
2∆t

2. solve the pressure problem

µ ¶
3ρf
△pn+1 = △pn
+ −µ ∇·v en+1 x ∈ Ωf ,
2∆t
∂pn+1 ³ ´
=µ ∇×∇×v en+1 · n x ∈ ∂Ωf ,
∂n

3. correct the velocity

2 ∆t ³ n+1 ´
en+1 −
v n+1 = v ∇ p en+1 .
− pn + µ∇ · v
3 ρf

Unlike the classical Chorin-Temam scheme, in the previous method the boundary con-
ditions for the pressure problem are consistent and no numerical boundary layer for the
pressure is observed. This is confirmed by the improved error estimate [232]

kv(tn , x) − v n (x)k[H 1 ]3 + kp(tn , x) − pn (x)kL2 . ∆t3/2 .

The previous two methods belong to the general class of pressure-correction methods,
see also, e.g., [111, 229, 230]. A different class is obtained by switching the role of velocity
and pressure in the splitting, i.e. the viscous term is now ignored or treated explicitly in
the first step and the velocity is then corrected accordingly (velocity-correction schemes,
[421, 285]). These schemes feature the same non-optimal error estimates of the pressure-
correction schemes due to artificial Neumann conditions for the pressure problem. Again,
an improvement could be obtained by considering a rotational-incremental variant [231].

38
 In hemodynamics it is often the case that Neumann boundary conditions are prescribed
at some artificial section. The extension of the differential splitting methods to this case
is addressed in [228]: on the Neumann boundary we have an artificial Dirichlet condition
for the pressure, which again deteriorates the optimal rate of convergence with respect to
∆t.

In view of the next methods we are going to review, it is convenient to introduce the
algebraic problem arising from the application of a Galerkin-like method to the semi-
discrete-in-time problem. First of all we notice that the solvability of the discretized-in-
space problem is guaranteed by a suitable compatible choice of the approximation spaces
for the velocity and the pressure in order to satisfy the discrete inf-sup stability condition
[479]. As it is well known, an example for Finite Elements (FE) for a tetrahedral mesh is
provided by piecewise polynomials of order 2 for the velocity approximation and of order
1 for the pressure approximation. This choice guarantees the existence and uniqueness of
the solution to the linearized fully discrete problem and are often used to provide a stable
solution in hemodynamics. In this case, we have the following optimal error estimate
kv n (x) − v nh (x)k[H 1 ]3 + kpn (x) − pnh (x)kL2 . h2 ,
provided that v n and pn are regular enough. (For other stable choices see [479, 64]). Al-
ternatively, suitable stabilization terms could be added to the problem, circumventing the
inf-sup condition and allowing the use of polynomials of equal order. In this case, addi-
tional terms are added to the mass conservation equation and, in case, to the momentum
conservation equation. Usually, these techniques allow one to stabilize also convected-
dominated problems arising when the Reynolds number is high, for example in aorta or
in stenotic carotids. One technique is Streamline Upwind-Petrov Galerkin (SUPG) [385].
A generalization of SUPG is the variational multiscale (VMS) method [266, 268], which
is based on the decomposition of the unknown into two terms, one accounting for the
large scales and another one for the small scales. The same decomposition is used for
the test functions, so that a system of two coupled problems is obtained. VMS is also
useful since it allows one to model the transitional to turbulence effects which may occur
in some pathological conditions such as stenoses [3]. For an application to hemodynamics,
see [195].
We introduce in what follows the algebraic problem related to the fully discretized
linearized problem. For the sake of exposition, we limit ourselves to the cases without
stabilization terms. We refer the interested reader to, e.g., [154, 48] for the more general
case. At each time step we have
· ¸· ¸ · ¸
A BT V Ff
= ,
B 0 P 0
where V and P are the vectors collecting the velocity and pressure unknowns, A =
α
ρf ∆t Mf +ρf N (V ∗ )+µK (with Mf the mass matrix, N the matrix related to the linearized
convective term, K the stiffness matrix), F f accounts for non-homogeneous Dirichlet and
Neumann conditions and the terms coming from time discretization, α depends on the
time discretization scheme, and where we have omitted the current temporal index n+1 .

39
 The previous non-symmetric linear system can be solved by a Krylov method, e.g. by
the GMRES method. Suitable preconditioners are mandatory. A classical choice is given
by block-preconditioners which again split in fact the solution of the velocity and of the
pressure, e.g. · ¸
PA B T
P = .
0 −PΣ
If PA = A and PΣ = Σ = BA−1 B T (Schur complement), then the solution is achieved
in three GMRES iterations [154]. In fact, this choice is equivalent to formally solve the
momentum equation for the velocity and to substitute its expression in the mass equation.
However, in practice this preconditioner is not efficient, since the linear system involving
the Schur complement is too onerous, Σ being a full matrix whose explicit construction
requires the knowledge of A−1 . Efficient preconditioners can be obtained by approximating
Σ (and, in case, A). For low Reynolds numbers (say, less than 10), an effective choice is
given by PΣ = µ1 MP , where MP is the pressure mass matrix (or even its diagonal) [155].
Thus, this is a good choice in hemodynamics for small vessels. For increasing Reynolds
numbers, the convergence properties of this preconditioner deteriorates since it does not
account for the convective term. A better choice for medium and large vessels is given by
PΣ = AP FP−1 MP , where AP is the pressure stiffness matrix and FP = µAP + ρf NP (V ∗ ),
NP being the matrix related to the convective term defined on the pressure space (pressure
convection-diffusion preconditioner, [154, 48]). As for the solution of the velocity problem,
suitable preconditioners for the advection-reaction-diffusion problem could be introduced.
Alternatively, fast solutions such as V-cycle multigrid can be considered as well [553].
Another class of preconditioners is obtained by an inexact block LU factorization of
the fluid matrix. The starting point is the exact factorization
· ¸ · ¸· ¸
A BT A 0 I A−1 B T
= .
B 0 B −Σ 0 I

b1 and A
Again, different preconditioners are obtained by suitable approximations A b2 of A
b
and Σ of Σ, leading to " #· ¸
Ab1 0 I A b−1 B T
P = 2 .
B −Σ b 0 I

A convenient choice is Ab1 = A b2 = DA and Σ b = BD−1 B T , where DA is the diagonal of

A
A (SIMPLE preconditioner, [432, 335]). Notice that in this case Σ b is sparse and could
be explicitly assembled. This is an effective choice when the fluid matrix is diagonally
dominant, i.e. when small values of ∆t are used. Another choice is Yosida preconditioner,
where Σb = ∆t BM −1 B T , A b1 = A and A b2 = ρf α Mf [560]. Again, the efficiency deteri-
ρf α f ∆t
orates for increasing ∆t. The Yosida preconditioner had been originally introduced as a
solver in [477, 476]. This lead to a splitting of the velocity and of the pressure compu-
tation, which could be seen as the algebraic counterpart of the Chorin-Temam method
(algebraic pressure-correction methods). In particular, we have the following steps:

1. AVe = F f (computation of the velocity);

40
 2. ∆tBMf−1 B T P = B Ve (computation of the pressure);

3. V = Ve − ∆t −1 T
ρf α Mf B P (correction of the velocity).

Again, an incremental version of the algebraic pressure-correction methods could be con-

sidered as well [477]. An extension to spectral methods is provided in [209].
For a recent comparison of the performance of different preconditioners (all described
above) used for hemodynamic applications, see [139].

In healthy conditions, blood flow is mainly laminar. Transitional flow may develop in
some pathological instances, or under the assistance of devices. In these circumstances,
suitable mesh refinement, possibly accompanied by the use of turbulence models, are often
employed. We mention the case of stenotic carotids, where Reynolds-average Navier Stokes
(RANS) models are used in [529, 224], Direct Numerical Simulations (DNS) in [326, 180,
99], and Large Eddy Simulations (LES) in [487, 316]. In [41] the VMS formulation is used
to describe transitional effect in the ascending aorta under the effect of the Left Ventricular
Assist Device (LVAD).
In Figure 8 we report some examples of numerical results obtained in four real geome-
tries reconstructed from radiological images (see the caption for details). These results
highlight the complex pattern of blood flow induced by the geometry and by the heart
pulsatility. To highlight the transitional effects in stenotic carotids, we plotted in one of
the figures the Q criterion, defined as
1X 2
Q=− Sij − Ω2ij ,
2
i,j

where S = ∇u+(∇u)T and Ω = ∇u−(∇u)T [326]. Positive values of Q indicate locations

where rotations dominates strain and shear.

4.6.2 Numerical methods for the vessel wall problem

In this section we review some of the most commonly used numerical approaches for the
solution of problem (10), endowed with its initial and boundary conditions.
As for the time discretization, a popular family of schemes is Newmark’s, which is
characterized by two parameters θ and ζ [399]. The special combination θ = 1/2 and
ζ = 1/4 yields the following semi-discrete form of (19) (set Pext = 0 and hs = 0 for the
sake of simplicity):
Z n+1
b bn Z ³ n+1 ´ Z
4d − 4d b b bn+1 · b
ρs 2
·b
e dω + Ts d : ∇b
e dω + αST d e dσ
Ωs ∆t Z µ Ω¶s Γext
n
b
4w
= ρs +abn · b e dω, (46)
Ωs ∆t

2 ³bn+1 bn ´ 4 ³bn+1 bn ´ 4 n
b n+1 =
w d −d −w b n, b n+1 =
a 2
d −d − bn,
b −a
w
∆t ∆t ∆t

41
Figure 8: Top left: velocity vectors in the aneurysm of an abdominal aorta (CT images
from the Vascular-surgery and Radiology Divisions at Fondazione IRCSS Cà Granda, Os-
pedale Maggiore Policlinico, Milan, Italy). Top right: velocity streamlines in a stenotic
carotid (MRI images from the Vascular-surgery and Radiology Divisions at Ospedale Mag-
giore Policlinico, Milan). Bottom left: coherent vortical structures by Q criterion in a
stenotic carotid (we report only the regions with Q > 50000 painted by the velocity mag-
nitude, CT images from the Vascular-surgery and Radiology Divisions at Ospedale Mag-
giore Policlinico, Milan). Bottom right: wall shear stress in an ascending aorta (MRI im-
ages from the Cardio-surgery and Radiology Divisions at Ospedale Borgo Trento, Verona,
Italy). These numerical results are obtained using the finite element library LifeV, P 2/P 1
Finite Elements , the backward Euler scheme for the time discretization with a semi-
implicit treatment of the nonlinear term, and the Yosida preconditioner. For the stenotic
carotids a LES model has been used

where wb n+1 and ab n+1 represent approximations of vessel wall velocity and acceleration,
respectively. This method is unconditionally absolutely stable and second order accurate
with respect to ∆t. An extension of Newmark schemes is provided by the generalized-alpha
method [105], see, e.g., [277] for an application to hemodynamics.

42
 Space discretization is typically based on Finite Elements. Whatever the implicit
temporal scheme chosen, a system of non-linear algebraic equations is obtained after space
and time discretization, reading
ρs β
Ms D + Γ(D) + αST Msext D = Gs ,
∆t2
where β depends on the time discretization (e.g., β = 4 for the Newmark method (46)),
D is the vector collecting the vessel wall displacement unknowns, Ms is the mass matrix,
Msext the boundary
³ ´ mass matrix related to Σext , Γ is the non-linear operator defined by
R
b b
Γi = Ωs T s d : ∇ebi dω, b ei being the i − th basis function, and Gs the vector related to
the right hand side of the discretized-in-time equation. Notice that we have suppressed
the temporal index that is understood. The previous system is linearized by means of the
Newton method, obtaining at each time step a sequence of linear systems of the form
µ ¶
ρs β ¡ ¢ ext
Ms + T D (k−1) + αST Ms δD (k) =
∆t2
ρs β ¡ ¢
Gs − 2
Ms D (k−1) − Γ D (k−1) − αST Msext D (k−1) ,
∆t

¡with k ≥ 1 the¢Newton iterations index, to be solved until convergence. Here δD (k) =

D (k) − D (k−1) , T is the matrix related to the linearization of the first Piola-Kirchhoff
R ³ ³ ´ ´
tensor, i.e. (T )ij = Ωs DF T bs d b(k−1) : ∇bej : ∇b ei , DF being the Gâteaux derivative
with respect to F .
For the solution of the previous linear system, Domain Decomposition (DD) methods
are often used as efficient preconditioners for iterative Krylov methods. Since matrix T is
symmetric, the Coniugate Gradient method is usually considered for iterations. Among
DD preconditioners, FETI methods [168] are very often used in structural mechanics.
In particular, all floating FETI methods have been considered for vessel wall problems,
e.g., in [17]. Like in classical FETI methods, Lagrange multipliers are introduced to glue
the solution at the subdomain interfaces. In addition, Lagrange multipliers are also used
to prescribe Dirichlet boundary conditions. This simplifies the implementation of FETI
method since all the subdomains are treated in the same way. A variant successfully used
for arterial vessel walls is the so-called Dual-Primal FETI method, see, e.g., [31]. Finally,
we mention as yet another class of DD methods considered for this problem: a two-level
overlapping Schwarz method with an energy minimization coarse space, see [145].
In Figure 9 we report a coupled of examples of numerical results obtained in real
geometries reconstructed from radiological images (see the caption for details). These
results highlight the anisotropic internal stresses characterizing vascular vessel walls.

4.6.3 Numerical methods for the fluid-structure interaction problem

The numerical solution of the coupled FSI problem (22) requires to manage three sources
of non-linearities, i.e.:
i) the fluid domain is unknown (geometric non-linearity);

43
Figure 9: Left: von Mises internal stresses in a carotid (MRI images from the
Vascular-surgery and Radiology Divisions at Ospedale Maggiore Policlinico, Milan,
Italy). Right: von Mises stresses in an abdominal aortic aneurysm (right, mesh
from [Link] These numerical re-
sults are obtained using LifeV (carotid) and the Finite Element library redbKIT v2.1
([Link]/redbKIT/redbKIT/releases) (AAA), P 2 Finite Elements, a Newmark un-
conditionally stable scheme for the time discretization and an exponential vessel wall law

ii) the fluid subproblem is non-linear (fluid constitutive non-linearity);

iii) the vessel displacement subproblem is non-linear (structure constitutive non-linearity),
together with two different kinds of coupling, i.e.:
iv) the displacement of the fluid domain at the FS interface needs to match the vessel
wall one (geometric adherence, see condition (22f));
v) the fluid and vessel displacement subproblems are coupled by means of the kinematic
and dynamic conditions (22c)-(22d) (physical coupling).

Arbitrary-Lagrangian formulation. As for points i) and iv), a classical numerical

strategy relies on extending the FS interface displacement df = d|Σt in the whole fluid
domain, thus associating a displacement also to its internal points. This is obtained by
solving an extra problem for the fluid mesh displacement, usually an harmonic extension
of the FS interface datum df , with homogeneous Dirichlet conditions at ∂Ωtf \ Σt . The
fluid domain displacement is then used to move the points of the fluid mesh accordingly,
obtaining the new computational fluid domain. With this aim, the Navier-Stokes equations
are reformulated on a frame of reference which moves with the fluid mesh. This is neither
a Lagrangian description (where the frame moves with the fluid particles) nor an Eulerian
description (where a fixed frame would be used). For this reason, it is called Arbitrary
Lagrangian-Eulerian (ALE) approach [253, 148]. To write the Navier-Stokes equations
in ALE configuration, the Reynolds transport formula to express the ALE material time

44
 δ
derivative δt of a function z with respect to the Eulerian one is used, i.e.

δz ∂z
= + (v f · ∇)z,
δt ∂t

v f = ḋf being the velocity of the points of the fluid mesh. Thus, the FSI problem together
with its initial and boundary conditions becomes:
µ ¶
δv
ρf + ((v − v f ) · ∇)v − ∇ · T f (v, p) = 0 in Ωtf , (47a)
δt
∇·v =0 in Ωtf , (47b)
∂d
v= on Σt , (47c)
∂t
T s (d)n = T f (v, p)n on Σt , (47d)
∂2db ³ ´
ρs 2 − ∇ · T bs d b =0 in Ωs , (47e)
∂t
bf = d
d b on Σ, (47f)
− △d bf = 0 in Ωf . (47g)

This choice is particularly suited when Finite Elements are considered for the space dis-
cretization. Indeed, the terms involving spatial derivatives are as usual expressed with
respect to the current configuration, whereas the (material) time derivative term is writ-
ten in the reference configuration. In particular, given the nodal basis functions φj , we
have  
δ(v h (t, x)) δ X j X dv j (t)
= vh (t)φj (x(t)) = h
φj (x(t)),
δt δt dt
j j

since the time variations of the basis functions with respect to the reference domain vanish.
This makes the computation of the fluid velocity on the nodes of the fluid mesh easy, see
e.g. [409].
For time discretization of (47), a common choice is to discretize the fluid and the
vessel wall problems with two schemes of equal order (let us say of order p), e.g., BDF2
or Crank-Nicolson for the fluid and the Newmark scheme (46) for the vessel problem.

Treatment of geometric coupling, geometric adherence, and fluid non-linearity.

As for issue ii) above (constitutive fluid non-linearity), following the approach commonly
used for a stand-alone fluid problem, the convective field is usually treated explicitly, by
using a suitable extrapolation v ∗ − v ∗f of order p from previous time steps. This choice
introduces a CFL-like condition for ∆t to preserve absolute stability, which however is al-
ways satisfied for the values of ∆t usually considered in hemodynamics. Another common
choice in hemodynamics is the explicit treatment of the geometric coupling and adherence
(issues i) and iv) above). In particular, the fluid problem coupled with the structure one
is solved in a domain Ω∗f obtained by a suitable extrapolation from previous time steps

45
of order p and the fluid geometry problem is then solved in sequence [536, 174, 26, 413].
Again, a limitation on ∆t is required to ensure absolute stability. In particular, in [174]
it is proven for a model problem that stability is guaranteed under a CFL-like condition.
Numerical evidence in real scenarios highlighted that also this is a mild condition in the
hemodynamics regime, see, e.g. [380, 411]. Based on the considerations above, the follow-
ing temporal discretization of the FSI problem (47) can be considered:

Explicit scheme for the geometric coupling and adherence. For n ≥ 1, at time step tn :

1. solve the FSI problem:

ρf α n
v + ρf ((v ∗ − v ∗f ) · ∇)v n − ∇ · T f (v n , pn ) = g nf in Ω∗f , (48a)
∆t
∇ · vn = 0 in Ω∗f , (48b)
α n
vn = d + g nfs on Σ∗ , (48c)
∆t
T s (dn ) n∗ = T f (v n , pn ) n∗ on Σ∗ , (48d)
ρs β b n ³ n´
d − ∇ · bs d
T b =g bns in Ωs , (48e)
∆t2
where α still depends on the time discretization scheme and g nf = g nf (v n−1 , v n−2 , ...),
g ns = g ns (dn−1 , dn−2 , ...), and g nfs = g nfs (dn−1 , dn−2 , ...) account for the terms at
previous time steps coming from time discretizations of order p of the corresponding
equations (47a),(47e), and (47c), respectively;
2. then, solve the fluid geometry problem:
bn = 0
− △d in Ωf , (49a)
f
bn
d b
=d
n
on Σ, (49b)
f

and then build Ω∗f accordingly.

In the previous substeps, the FSI problem (48) is still coupled by means of the physical
coupling given by the interface conditions (48c)-(48d), see issue v) above. For the solu-
tion of this problem, both partitioned and monolithic procedures have been successfully
considered so far in hemodynamics. In partitioned schemes, the fluid and vessel wall sub-
problems are solved separately, one or more times per time step. Each of the two problems
is equipped with a suitable boundary condition at the FS interface Σ∗ derived by splitting
the physical interface conditions (48c)-(48d).

Partitioned algorithms of explicit type. From the computational point of view, an

attractive class of partitioned schemes is that of the loosely-coupled algorithms where the
two subproblems are solved only once per time step. The most classic loosely-coupled
algorithm, widely used in aerodynamics [455], is the explicit Dirichlet-Neumann scheme,
where condition (48c) is prescribed explicitly as a Dirichlet condition to the fluid sub-
problem, whereas condition (48d) is prescribed as a Neumann condition to the vessel wall

46
subproblem, leading to the following

Explicit Dirichlet-Neumann scheme. For n ≥ 1, at time step tn :

1. solve the fluid Oseen problem with a Dirichlet condition at the FS interface:
ρf α n
v + ρf ((v ∗ − v ∗f ) · ∇)v n − ∇ · T f (v n , pn ) = g nf in Ω∗f , (50a)
∆t
∇ · vn = 0 in Ω∗f , (50b)
α n−1
vn = d + g fn−1
s on Σ∗ ; (50c)
∆t
2. then, solve the non-linear vessel wall problem with a Neumann condition at the FS
interface:
ρs β b n ³ n´
d − ∇ · bs d
T b =g bns in Ωs , (51a)
∆t2³ ´
Tbs d bn n b=T b f (b
v n , pbn ) n
b on Σ. (51b)

Notice that the discretized-in-time kinematic condition (50c) differs from (48c) since now
we are considering an explicit Dirichlet condition for the fluid subproblem, so that the
right hand side is computed at the previous time step.
A ”parallel” version b v n , pbn ) n
b
³ ´ of the previous scheme is obtained by substituting T f (b
b
with T f v b n−1 n−1
, pb b in (51b). Notice that in the monolithic FSI problem (48), the
n
dynamic continuity condition (48d) is written in the current configuration Σ∗ , whereas
for the structure subproblem alone (51) is written in the reference configuration Σ. Ac-
cordingly, in what follows the structure interface quantities will be written in the current
configuration in monolithic FSI problems and in the reference configuration when the
structure problem is uncoupled in view of a partitioned scheme.
Unfortunately, the explicit Dirichlet-Neumann scheme can be unconditionally abso-
lutely unstable. In particular, in [85] it is proven that this happens if the fluid and
structure densities are comparable, which is precisely the case of hemodynamics (high
added mass effect, see also [194] for a discrete analysis and [414] for an analysis of the
added mass effect arising from a temporal discretization of order p ≥ 2).
Stable loosely-coupled algorithms have been recently introduced. To this aim, replace
in the FSI problem (48) the interface conditions (48c)-(48d) with two linear independent
combinations
³α ´
σf v n + T f (v n , pn ) n∗ = σf dn + g nfs + T s (dn ) n∗ on Σ∗ , (52a)
³α ´ ∆t
σs dn + g nfs + T s (dn ) n∗ = σs v n + T f (v n , pn ) n∗ on Σ∗ , (52b)
∆t
where σf 6= σs are, in general, two functions of space and time. This naturally leads to
the following

Explicit Robin-Robin scheme. For n ≥ 1, at time step tn :

47
 1. solve the Oseen problem with a Robin condition at the FS interface:
ρf α n
v + ρf ((v ∗ − v ∗f ) · ∇)v n − ∇ · T f (v n , pn ) = g nf in Ω∗f , (53a)
∆t
∇ · vn = 0 in Ω∗f , (53b)
³α ´ ¡ ¢
σf v n + T f (v n , pn ) n∗ = σf dn−1 + g fn−1s + T s dn−1 n∗ on Σ∗ ; (53c)
∆t

2. then, solve the non-linear vessel wall problem with a Robin condition at the FS
interface:
ρs β b n ³ n´
d − ∇ · T b =g
bs d bns in Ωs , (54a)
∆t2 ³ ´
σs α b n b b n b f (b
d + Ts d n bn + T
b = σs v v n , pbn ) n bnfs
b − σs g on Σ. (54b)
∆t

In [80], a Discontinuous Galerkin (DG)-like mortaring of the interface coupling condi-

tions (48c)-(48d) is proposed. The corresponding block Gauss-Seidel explicit algorithm
could be reinterpreted as an explicit Dirichlet-Robin scheme, where in (53c)-(54b) we
have σf = +∞, σs = − γµ h , γ being the DG penalty parameter. It is shown that the
DG interface penalty and the viscous dissipation are not able to control the pressure fluc-
tuations at the FS interface appearing in the discrete energy estimate. For this reason,
it is proposed to add to the fluid problem a consistent stabilization term penalizing the
pressure fluctuations, that is proven to be absolutely stable under a CFL-like condition. A
stable explicit Robin-Robin scheme has been introduced in [34], where the parameters in
the Robin interface conditions (53c)-(54b) were set after analyzing the incoming and out-
coming characteristic variables of the vessel wall problem. In particular, they are defined
in terms of the outgoing characteristic variable for the fluid subproblem and in terms of
the incoming characteristic variable for the vessel
p wall subproblem. For linear elasticity,
√
the following values are obtained: σf,norm = ρs (µ1 + 2µ2 ) and σf,tang = ρs µ2 in the
normal and tangential direction, respectively (µ1 and µ2 being the Lamé constants), and
σs = −σf . This choice allows the traveling information brought by the characteristic
variables to provide a more tightly coupling of the fluid and structure problems than the
one enforced by (48c)-(48d).

Partitioned algorithms of semi-implicit type. In [174], it is proven that to achieve

stability without stabilization terms for the Dirichlet-Neumann scheme, at least an im-
plicit coupling between the fluid pressure and the vessel wall displacement is mandatory.
In particular, a projection scheme is used where the fluid ALE-advection-diffusion step is
solved explicitly, whereas the fluid pressure and vessel wall subproblems are coupled and
solved in an iterative framework until convergence (see also [15] for a convergence analysis):

Semi-implicit pressure-vessel wall coupled scheme. For n ≥ 1, at time step tn :

48
 1. solve the ALE-advection-diffusion problem with a Dirichlet condition at the FS in-
terface:
ρf α n ³ ´
v n − ∇ · µ ∇e
e + ρf ((v ∗ − v ∗f ) · ∇)e
v v n )T = g nf
v n + (∇e in Ω∗f , (55a)
∆t
α n−1
en =
v d + g fn−1
s on Σ∗ ; (55b)
∆t

2. then, solve the coupled pressure-vessel wall problem. To this aim, introduce the
following iterations on index k ≥ 1:

a) solve the pressure problem with a Neumann condition at the FS interface:

ρf α
△pn(k) = en(k)
∇·v in Ω∗f ,
∆t
∂pn(k) α n
= d + g nfs on Σ∗ ;
∂n∗ ∆t (k−1)
b) then, solve the non-linear vessel wall problem with a Neumann condition at the
FS interface:
ρs β b n ³ n ´
− ∇ · bs db bns
d T (k) = g in Ωs ,
∆t2³ (k) ´ ³ ´
n
bs d bn n b b e , pbn(k) n
T (k) b = T f v b on Σ.

In [13], the DG-mortaring approach is applied to this projection scheme, leading to a

Robin-Robin-like scheme. An algebraic version of the projection scheme proposed in [174]
is introduced in [26].

Partitioned algorithms of implicit type. More in general, a fully implicit treatment

of (48c)-(48d) (or more in general of (52)) by means of partitioned algorithms is often con-
sidered. In this case, the fluid and vessel wall subproblems are solved iteratively until the
whole interface conditions are satisfied within a prescribed tolerance. A general scheme is
given by the following

Implicit Robin-Robin scheme. For n ≥ 1, k ≥ 1, at time step tn /iteration k:

1. solve the Oseen problem with a Robin condition at the FS interface:

ρf α n
v + ρf ((v ∗ − v ∗f ) · ∇)v n(k) − ∇ · T f (v n(k) , pn(k) ) = g nf in Ω∗f ,
∆t (k)
(58a)
∇· v n(k) =0 in Ω∗f ,
(58b)
³ ´ ³α ´ ³ ´
σf v n(k) + T f v n(k) , pn(k) n∗ = σf dn(k−1) + g nfs + T s dn(k−1) n∗ on Σ∗ ;
∆t
(58c)

49
 2. then, solve the non-linear vessel wall problem with a Robin condition at the FS
interface:
ρs β b n ³ n ´
− ∇ · bs d b bns
d T (k) = g in Ωs , (59a)
∆t2 (k) ³ ´ ³ ´
σs α b n bs d bn n bf v
d(k) + T (k) b = σs vbn(k) + T bn(k) , pbn(k) n bnfs
b − σs g on Σ. (59b)
∆t
As proven in [85], a small relaxation parameter is needed to achieve convergence in the
implicit Dirichlet-Neumann scheme (corresponding to setting σf = +∞, σs = 0 in (58c)-
(59b)). In practice, often an Aitken relaxation procedure is used to dynamically estimate
an efficient relaxation parameter [136, 313]. A better situation is obtained by properly
selecting the parameters in the Robin interface conditions (58c)-(59b). In particular, the
choice σf = ρs∆t
Hs EHs ∆t
+ (1−ν 2 )R2 , σs = 0 (where, as usual, E and ν are the Young modulus and

the Poisson ratio for the vessel material at small deformations, Hs and R representative
thickness and radius of the vessel) yields fast convergence without any relaxation (Robin-
Neumann scheme, [23, 410]). An optimization of σs is performed in [207, 211], leading to
a further improvement in the convergence history. A Dirichlet-Robin scheme is derived in
[610] by means of a generalized fictitious method, where the coefficients of the fluid pressure
and vessel wall acceleration are changed to account for the added mass effect. This allows
one to obtain again good convergence properties for hemodynamic parameters without any
relaxation. Another class of implicit methods with good convergence properties for high
added mass effect is based on adding a suitable interface artificial compressibility (IAC)
consistent term to the fluid problem, proportional to the jump of pressure between two
successive iterations [131]. In [130], it is showed that for a finite volume approximation,
IAC method based on Dirichlet-Neumann iterations is equivalent to a Robin-Neumann
scheme for a suitable choice of the parameter σf .

Partitioned algorithms for the FSI problem with a membrane structure. Re-
cently, several papers have analyzed algorithms for FSI problems featuring a reduced mem-
brane model for the vessel wall. In this case, the FSI problem is given by the fluid problem
(22a), by the kinematic continuity condition (26a) and by the membrane equation (26b),
which in this case plays also the role of dynamic continuity condition. Moreover, homoge-
neous Dirichlet or Neumann conditions in the tangential direction need to be prescribed
for the fluid problem at the interface Σ. By considering an implicit time discretization of
(26b) and an explicit treatment of the geometry coupling, we obtain at each time step the
following linearized FSI problem:
ρf α n
v + ρf ((v ∗ − v ∗f ) · ∇)v n − ∇ · T f (v n , pn ) = g nf in Ω∗f , (60a)
∆t
∇ · vn = 0 in Ω∗f , (60b)
α
v n · n∗ = dr n + gfns on Σ∗ , (60c)
∆t
v n − (v n · n∗ )n∗ = 0 on Σ∗ , (60d)
ρs H s β n
d − ∇ · (P ∇dnr ) + χHs dnr = −T f (v n , pn ) n∗ · n∗ + gsn on Σ∗ , (60e)
∆t2 r

50
where, as usual, g nf , gsn and gfns account for the terms at previous time steps coming from
time discretization.
The explicit Dirichlet-Neumann scheme applied to the previous monolithic problem
reads:

Explicit Dirichlet-Neumann scheme for the FSI problem with membrane structure. Given
the quantities at previous time steps, at time step tn :

1. solve the Oseen problem with a Dirichlet condition at the FS interface:

ρf α n
v + ρf ((v ∗ − v ∗f ) · ∇)v n − ∇ · T f (v n , pn ) = g nf in Ω∗f ,
∆t
∇ · vn = 0 in Ω∗f ,
α n−1
v n · n∗ = d + gfn−1 on Σ∗ ,
∆t r s
n n ∗ ∗
v − (v · n )n = 0 on Σ∗ ;

2. then, solve the membrane problem:

ρs Hs β bn
d − ∇ · (P ∇dbnr ) + χHs dbnr = −T
b f (b
v n , pbn ) n b + gbsn
b ·n on Σ.
∆t2 r

As in the case of scheme (50)-(51), unfortunately this scheme is unconditionally absolutely

unstable in the hemodynamic regime [85].
Different algorithms are obtained by linearly combining the interface conditions (60c)
and (60e) and by substituting the new condition to (60c). In this case, we have to solve a
coupled problem consisting of:

Robin-Neumann coupling for the FSI problem with membrane structure. Given the quan-
tities at previous time steps, at time step tn :

1. solve the Oseen problem with a Robin condition at the FS interface:

ρf α n
v + ρf ((v ∗ − v ∗f ) · ∇)v n − ∇ · T f (v n , pn ) = g nf in Ω∗f , (62a)
∆t
∇ · vn = 0 in Ω∗f , (62b)
n n n ∗ ∗
(σf v + T f (v , p ) µn )·n = ¶
³α ´ ρs H s β n on Σ∗ , (62c)
n n n n
σf d + gf s − d − ∇ · (P ∇dr ) + χHs dr
∆t r ∆t2 r
v n − (v n · n∗ )n∗ = 0 on Σ∗ ; (62d)

2. solve the membrane problem:

ρs Hs β bn
2
dr − ∇ · (P ∇dbnr ) + χHs dbnr = −T
b f (b
v n , pbn ) n b + gbsn
b ·n on Σ. (63)
∆t

51
The previous problem could be solved either monolithically or by means of a block Gauss-
Seidel method that in fact introduces subiterations splitting the solution of (62) and (63).
When P = 0, the special choice
ρs Hs β χHs ∆t
σf = σ
ef = + , (64)
α∆t α
introduced in (62c), yields to

σf v n + T f (v n , pn ) n∗ ) · n∗ = σ
(e ef gfns on Σ∗ . (65)

At this stage, this is a Robin condition for the fluid problem without any explicit de-
pendence on dnr . Thus, the monolithic problem given by (62)-(63)-(64) is equivalent to
the stand-alone fluid problem (62a)-(62b)-(62d)-(65), see [413]. The solution of this fluid
problem can then be used to feed the right hand side of (63) and to get the structure
displacement dbnr . In this way, the fluid and structure problems are in fact decoupled, even
if the coupling conditions are treated implicitly. This provides a smart and efficient way
to solve the monolithic problem (60) exactly, at the expense of a single fluid problem solve
(note that the membrane problem (63) is solved very cheaply).
Starting from this result, a stable Robin-Neumann scheme based on an operator-
splitting has been proposed in [233] for a general membrane law (P 6= 0). The inertial
vessel wall term is treated implicitly as in the previous case leading to a Robin boundary
condition for the fluid with σf = ρsα∆t Hs β
, whereas the elastic and algebraic contributions
are treated explicitly. In [173], an incremental version of this scheme is proposed, where
the elastic and algebraic parts of the membrane law are included in the Robin condition
for the fluid problem by means of a suitable extrapolation from previous time steps. Fi-
nally, we mention [113], where the whole membrane law is treated implicitly, leading to a
generalized Robin condition, which however requires an ad-hoc implementation.

Partitioned algorithms based on the Schur complement. We discuss here parti-

tioned schemes arising from an interface equation written for the FSI problem and intro-
duced in [137]. For the sake of exposition, we introduce the algebraic counterpart of (48)
related to a Finite Element discretization for the case of linear elasticity for the vessel wall
problem:
 ff  f   f 
C CfΣ 0 0 V b
    
  Σ   
 0 M Σ −M Σ 0  V   0 
   
 
  = , (66)
    
 C Σf C ΣΣ GΣΣ GΣs   U Σ   bΣ 
    
    
0 0 GsΣ Gss Ds bs
where the subscript f refers to internal degrees of freedom (d.o.f.) for the fluid, s to internal
d.o.f. for the vessel wall, Σ to the FS interface d.o.f.; the notation C (with subscripts)
refers to fluid matrices, G to vessel wall matrices, M Σ is the FS interface mass matrix; V
collects the fluid d.o.f. (in this case also the pressure ones), D s the vessel wall displacement

52
internal d.o.f., and U Σ the vessel wall velocity interface d.o.f.; vectors b represent the right
hand sides. The first row corresponds to the momentum and mass conservation for the
fluid, the second and third rows to the interface kinematic and dynamic conditions, and
the last row to the vessel wall problem. By eliminating V f , V Σ and D s from (66), we
obtain the following interface equation
eΣ + G
(C eΣ,
e Σ )U Σ = b (67)

where Ce Σ = C ΣΣ − C Σf (C f f )−1 C f Σ and G

e Σ = GΣΣ − GΣs (Gss )−1 GsΣ are the fluid and
vessel wall Schur complement matrices and b e Σ = bΣ − C Σf (C f f )−1 bf − GΣs (Gss )−1 bs
the corresponding right hand side.
The Robin-Robin method (58)-(59) could be obtained by applying the Richardson
method to the interface equation (67) preconditioned by the matrix
1 ³ ´ ³ ´
P = e Σ + σf M Σ (M Σ )−1 G
C e Σ + σs M Σ . (68)
σf + σs
This leads to a new family of partitioned schemes obtained by applying other Krylov
methods to (67) with the same preconditioner. For example, GMRES preconditioned
by the Dirichlet-Neumann preconditioner P = G e Σ is considered in [25], whereas GMRES
preconditioned with the RR preconditioner (68) is introduced in [24]. The first of these two
schemes highlighted better convergence properties with respect to the classical Dirichlet-
Neumann method, whereas the second one is more robust with respect to the choice of
the interface parameters σf , σs than the classical Robin-Robin method.
In some works, a non-linear interface equation is written directly for the non-linear
problem (47) and a Newton method is then applied to this equation [176, 132].
Remark 2. In the partitioned schemes introduced above, the vessel wall subproblem ap-
pearing at each iteration is still non-linear, i.e. we did not discuss how to tackle point iii)
in the list of issues reported at the begin of the paragraph. The simplest way to treat the
vessel wall non-linearity, very appropriate when a non-linear structural solver is available,
is to introduce inner Newton subiterations at each Robin-Robin iteration. Alternatively, a
scheme based on managing together the five sources of non-linearities/coupling (geometric
coupling and adherence, fluid and structure constitutive non-linearities, physical coupling)
in a single inexact-Newton loop is proposed in [312]. Other possible combinations will be
discussed later on as specific monolithic procedures. For a detailed overview with numerical
comparisons we refer to [411].

Monolithic solvers. As an alternative to partitioned schemes, monolithic procedures

have been successfully considered in hemodynamics for the solution of the FSI problem
(48). These methods consist in applying an exact or inexact Newton method to the whole
non-linear FSI problem. Referring to the notation introduced above, by considering a
Finite Element space discretization and the inexact Newton method, we obtain from (48)
(we suppress for the sake simplicity the temporal index n)
e (k−1) )δx(k) = G − A(x(k−1) ),
J(x (69)

53
where Je is the Jacobian matrix or a suitable approximation of it, x = [V f V Σ U Σ D s ]T ,
and A(x) = G is the non-linear system related to (48). The exact Jacobian matrix is
given by  ff 
C CfΣ 0 0
 
 
 0 MΣ −M Σ 0 
 
 
J(x(k−1) ) =  , (70)
 C Σf C ΣΣ J ΣΣ (U Σ Σs s 
 G (k−1) ) JG (D (k−1) ) 
 
 
sΣ Σ ss s
0 0 JG (U (k−1) ) JG (D (k−1) )
where the submatrices JG are the exact Jacobians of A related to the structure d.o.f. For
the solution of the linear systems (69) with the exact Jacobian, classical strategies used so
far for hemodynamics are, e.g., GMRES preconditioned by a one-level additive Schwarz
method [36] and global algebraic multigrid [203]. In [37], a two-level Newton method is
used in combination with a two-level hybrid Schwarz preconditioner, where the solution on
a coarse grid is used to provide a good initial guess to the Newton method. Alternatively,
inexact Newton methods have been considered, e.g. by means of block approximations Je
of J splitting the fluid velocity, pressure and vessel wall unknowns [246, 125].
A particular class of inexact Newton methods is obtained by neglecting the term −MΣ
appearing in the upper-right block of the exact Jacobian (70). This yields in fact a class
of partitioned schemes of Dirichlet-Neumann type where issues iii) and v) (vessel wall
non-linearity and physical coupling) are treated within the same iterations. An example
is given by the preconditioner proposed in [125], where the fluid and structure blocks in
(70) are approximated by the corresponding algebraic additive Schwarz preconditioners.
This method is strongly scalable for hemodynamic applications. Recently, a variant of the
previous preconditioner has been introduced in [138]. This new preconditioner, named
FaCSI, is based on operating a static condensation of the fluid interface d.o.f. and using
a SIMPLE preconditioner for the fluid block.
For the sake of exposition, we have discussed numerical strategies for the FSI problem
based on Finite Elements for the space discretization and Finite Differences for the time
discretization. Other strategies considered so far in hemodynamics are space-time Finite
Elements, see, e.g., [544, 42], and the methods based on the iso-geometric analysis, see
[40, 41].
In Figure 10 we report a couple of examples of numerical results obtained in real
geometries reconstructed from radiological images (see the caption for details).

4.6.4 Numerical methods for defective boundary problems

For the numerical solution of the augmented formulation (30), one could rely either on a
monolithic strategy where the full augmented matrix is built and solved, or on splitting
techniques. As for the latter choice, in [184, 561] it is proposed to write the Schur com-
plement equation with respect to the Lagrange multiplier of the linearized and discretized
(in time and space) augmented formulation. This is a linear system whose dimension is

54
Figure 10: a) Blood velocity streamlines and vessel wall displacement vectors
in a stenotic carotid artery (MRI images from the Vascular-surgery and Radi-
ology Divisions at Ospedale Maggiore Policlinico, Milan, Italy); b-c-d) Results
of a FSI simulation in the ascending and thoracic aorta (MRI images from
[Link] =repository). Blood velocity mag-
nitude in the whole domain (b) and on a selected longitudinal section (c), vessel wall dis-
placements (d). All cases refer to the systolic peak. These numerical results are obtained
using LifeV, P 1−Bubble/P 1 Finite Elements for the fluid problem and P 1 Finite Elements
for the vessel wall problem; the backward Euler scheme and the midpoint method have
been used for the time discretization of the fluid and vessel wall problems, respectively.
The implicit Robin-Robin partitioned scheme has been used in case a) and the FaCSI
preconditioner in cases b-c-d)

equal to the number of flow rate conditions, say m ≥ 1. By applying the GMRES method
to iteratively solve this system, the exact solution is reached after exactly m iterations (in
exact arithmetics). At each iteration, the solution of a standard fluid problem with Neu-
mann conditions is needed (exact splitting technique). The solution of a further standard
fluid problem is required to compute the initial residual in the GMRES algorithm. This
approach is quite expensive, even in the case m = 1, which requires the solution of two fluid
problems per time step. However, it preserves modularity, indeed it can be implemented
using available standard fluid solvers in a black box mode. This is an interesting property
when applications to cases of real interest are addressed, see [576, 569, 453, 226, 227].
To reduce the computational time required by the exact splitting approach, in [562]
a different (inexact) splitting procedure is proposed, requiring the solution of m steady
problems out of the temporal loop and of one unsteady null flow rate problem at each
time step. This strategy introduces an error near the section which is smaller than the
one based on conjecturing the velocity profile in the original (non-null) flow rate problem,

55
see Sect. 4.4.1.
As for the numerical solution of the control-based approach described in Sect. 4.4.3,
classical iterative methods have been considered in [190, 191] for the solution of the re-
sulting KKT system.
Recently, a numerical approach based on the Nitsche method has been considered to
prescribe a flow rate condition. In particular, the original idea of prescribing Dirichlet con-
ditions with a consistent penalization approach [408] is extended to the case of flow rate
boundary conditions in [615]. This strategy does not introduce further variables, other
than those of the original problem, however it requires to properly tune a penalization
parameter. In addition, it deals with non-standard bilinear forms that need ad hoc imple-
mentation. However, it should be very effective if flow rate conditions are implemented in
a DG code. A similar approach has been considered to fulfill the mean pressure condition
(3) and the FSI case in [565], see also [459].
For a more comprehensive overview of numerical strategies for defective boundary
problems, we refer the reader to [192].

4.6.5 Numerical methods for the geometric reduced models and multiscale
approach
For the numerical solution of the 1D reduced model (33), in principle any convenient
approximation method for non-linear hyperbolic equations can be used. The peculiar fea-
ture of this model, however, is the lack of discontinuous solutions. A common approach
relies on the Finite Element version of the Lax-Wendroff scheme, thanks to its excellent
dispersion properties [183]. Being this scheme explicit, a CFL-like condition is required to
ensure absolute stability. In presence of a visco-elastic term, the 1D model is usually dis-
cretized by means of a splitting procedure where the solution is split into two components,
one satisfying the pure elastic problem and the second one the visco-elastic correction
[185, 352]. High order method are suitable to capture the (physical) reflections at bifur-
cations induced by the vessel tapering, see e.g. [519, 518] for a Discontinuous Galerkin
discretization and [387] for a finite volume scheme.
Regarding 0D models, they are in general described by systems of DAE (differential
and algebraic equations), possibly non-linear due to the presence of diodes to represent
the valves [188]. Usually, for hemodynamic applications, these systems can be reduced to
classical Cauchy problems and solved by classical Runge-Kutta methods.
As for the solution of the 3D-1D coupled problems described in Sect. 4.5.2, we can
in principle identify three different strategies, namely partitioned schemes, monolithic
approaches, and methods based on the solution of an interface equation. In partitioned
schemes, the 3D and 1D problems are solved separately in an iterative framework. The
coupling interface conditions can be enforced in many different ways: for example, we
can prescribe the flow rate condition (42a) to the 3D problem and the pressure condition
(45) to the 1D problem. Different algorithms are obtained by switching the role of the
interface conditions in the iterative algorithm or by considering other interface conditions
(e.g. (42b)). This is also the case when one of the two interface conditions is replaced
by a condition expressing the continuity of the characteristic variable W1 entering the 1D

56
domain [183, 431], i.e., according to (36),
à Z !
¯ ¯
W1 (t)|z=0 = ζ1 ¯Γtf ¯ , v · n dγ .
Γtf

In any case, each of these approaches yields a 3D problem with a defective boundary
condition, which could be dealt with one of the strategies described in Sect. 4.4. Explicit
algorithms based on the solution of the 3D and 1D problems only once per time step have
been successfully considered in [186, 431]. These algorithms enforce a limitation on ∆t,
which, however, is milder with respect to the one imposed by the numerical scheme adopted
for the 1D model. Alternatively, iterative methods applied directly to the monolithic
linearized system have been introduced in [57, 58]. A different approach to solve the 3D-1D
coupled problem relies on writing an interface equation involving only the 3D-1D interface
unknowns. We can interpret this equation as the geometric heterogeneous counterpart
of the Schur complement equation. For its numerical solution, either the Broyden or the
Newton method have been used in [330, 351, 55], in combination with GMRES. Methods
relying on the numerical solution of the interface equation are simple to implement in the
case of multiple interfaces, such as those that arise in complex arterial networks.
In Figure 11 we report a numerical result obtained by the coupling between the 3D
model of an ascending aorta and a 1D model of the systemic circulation. This result
highlights the suitability of the 1D model in providing absorbing conditions to the 3D
model and in propagating the pressure wave along the whole network [350].

Figure 11: Pressure wave propagation in an ascending aorta (3D model) and in the 1D
model of the systemic circulation. These numerical results are obtained using LifeV; the
Newton method has been used for the interface equation. Courtesy of C. Malossi

57
Part II
THE HEART FUNCTION

5 Basic facts on quantitative physiology

5.1 Basic anatomy
The heart is a hollow organ that pumps the blood into the arteries of the systemic and
pulmonary circulations and collects it after its return through the veins. It is formed by
the left and the right heart, each of them being composed by two chambers, an atrium and
a ventricle. The left and right hearts are separated by the interatrial and interventricular
septa, which do not allow the transfer of blood, whereas the atria and the ventricles are
connected by the atrio-ventricular valves (tricuspid valve in the right heart, mitral valve
in the left heart) that either allow or prevent the blood transfer from the atria to the
ventricles depending on the their position (open or closed, respectively), see Figure 12. In
particular, they open when the atrial pressure is higher than the ventricular one and close
as soon as the blood flow rate, normally going from the atrium to the ventricle, becomes
negative, i.e. coming back in the atrium. The papillary muscles, located in the ventricles,
attach to the cusps of these valves via the chordae tendineae preventing their inversion
(prolapse) during the closure.

Figure 12: Schematic representation of the heart

All the four chambers are connected with the circulatory system: the left ventri-
cle through the aorta, the right ventricle through the pulmonary artery, the left atrium
through the pulmonary veins, and the right atrium through the superior and inferior ve-
nae cavae. The ventricles are separated by the circulatory system by two further valves,

58
the aortic valve on the left side and the pulmonary valve on the right side, whose open-
ing/closure mechanism is similar to the one of the atrio-ventricular valves, i.e. they open
when the pressure is higher in the ventricle with respect to the corresponding connected
artery, whereas they close when the flow rate becomes negative (i.e. going from the artery
to the ventricle). No valves are located between the atria and the corresponding terminal
veins.
The heart wall is composed by three layers: the internal thin endocardium, the thick
muscular myocardium, and the external thin epicardium. The myocardium of the left
ventricle is almost two times thicker than the one of the right ventricle. The epicardium
is surrounded by the pericardium, a serous membrane that isolates the heart from the
closest organs, facilitating its movements.

5.2 The cardiac cycle

The main purpose of the heart is to pump the blood in the circulatory system through
the aorta and the pulmonary artery. To do this, it needs to win the resistances in the
arteries where blood has a non-null pressure (about 70 mmHg in the aorta, 10 mmHg in
the pulmonary artery) due to the reaction of the elastic vessel wall to the deformation
induced by blood inside. Moreover, the heart supplies to blood the energy needed to reach
the microvasculature or the lungs. The total work made by the heart per heartbeat is
1
W = QP + mv 2 ,
2
Q and m being the blood flow rate and mass ejected, P the arterial blood pressure, and
v the blood velocity in the aorta or pulmonary artery. The kinetic component of the
work is quite negligible with respect to the potential one (about 2% for the left heart and
about 5% for the right heart), although it could become more relevant (up to 25%) under
physical effort. The flow rate Q in normal conditions is about 5 · 103 cm3 /min, so that,
assuming 60 − 90 heartbeats per minute, we have about 55.5 − 83.3 cm3 of blood expelled
at each heartbeat. The energetic requirements to pump blood are obtained, as it happens
in all the organs, by the consumption of oxygen that is provided to the heart by blood in
the coronary arteries. The efficiency of the heart, i.e. the ratio between the energy W and
the total energy consumption, is in physiological cases about 25%.
The cardiac cycle comprises different phases. For its description, we consider the left
heart and we start from the situation where the left atrium is filled by oxygenated blood
coming from the pulmonary veins, the mitral valve is closed, and the left ventricle has
reached its maximum contraction.

1. Ventricular filling. Due to the continuous inflation of blood, when the pressure in the
atrium exceeds the ventricular one (about 10 mmHg) the mitral valve opens. Blood
starts filling the left ventricle which relaxes. In a first stage (∼ 0.15 s), there is a fast
inflation of blood due to the pressure gradient between the atrium and the ventricle,
where probably the ventricle exerts also a suction due to its expansion (phase 1a
in Figure 13). In any case, this is a passive phase. In a second stage (∼ 0.25 s,

59
 phase 1b in Figure 13), the atrium actively contracts (atrial systole) producing a
slow ventricular inflation of the remaining blood;
2. Isovolumic contraction. After the atrial systole, the ventricular active contraction
starts. This produces an increase of the ventricular pressure causing retrograde flow
that accordingly closes the mitral valve. However, the ventricular pressure is still
lower than the aortic one, so that also the aortic valve is closed. Thus, during this
phase, there is a continuous and fast increase of the ventricular pressure without any
change of blood volume due to blood incompressibility (∼ 0.05 s, phase 2 in Figure
13);
3. Ventricular ejection. As soon as the ventricular pressure reaches the aortic pressure
(about 70 mmHg), the aortic valve opens and blood is ejected in the systemic cir-
culation. Since the ventricular contraction carries on also after the valve opening,
the ventricular pressure continues to increase. Accordingly, also the aortic pressure
increases due to the elastic adaption of the vessel wall. However, at each time, there
is a pressure difference between ventricle and aorta that allows blood to accelerate
(∼ 0.05 s, phase 3a in Figure 13). When the ventricle stops its active contrac-
tion, this pressure difference reduces and, after a short period (but not immediately
due to inertial effects), becomes zero, allowing the flow rate to reach its maximum
(∼ 0.05 s, phase 3b in Figure 13). Then, the pressure difference starts to become
negative (higher in the aorta) provoking a deceleration of blood that however con-
tinues to enter the aorta. When, due to this deceleration, the flow rate becomes
negative, the aortic valve closes and the blood ejection stops (∼ 0.15 s, phase 3c in
Figure 13). Notice that, once the active contraction stops, the ventricle starts to
relax releasing the elastic energy accumulated during the contraction. However, this
energetic relaxation is not immediately followed by a mechanical relaxation which
should result in an increase of the ventricular volume. Indeed, due to inertial effects,
there is a “passive” mechanical contraction which allows to eject all the blood.
4. Isovolumic relaxation. When the aortic valve closes, the release of energy of the
ventricle continues with both valves closed, so that no changes of ventricular volume
are produced resulting in a fast decrease of the ventricular pressure (∼ 0.10 s, phase 4
in Figure 13). After the valve closure, the aortic pressure keeps (slightly) increasing,
due to the elastic recoil of the closing valve (dicrotic wave, see Figure 13).
All these mechanisms apply to the right heart as well, the only difference being the
pressure values, smaller in this case. This justifies the thicker myocardium of the left heart
since higher values of resistances need to be won.

5.3 Electric propagation

As seen in the previous paragraph, the main responsible for blood ejection in the circula-
tory system is the active ventricular contraction. The heart is able to produce itself the
electric impulse that determines this contraction, triggered by an electric potential that
propagates along all the myocardium. This is possible owing to the excitability of the heart

60
Figure 13: Aortic pressure, ventricular pressure, atrial pressure, and ventricular volume
during an heartbeat

cells, the cardiomyocites, that, when suitably stimulated, are able to produce a variation of
the membrane voltage. At rest, the membrane potential is negative (∼ −90 mV ), whereas
when stimulated it reaches a positive value (∼ 20 mV ) in a very short period (about
2 ms). After this depolarization, a plateau around 0 mV is observed that corresponds to
the refractory period (see below). Then, the repolarization phase starts that brings the
potential back to the rest value allowing for a new excitation (see Figure 14, left). This
action potential is generated by several ion channels that open and close and by the result-
ing currents passing through the membrane. The most important channels are those of
calcium, sodium, and potassium. In particular, a fast inward sodium current is the main
responsible for the rapid depolarization, a slow inward flux of extra-cellular calcium ions
is the main responsible for the characteristic plateau appearing after the depolarization,
whereas the outward potassium currents are responsible for the repolarization.
Unlike other cells in the human body, the cardiomyocites obey to the “all-or-none” law,
meaning that if the stimulus is above a suitable threshold, a complete action potential with
peak value independent of the stimulus is generated, otherwise no response is provided by
the cell. Another characteristic of the heart cells is the presence of a refractary period after
the generation of an action potential, which inhibits any further stimulus independently of
its intensity. Thus, during this period the cell is non-excitable at all (absolute refractary
period, infinite threshold). Afterwards, the cardiomyocites recover their excitability with
a value of the threshold needed to generate the action potential which decreases in time
(relative refractory period). Finally, once the threshold value reaches its minimum, the
cells returns to its complete excitable state and the threshold remains constant.
The cardiomyocites act as a syncytium, i.e. the stimulation of an individual cell
produces the action potential and the corresponding current that result in the excitation

61
Figure 14: Characteristic action potential of cardiomyocites (left) and anatomy of the
cardiac conduction system ([Link] right)

of the neighbor cells, and thus of the whole myocardium. This is allowed by the gap
junctions, intercellular channels characterized by a low resistance and located between
cardiomyocites that permit the electric potential to travel on the cellular membranes from
cell to cell.
In normal conditions, the signal spontaneously originates at the sinoatrial node, lo-
cated in the right atrium at the junction with the superior vena cava. It represents the
natural pacemaker of the heart and imposes its rhythm to all the myocardium (sinusal
rhythm, ∼ 60 − 90 heartbeats per minute). The impulse generated by the sinoatrial node
propagates through all the cardiomyocites of the atria, activating all their regions and
allowing their contraction. The propagation is faster in the direction of the ventricles
(∼ 200 cm/s) allowing to reach the atrioventricular node, located between the atria and
the ventricles. When the signal arrives at this node, it is subjected to a delay (∼ 0.12 s)
that allows the complete contraction of the atria before the propagation in the ventricles
starts. Moreover, this node provides a filter to possible high frequencies of the atrial signal,
induced e.g. by atrial fibrillation, protecting the ventricles. This node, when the sinoatrial
node looses its automatism, becomes the leading pacemaker and takes on the role of giving
the pace to all the ventricle stimulation. Then, the electric signal enters the bundle of His,
propagating in the ventricles through the two (left and right) bundle branches and then
through the Purkinje fibers (see Figure 14, right). The bundle of His, bundle branches and
Purkinje fibers form the cardiac conduction system (CCS), a subendocardic specialized
network responsible for the fast and coordinated propagation of the electric impulse in
the ventricle. The propagation in the CCS is very fast (∼ 350 cm/s) and its role is to
reach the entire endocardium through the dense network of Purkinje fibers, activating it
almost simultaneously. Notice that the cells of the CCS are specialized in the electric
propagation so that they are not involved in the muscular contraction. Then, the electric
signal enters the myocardium through the Purkinje muscle junctions (PMJ) which are the
terminal points of the Purkinje network. At the PMJ, the signal is subjected to a delay
(∼ 0.01 s) and then propagates into the ventricular muscle towards the epicardium with a

62
reduced velocity (∼ 80 cm/s).
To better understand how the propagation of the electric potential spreads into the
ventricles, we observe that cardiomyocites are of cylindrical type. This allows us to define
the fiber direction of the cell, resulting in a macroscopic fiber direction intended as the
average cell orientation in a sufficiently small control volume. A transmural variation of
the fiber direction is measured between the epicardium and the endocardium (∼ −70o and
∼ 80o with respect to the normal direction to the surface, respectively). The fibers are in
turn organized in sheets of collagen. The velocity of propagation of the electric potential
is about two times faster along the fiber direction than in the directions tangential to the
fibers.

5.4 Mechanisms of contraction and cardiac blood fluid-dynamics

The propagation of the electric signal through the cardiomyocites is responsible of their
contraction (electro-mechanical coupling), resulting in the atrial and, more importantly,
ventricular contractions. The characteristic connections between the cardiomyocites re-
sulting in a syncytium allows for a coordinated contraction of the heart.
At the cellular level, as discussed in the previous paragraph, there is an inward flux of
extra-cellular calcium ions just after the depolarization of the cell. Once in the intracellular
space, calcium ions bind to troponin, which allows myosin to bind to actin and contraction
of the cell to occur.
At the macroscopic level, the ventricles contraction results in a longitudinal (apex-
to-base) shortening of about 15% from the diastolic configuration, in a ventricular wall
thickening of about 30%, and in a torsion around the longitudinal axis of about 15o . This
is due to the particular fiber orientation of the cardiomyocites which highly influences also
the mechanical response of the heart.
The heart obeys to the Frank-Starling law, stating that an increase (for any reason) of
the end diastolic volume (i.e. the maximum ventricular expansion) results in an increase
of the stroke volume, i.e. the volume of blood pumped by the left ventricle per heartbeat.
This is due to an increase of load experienced by the cardiomyocites as a result of the
increased volume of blood filling the ventricle. This extra-stretching of the cells produces
a greater number of actin-myosin bridges resulting in an augmented muscle contraction
power. It should be noticed however that the Frank-Starling law holds true only up to a
certain level of increased stretch: for higher values, the contractility power of the heart,
and thus the stroke volume, decreases.
Besides the influence of electric processes on the mechanical behavior, cardiomyocites
experience also a mechano-electric feedback. This is due to the formation of stretch-
activated ion channels and to changes of cells conductivity by means of the stretch of
the cell that changes the membrane shape and distance between gap junctions. This
mechanism seems to be of utmost importance to describe evolution of spiral waves and
thus arrhythmias [549, 289].
Blood flow in the cardiac chambers features different characteristics with respect to
the vascular one. In particular, peculiar challenges are the large-scale motion and complex
deformation of the myocardium, the complex interaction with valves leading to specific

63
flow patterns, and the higher Reynolds number (about 4000). For some authors, all
these aspects are responsible for transition to turbulence effects, especially in pathological
conditions, even if not all the authors agree with the fact that also in normal conditions
turbulence could occur. Another important aspect of cardiac blood flow is given by the
formation of a large vortex ring in the long axis plane of the left ventricle due to the
asymmetry of the mitral valve leaflets that forces the blood flow jet to impinge on the
posterior wall [91]. The interaction of this ring with the wall gives rise to a complex flow
pattern. Also in the left atrium, complex vortex rings ejected by the pulmonary veins can
be observed [375].

5.5 A brief summary of heart diseases

The principal heart diseases (cardiopaties) are the ischemic cardiopathy, the cardiomy-
opaties, the hypertensive cardiopathy, and the valvular pathologies.
The ischemic cardiopathy is the most important, for frequency and clinical impact,
among cardiopathies. It is caused by a reduced coronary flow rate (due to the atheroslero-
sis of coronaries) with consequent malnutrition of the myocardium. When, due to the
coronary occlusion, the decrease of the oxygen supply is abrupt, long-lasting, and total,
the infarct of the myocardium occurs. This is the most classical event of the ischemic
cardiopathy resulting in the necrosis of cardiomyocytes and leading, at the end of the
process, to the formation of a scar.
In the dilated cardiomyopathy the ventricle wall becomes thinner due to the partial
substitution of cardiomyocites by fibrosis and small scars, leading to the dilation of the
ventricle. In the hyperthrophic cardiomyopathy the ventricular wall becomes thicker, with
a consequent increase of the ventricular blood pressure resulting in a decreased filling of
the ventricle. As in all the cases of ventricular hyperthropy, this produces a malnutrition of
the myocardium, since the amount of blood supplied by the coronaries remains unchanged.
Moreover, in this specific hypertrofic cardiomiopathy, the orientation of the fibers is not
coordinated as in normal conditions (this is called electric disarray).
In the hypertensive cardiopathy, there is an increase of blood pressure resulting in a
increased work needed by the heart to pump blood, leading again to the thickening of the
myocardial wall and to a malnutrition of the myocardium.
Valvular main pathologies are stenosis and insufficiency. In the aortic valve stenosis,
the narrowing of the maximum opening of the aortic valve causes the heart to require an
increased work to properly pump blood into the aorta, resulting again in the thickening of
the myocardial wall and malnutrition. In the mitral valve stenosis, there is an increase in
the pressure of the left atrium and in the pulmonary circulation, with consequences on the
right heart. Aortic and mitral valve insufficiency is due to the partial reversal flow (going
from the aorta to the left ventricle in case of aortic insufficiency, from the left ventricle to
the left atrium in mitral insufficiency), that occurs as a consequence of the non adequate
closure of the valve. In this case, the heart dilates in order to supply this retrograde
flow resulting in an increase of its contraction power owing to the Frank-Starling law.
However, sooner or later, the diastolic cell stretching overcomes the threshold of validity
of the Frank-Starling law, resulting in a decreased contractility power.

64
 The two main events, possibly fatal, that heart may encounter as a consequence of
these cardiopaties are the cardiac arrest, due to ventricular fibrillation, and the heart
failure.
In the ventricular fibrillation, the cardiomyocites are not excited in a coordinate way
and thus do not contract homogeneously, but, continuously, groups of cells contract
whereas other groups relax. This chaotic excitation, if the fibrillation perpetuates, in-
hibits the normal functioning of the heart pump that is no longer able to regularly pump
the blood in the circulatory system, finally leading to death. All the cardiopaties described
above could yield altered electric properties (i.e. velocity of conduction and refractariness)
which could lead to ventricular fibrillation. For example, in the myocardial infarct, the
arrhythmogenic substrate is given by the vicinity of healthy cardiomyocites, necrotic car-
diomyocites, and scars, whereas in the hypertrofic cardiomiopathy it is provided by the
electric disarray.
When the blood pumped by the heart at each heartbeat is below the required amount
needed by the body (but not absent as in ventricular fibrillation), the heart on the one
hand increases its frequency so as to guarantee that the blood ejected per minute (cardiac
output) is almost normal, and on the other hand it increases its diastolic filling to increase
the stroke volume. However, the increase of the frequency (tachycardia) is energetically
disadvantageous for the heart, and a possible excessive diastolic cell stretching leads to
the loss of validity of the Frank-Starling law. When these two compensating mechanisms
become no longer effective, the cardiac output dramatically decreases leading to heart fail-
ure. In absence of other causes of death (cardiac arrest or death for non-cardiac reasons),
all the cardiopaties described above, sooner or later, leads to heart failure.

6 Landscape on data
As discussed in Section 3, the use of patient-specific data is mandatory in view of numerical
modeling aiming at understanding biophysical processes and supporting clinicians. This
holds true also for the heart modeling. The latter accounts for three processes that are
intimately coupled: the electrical propagation, the mechanical contraction and relaxation,
and the blood fluid-dynamics. The first two processes occur in the cardiac tissue (here
called muscle region), whereas the latter occurs in the four hollow regions (chambers)
delimited by the endocardium. As done in Section 3 for the circulatory system, in what
follows we briefly discuss how to obtain geometric, boundary, and biological cardiac data.

6.1 Cardiac geometric data

Referring to Figure 15, we need to build two computational domains, the muscle region
Ωmus , delimited by the external surface of the epicardium Σepi and by the internal surface
of the endocardium Σendo , and the cavities Ωcav , referred in what follows to as heart
cavities or chambers, delimited by the endocardium. The cavities are easily obtained once
the endocardium has been reconstructed, so that in fact the cardiac image reconstruction
process relies on identifying the endocardium and the epicardium surfaces.

65
 Figure 15: Longitudinal section of a complete heart domain

This problem features several challenging issues. First of all, unlike the vascular case
where the external wall surface is usually obtained by extruding the internal one under
the assumption of constant (or in any case known) wall thickness, for the heart it is
crucial to identify and reconstruct the epicardium. This is mandatory because of the high
variability of the myocardial thickness (both in terms of latitude and of patient variability)
that makes the extrusion of the endocardium not meaningful. Second, the motion of the
heart produces large displacements. This means that if one is interested in obtaining
a reconstruction of the heart not only at the end-diastolic phase, a dynamic acquisition
procedure (allowing for the acquisition of several frames per heartbeat) is required. Third,
the presence of the papillary muscle and wall irregularities given by trabeculations makes
the reconstruction of the endocardium very problematic.
As in the vascular case, the two most common radiological techniques used for acqui-
sition of cardiac images are MRI and CT. Often, due to the heart motion, temporally
resolved acquisitions are performed allowing to obtain 20-30 frames per heartbeat.
The main interest is for the left ventricle, due to its vital importance and to its pro-
nounced thickness, ranging between 6 and 16 mm. The shape of its cavity is often approx-
imated by an ellipsoid. Instead, the right ventricle and the atria are characterized by a
thickness that usually does not reach the spatial resolution of the acquisition technologies,
thus their reconstruction is hard and for this reason less studied.
The standard cardiac acquisition plane is orthogonal to the long (apex-base) axis (short
axis plane). Blood appears brighter whereas the myocardium and the surrounding tissue
darker, see Figure 16. Not all the slices on the short axis plane share the same degree of
complexity in their reconstruction. In particular, apical and basal slices images are more
difficult to segment than mid-ventricular ones.
Before the ventricle segmentation starts, a preliminary localization procedure is per-
formed, in order to identify a region of interest and reduce the computational effort. This
step is usually performed automatically, taking advantage of the movement of the heart
on a fixed background, see, e.g., [110]. Another automatic heart location is based on
extracting rectangular subwindows from the image and to compute for them specific fea-
tures. Then, based on a priori chosen class, the subwindows satisfying specific features

66
Figure 16: Left: Longitudinal CT slice of the heart. Right atrium (top left), right ventricle
(bottom left), left atrium (top right), left ventricle (bottom right). Right: CT slice in the
short axis plane. In both figures, on the right the thick left ventricle myocardium is
detectable in darker grey. Radiological images from Ospedale Sacco, Milan, Italy

are recognized as belonging to the heart.

A first class of ventricle segmentation methods makes use of few (or even none) a priori
information. Usually in these methods the endocardium is first segmented by means of
thresholding (see Section 3 and refer, e.g., to [219]). Alternatively, dynamic programming
methods have been considered, where the optimal path in a cost matrix, assigning a low
cost to frontiers, is searched. Taking advantage of the circular shape of the ventricle,
polar coordinates are used and one-dimensional search is performed [237]. In order to
exclude papillary muscles in the segmentation, possible strategies are the computation of
the convex hull of the contour [557] or a smoothing of the latter by means of the fitting of
a parametric curve [582]. As for the vascular case, deformable models are used for cardiac
segmentation (see Section 3). In particular, the functional to be minimized often includes
a regularization term to control the smoothness of the reconstructed curves [451].
Specific automatic methods that have been developed for cardiac image segmentation
are based on a strong a priori information on the shape of the ventricles that is included
in the segmentation algorithm by means of statistical models. These strategies are suited
for cardiac segmentation since the variability among patients of the heart shape is, unlike
the arteries, very small in normal conditions. These statistical model-based segmentation
strategies rely on identifying an average shape of available geometries forming a training
set and on modeling the variability within the latter. This is usually done by means of
the principal component analysis of positions and, in case, displacements, that allows to
compute the eigenvalues and eigenvectors of the covariance matrix related to the training
set. These strategies allow an automatic segmentation that does not need any interaction
of the user, at the expense of having at disposal a training set. For example, deformable
models have been extended to this framework by adding to the functional to be minimized

67
a term that penalizes the distance to a reference model (for example the mean shape of the
training set). Another very common statistical model-based strategy is the atlas-guided
segmentation. Given an atlas, i.e. an integrated image from multiple segmentations, a
registration procedure is performed based on mapping the coordinates of the image under
investigation to those of the atlas [341]. This transformation is then applied to the atlas
obtaining the final segmentation. The registration process could be based on non-rigid
transformations that account for elastic deformations. For a recent review of cardiac
segmentation methods we refer to [449].
Finally, we observe the importance of including the fibers orientation in the recon-
structed geometries in view of modeling the electrical propagation and mechanical con-
traction in the muscle region. Indeed, as discussed in Section 5, the conduction velocity of
the action potential propagation assumes different values along the fibers than in the tan-
gential direction. Moreover, the stretching ability of the myocardium is facilitated along
the fibers direction.
Diffusion-tensor-MRI is a MRI technology able to identify the fibers orientation, how-
ever, not yet daily used in the clinical practice, and hardly applicable because of the
heart movement. Since the fibers direction is unfortunately hardly detectable by the com-
mon radiological acquisition technologies, analytical representations of the fibers suitably
mapped onto the geometry under investigation have been proposed so far to supply the
lack of information provided by the imaging. For example, in [447] fibers are described
as geodesics, whereas in [461] they are represented by means of spiral surfaces. Other
strategies are based on a computational generation of the fibers orientation to provide a
plausible configuration, e.g. by means of the solution of a Poisson equation [39, 496, 596],
or by using the unscented Kalman filter [395].
As for the Purkinje fibers, they are not detectable from classical radiological acquisi-
tions since their thickness falls below the spatial resolution. For this reason, in [1, 513, 276]
it has been proposed to exploit the fractal nature of these fibers to generate a realistic
Purkinje network, whereas in [567, 427, 428] such a fractal network has been personalized
to the patient at hand by including patient-specific measures of the activation times (see
Section 6.2 for a description of the latter).
For the muscle region mesh generation, the strategies described in Section 3 could be
applied as well as to cardiac geometries. For ideal ellipsoid geometries, usually considered
in numerical experiments to test the performance of the numerical algorithms, structured
mesh of hexahedra are often used exploiting the symmetry of the ventricles around the
long axis, see e.g. [436]. However, also unstructured meshes composed by tetrahedra have
been considered so far [218], in particular for real geometries reconstructed by MRI or CT,
see e.g. [496, 595, 566], or for the atria [575]. Hybrid unstructured meshes composed by
tetrahedra and hexahedra have also been successfully considered [238]. We notice that for
the solution of the mechanical problem in the muscle region, no particular requirements
are needed for the mesh generation, whereas for the electrical propagation problem, due to
the very steep front (about 200 µm), the required mesh resolution should be at least of the
order of 100 µm in order to spatially resolve this front [109]. For the mesh generation of
the heart chambers in view of the numerical solution of the fluid-dynamic problem, often
unstructured tetrahedral elements are considered, see e.g. [375]. In this case the mesh

68
resolution needs to be very fine, even smaller than the one of CT or MRI technologies
(≃ 0.5 mm), in order to capture the complex flow structures arising in particular in the
left atrium and left ventricle.

6.2 Cardiac boundary data

The acquisition of boundary data on the electrical activation in the muscle region is nowa-
days possible owing to specific therapeutic techniques. For example, the NavX system,
is able to accurately locate any electrode catheter inserted in the heart chambers, pro-
viding accurate, real-time recording of the local endocardial electrical activity in a point
in contact with the catheter [153, 567]. In particular, this allows one to acquire endo-
cardial maps of the activation times, defined as the time at which the action potential in
a point reaches an intermediate value between the rest and the plateau ones. However,
this acquisition is performed only for specific therapeutic purposes, such as the ablation
of anomalous electrical pathways.
For the mechanical problem involving the muscle region, commonly data that could
be available are the stresses exerted by the blood on the endocardium of the left ventricle
and the endocardial and/or epicardial vessel wall displacements. As for the first type of
data, they are usually extracted from a measure of the aortic pressure (see Section 3.2).
This represents a good approximation; as a matter of fact the ventricular and circulatory
pressures are not the same (thus allowing the acceleration and decelaration of blood) but
they are very similar, see Figure 13. As for the vessel wall displacements, they can be
obtained from dynamic MRI or CT images, which allows one to have 20/30 frames per
heartbeat, providing the position of the endocardium and epicardium at several instants.
After suitable post-processing, these techniques can provide an estimate of the vessel wall
displacement (and thus velocity) by comparing two consecutive frames. The endocardial
vessel wall velocity, thanks to a continuity argument, could also be interpreted as the
blood velocity at the interface with the endocardium [295, 375].
Another useful measure that could be provided quite easily by means of Doppler
echocardiographic methods or PC-MRI (see Section 3.2) is the flow rate at the mitral
and aortic valve orifices. With PC-MRI technology the measure of blood velocity is pos-
sible in principle in any point of the ventricles and atria chambers.
In Section 7 we will see how these data can be used to provide boundary conditions
for the different cardiac models that we are about to introduce.

6.3 Cardiac biological data

The classical monodomain and bidomain models, widely used to describe the electrical
propagation in the muscle region and Purkinje network (see Section 7.1), would require
the following data: the membrane capacitance per unit area C, the membrane surface-
to-volume ratio χ, and the conductivities σ (see next Section). Their subject-specific
acquisition is in general out of reach. An acceptable range of values for the capacitance
Cm in the myocardium is (0.8−1.0) µF/cm2 [497, 269]; the capacitance Cp in the Purkinje
network features a higher value (∼ 1.3 µF/cm2 [327]). The value of membrane surface-

69
to-volume ratio in the muscle region χm varies in the range (200, 3000)cm−1 [403, 460],
whereas for the Purkinje network, a measure in a porcine heart lead to χp = 1467 cm−1
[523]. In the myocardium, we need to distinguish between the conductivities along the
fibers direction (index f ) and that in the direction of sheets (index s). Sometimes, also
a third direction (orthogonal to sheets) is considered (if this is the case, we will use
the index n). Moreover, in view of the bidomain model, we also distinguish between
the intra-cellular (index i) and extra-cellular (index e) conductivities. Following [498],
acceptable ranges used in the numerical experiments are: σfi ∈ (0.17, 0.34) (Ω m)−1 , σsi ∈
(0.02, 0.06) (Ω m)−1 , σfe ∈ (0.12, 0.62) (Ω m)−1 , σse ∈ (0.08, 0.24) (Ω m)−1 .
Another classical model used so far to describe the activation of Purkinje and muscle
cells is provided by the Eikonal equation (see Section 7.1). In this case, one needs to
prescribe explicitly the conduction velocity V , unlike in the monodomain and bidomain
models where this is determined by the conductivity and the membrane capacitance.
Acceptable ranges of values of the conduction velocity are Vf ∈ (0.6, 1.0) m/s [293], Vs ≃
Vf /2 [197], Vn ≃ Vf /4 [108], and Vp ∈ (3.0 − 4.0) m/s [273], where again index f refers to
the direction along the fibers in the muscle region, s to the direction along sheets, and p
to the Purkinje network.
All these electrical data are hardly measurable in vivo so that no patient specific
measures are usually available. Nevertheless, the use of extra data such as the activation
time at the endocardium provided by the NavX system could be used to estimate some of
these parameters by solving a suitable inverse problem, see e.g. [515, 567] for the case of
the Purkinje network.
Regarding the parameters involved in the cardiac mechanic model, they depend on the
chosen constitutive law. In general, the linearization of the stress-strain curves gives the
following values for the corresponding varying-in-time Young modulus E [495]: during a
traction/compression test along the fibers direction, we have E ≃ 20 kP a during traction
at small deformation and during compression, and E ≃ 1500 kP a during traction for
deformation of about 15%; during traction the material behaves as transversally isotropic
so that the stiffness in the two tangential directions (along sheets and orthogonal to fibers
and sheets) is the same, whereas during compression the stiffness in the sheets direction
is higher than that in the direction orthogonal to fibers and sheets. As for the Poisson
modulus, usually a value of 0.45 is considered.
Regarding the aortic valve leaflets, measures in dogs featured a low stiffness during
systole (≃ 240 kP a) and an increased stiffness during diastole (≃ 5000 kP a). For the
mitral valve, a different stiffness behavior is observed for the two leaflets (the anterior
and the posterior ones). In particular, in [311] the following values of the Young modulus
are proposed: for the anterior leaflet E = 6200 kP a and E = 2100 kP a in the directions
parallel and perpendicular to the annulus, respectively, whereas for the posterior leaflet
E = 2300 kP a and E = 1900 kP a, respectively.
Blood in the heart features the same behavior than in the circulatory systems so that
its parameters (density and viscosity) are taken in the ranges reported in Section 3.3.

70
7 Modeling the heart
In this section we discuss the main mathematical models introduced so far to describe
the heart function and the related numerical strategies developed for their solution. In
particular, Section 7.1 is devoted to the electrical propagation, Section 7.2 to the cardiac
mechanics and electro-mechanical coupling, Section 7.3 to the ventricular fluid-dynamics,
Section 7.4 to the valve modeling and its interaction with blood fluid-dynamics, and finally
Section 7.5 to their integration.

7.1 Cardiac electric activity

7.1.1 The bidomain model
As observed in Section 5.3, the electric activation of the heart is the result of two processes:
at the microscopic scales, the generation of ionic currents through the cellular membrane
producing a local action potential, and at the macroscopic scales, the travelling of the
action potential from cell to cell allowed by the presence of the gap junctions. The former
is a discrete process, in the sense that there is a delay between the depolarization of
a cardiomyocyte and its neighbours, whereas the latter can be assimilated to a smooth
process [150].
At the macroscopic level, the propagation of the potentials is described by means of
partial differential equations, suitably coupled with ordinary differential equations model-
ing the ionic currents in the cells. In particular, the single membrane cell can be modeled
as a capacitor separating charges that accumulate at its intracellular and extracellular
surfaces. Moreover, as observed in Section 5, ionic currents cross the membrane through
suitable channels which open and close during the excitation. A suitable model can there-
fore be expressed by the simple electric circuit depicted in Figure 17, for which
µ ¶
∂Vm
Im = χm Cm + Iion ,
∂t
where Im is the membrane current per unit volume, Cm the membrane capacitance, χm the
surface area-to-volume ratio (see Section 6.3 for a quantification of the latter two), Vm (t, x)
the transmembrane potential, and Iion (t, x) the ionic currents. Due to the conservation of
current and charge, this current should equal the divergence of both the intracellular and
extracellular current fluxes j i and j e

∇ · j i = −Im , ∇ · j e = Im . (71)

The Ohm’s law in the intracellular and extracellular regions give

j i = −Σi ∇φi , j e = −Σe ∇φe , (72)

Σi , Σe being the conductivity tensors and φi (t, x), φe (t, x) the intracellular and extracel-
lular potentials, so that
Vm = φi − φe . (73)

71
Figure 17: Electric circuit for the sequence of two cardiac cells. Each of them is composed
by a capacitor and a series of resistances, one for each ionic current (here only sodium and
potassium ionic channels are depicted). In the intracellular region, two adjacent cells are
connected by a resistance representing a gap junction. However, the latter is not explicitly
modeled at the macroscopic scales, rather its effect is hidden in the conductivity tensor,
see the text

Notice that, due to the anisotropy of the cardiac tissue induced by the presence of
fibers and sheets, each conductivity tensor is in general expressed in terms of three scalar
quantities representing the conductivities along the fiber direction af (x), the direction
as (x) orthogonal to af and tangential to sheets, and the direction an (x) orthogonal to
sheets, i.e.
Σβ = σfβ af aTf + σsβ as aTs + σnβ an aTn , β = i, e. (74)
Putting together all the previous equations and using a homogenization procedure (see
e.g. [118] for a rigorous derivation), we obtain for each t > 0 the following system of two
partial differential equations called parabolic-parabolic (PP) formulation of the bidomain
equations:
∂Vm
χm Cm − ∇ · (Σi ∇φi ) + χm Iion = Iiext in Ωmus , (75a)
∂t
∂Vm
− χm Cm − ∇ · (Σe ∇φe ) − χm Iion = −Ieext in Ωmus , (75b)
∂t
where Iiext (t, x), Ieext (t, x) are applied currents per unit volume.
Thanks to (71), ∇ · (j i + j e ) = 0; thus, using (72) and (73), we obtain the following
parabolic-elliptic (PE) formulation of the bidomain equations:
∂Vm
χm Cm − ∇ · (Σi (∇φe + ∇Vm )) + χm Iion = Iiext in Ωmus , (76a)
∂t
− ∇ · (Σi ∇Vm ) − ∇ · ((Σi + Σe ) ∇φe ) = Iiext − Ieext in Ωmus . (76b)

Due to the homogenization procedure, the effect of the gap junctions, which at the
cellular level contributes in determining the current flux j i , is hidden in the conductivity

72
tensor Σi . We also notice that both bidomain problems (75) and (76) hold in the whole
computational domain Ωmus given by the union of the myocardium with endocardium and
epicardium, see Figure 15. Indeed, again because of the homogenization procedure, no

Figure 18: Left ventricular myocardial domain obtained by the cut at the basis (left), and
corresponding fluid cavity domain (right)

geometrical distinction is made between the intracellular and extracellular regions, even if
their different functionality is maintained in the bidomain models.

7.1.2 Cardiac cell models

In order to close equations (75) and (76) we need to provide a model for the ionic current
Iion . In what follows we briefly describe three families of models, featuring different levels
of complexity and accuracy.
The first family, the so-called reduced ionic models only provide a description of the
action potential and disregard sub-cellular processes:

Iion = f (Vm , w), (77a)

∂w
= g w (Vm , w), (77b)
∂t
where f, g w are suitable functions, while w : [0, T ] × Ωmus → RM collects the so-called
gating variables which represent the percentage of open channels per unit area of the
membrane. The most celebrated reduced model for ventricular cells is FitzHugh-Nagumo’s
[182] where f (Vm , w) = −kVm (Vm − a)(Vm − 1) − w and gw (Vm , w) = ǫ(Vm − γw), for
suitable constant parameters k, a, γ. In this case the gating variable w plays in fact
the role of a recovery function that allows to model the refractariness of cells. Other,
more sophisticated ventricular cell models of this family are, e.g., the Roger-McCulloch
[494], the Aliev-Panfilov [5], the Fenton-Karma [172], and the Bueno-Orovio [78] models.

73
Like FitzHugh-Nagumo’s, the first two models are characterized by the dynamics of one
gating variable and by a cubic non-linear expression of the ionic current. Instead, the
Fenton-Karma model and its Bueno-Orovio variant specifically addressed for the human
ventricular cells, feature two and three gating variables, respectively, and a more complex
non-linearity in the ionic current expression. These simple models are very appealing,
especially because their parameters have a direct physical interpretation, such as the action
potential duration, allowing for an easy setting of the model properties. For example, the
Aliev-Panfilov model has been used successfully in the first simulations of ventricular
fibrillation in a real geometry [429]. However, they are not able to describe any process
occurring at the level of the ionic channels and of the cell, so that they are recommended
when one is only interested in the heart electric activity.
The second family of ventricular cell models we consider is that of the so-called first-
generation models. Unlike reduced models that surrogate the ionic current by means of
the sole function f , they allow for an explicit description of the kinetics of different ionic
currents by using several gating variables. They read
 
XN M
Y pj ∂w
Iion = Ik (Vm , w), Ik = Gk  wj k  (Vm − Vk ) , = g w (Vm , w),
∂t
k=1 j=1

where N is the total number of ionic currents and M the total number of gating vari-
ables, Vk is the Nernst potential of the k − th ion (a constant value corresponding to the
thermodynamical equilibrium of the ion at hand), Ik is the current related to the k − th
ion, pjk accounts for the influence of the j − th gating variable on the k − th ionic current
(possibly vanishing), Gk is the maximal conductance of the k − th ion. The M compo-
nents of g usually have the expression gw,j = (wj∞ (Vm ) − wj )/τj (Vm ), where wj∞ is the
equilibrium state and τj the characteristic time constant. The most famous model of this
family, Hodgkin-Hukley’s (HH) [254], depends on three ionic currents, namely the sodium,
potassium and leakage ones, and three gating variables

IN a = GN a w13 w2 (Vm − VN a ), IK = GK w34 (Vm − VK ), IL = GL (VM − VL ).

Although introduced to describe the action potentials in nerves, the HH model inspired
all the following models introduced specifically for the ventricle. Among these, we cite
the Beeler-Reuter [43], the Luo-Rudy I [343], and the TenTusscher-Panfilov [554] mod-
els. These models were widely used to study specific features of the ventricular electric
activation, such as re-entry and fibrillation [598].
Finally, we mention the family of the second-generation ventricular cell models, like the
Luo-Rudy dynamic model [344, 345], where, unlike in first-generation models, a detailed
description of some ion concentration variables c is provided. They provide a detailed
description of many processes allowing for the study, e.g., of channelopathies and of drug
action. However, due to their increased complexity, the required computational time is
very huge for a complete heart model and often the tuning of the parameters is very
demanding. We refer to [109] for a discussion of second-generation models and to [501]
for a general review of cardiac cell models.

74
 Although most of research studies focused on ventricular cell models as those mentioned
above, specific models have been introduced also for the atrial cells, see e.g. [249], and for
the sinoatrial node cells, see e.g. [605].
All the cardiac cell models belonging to the three families described above, used in
combination with the bidomain problem (75) or (76), leads to a system of two PDEs
coupled with two systems of ODEs, i.e. the equations for the gating variables and ion
concentrations written for each point x (the latter, in case, vanishing). The general
expression of such coupled problem reads (we only detail the PP case): Find at each time
t > 0 the potentials VM , φi and φe , the gating variable w and the ion concentrations c
such that
∂Vm
χm Cm − ∇ · (Σi ∇φi ) + χm Iion (Vm , w, c) = Iiext in Ωmus , (79a)
∂t
∂Vm
− χm Cm − ∇ · (Σe ∇φe ) − χm Iion (Vm , w, c) = −Ieext in Ωmus , (79b)
∂t
XN
Iion = Ik (Vm , w, c) in Ωmus , (79c)
k=1
 
M
Y pj
Ik = Gk  wj k  (Vm − Vk (c)) in Ωmus , (79d)
j=1
∂w
= g w (Vm , w) in Ωmus , (79e)
∂t
∂c
= g c (Vm , w, c) in Ωmus , (79f)
∂t
where, together with the notation introduced above for the reduced and first-generation
models, c : [0, T ] × Ωmus → RS collects the S ionic concentration variables and g c is a
suitable function, see e.g. [118]. We observe, in general, the dependence of the Nernst
potential Vk on the variable c. Well-posedness results of the previous coupled problem
are provided, e.g., in [121], where the existence and uniqueness of the solution of the PP
formulation coupled with the FitzHugh-Nagumo model is proved, and in [76], where a
Faedo-Galerkin technique is applied to the PE formulation coupled with a general first-
generation cell model.
The ODE systems modeling the gating variables and the ionic concentrations vari-
ables are in general stiff since the Jacobians ∂g w /∂w and ∂g c /∂c feature a wide range of
eigenvalues.
For each t > 0, the weak formulation of the bidomain model (79) together with ho-
mogeneous Neumann conditions and initial conditions (see Section 7.1.4) reads: Given
Iiext (t), Ieext (t) ∈ L2 (Ωmus ), find Vm (t), φe (t), φi (t) ∈ H 1 (Ωmus ), w ∈ [L2 (Ωmus )]M and

75
c ∈ [L2 (Ωmus )]S such that
Z Z Z
∂Vm
χm Cm z dω + Σi ∇φi · ∇z dω + χm Iion (Vm , w, c)z dω =
Z Ωmus ∂t Ωmus Ωmus (80a)
Iiext z dω,
Ωmus
Z Z Z
∂Vm
−χm Cm z dω + Σe ∇φe · ∇z dω − χm Iion (Vm , w, c)zdω =
Z Ωmus ∂t Ωmus Ωmus
− Ieext zdω,
Ωmus
(80b)
Z Z
∂w
· y dω = g(Vm , w) · y dω, (80c)
Ωmus ∂t Ωmus
Z Z
∂c
· ζ dω = g c (Vm , w, c) · ζ dω, (80d)
Ωmus ∂t Ωmus

for all z ∈ H 1 (Ωmus ), y ∈ [L2 (Ωmus )]M and ζ ∈ [L2 (Ωmus )]S , together with (79c)-(79d).

7.1.3 Reduced continuous models: the monodomain and Eikonal equations

To reduce the complexity of the bidomain models (75) and (76), an assumption of pro-
portionality between the intracellular and extracellular conductivities, i.e. Σe = λΣi for
a suitable constant λ, is introduced. Substituting this relation in (76b), eliminating Σe
and substituting the corresponding relation for Σi in (76a), we obtain the monodomain
equation: Find for each t > 0 the transmembrane potential Vm such that
∂Vm
χm Cm − ∇ · (Σ∇Vm ) + χm Iion = I ext in Ωmus , (81)
∂t
where
λ
Σ= Σi (82)
1+λ
λI ext +I ext
is the effective conductivity and I ext = i 1+λ e . Again, a ventricular cell model is needed
to provide the ionic current Iion . The same models discussed above for the coupling with
the bidomian problem are used in combination with the monodomain problem as well.
Once the transmembrane potential Vm has been computed, the extracellular potential φe
could be computed as a post-processing by solving the elliptic problem (76b).
For each t > 0, the weak formulation of the monodomain problem (81) together with
homogeneous Neumann conditions and initial conditions (see Section 7.1.4) reads: Given
I ext (t) ∈ L2 (Ωmus ), find Vm (t) ∈ H 1 (Ωmus ), w ∈ [L2 (Ωmus )]M and c ∈ [L2 (Ωmus )]S such

76
that
Z Z Z
∂Vm
χm Cm z dω + Σ∇Vm · ∇z dω + χm Iion (Vm , w, c)z dω =
ΩZ ∂t Ωmus Ωmus
mus (83a)
I ext z dω,
Ωmus
Z Z
∂w
· y dω = g(Vm , w) · y dω, (83b)
∂t
ZΩmus Z Ωmus
∂c
· ζ dω = g c (Vm , w, c) · ζ dω, (83c)
Ωmus ∂t Ωmus

for all z ∈ H 1 (Ωmus ), y ∈ [L2 (Ωmus )]M and ζ ∈ [L2 (Ωmus )]S , together with (79c)-(79d).
Although the hypothesis underlying the monodomain model, i.e. the proportionality
between the internal and external conductivities, is not physiological as shown by some
experiments, in some cases this model provides a very accurate solution in comparison with
the bidomain one. In particular, this is true when there is no injection of current in the
extracellular region [120, 460]. On the contrary, when an external current is injected such
as in defibrillation, the monodomain solution is not anymore accurate and the bidomain
model is mandatory since the unequal anisotropy is fundamental to successfully describe
these scenarios [548].
A further simplification is provided by the eikonal equation. Starting from the bido-
main model coupled with a simplified representation of the ionic current which does not
consider any gating variable and allows for the description only of the depolarization phase,
in [115] the following eikonal-diffusion equation is derived:
p
co ∇ψ · M ∇ψ − ∇ · (M ∇ψ) = 1, (84)

where ψ(x) is the unknown activation time (see Section 6.2), co represents the velocity
of the depolarization wave along the fiber direction for a planar wavefront and M =
Σ/(χCm ). A different derivation has been provided in [286], leading to the following
eikonal-curvature equation
µ ¶
p p M ∇ψ
co ∇ψ · M ∇ψ − ∇ψ · M ∇ψ∇ · √ = 1. (85)
∇ψ · M ∇ψ

They are both steady equations providing an information on the activation of each cell.
The contours of ψ(x) give the position of the wavefront at time t = ψ. The eikonal-
diffusive model (84) is an elliptic equation, where the propagation speed is influenced by
the tissue surrounding the wavefront. Once the activation time ψ has been computed, it
is possible to obtain an approximate value of the extracellular potential φe by solving at
each time step a suitable elliptic problem, see [114].
Instead, the eikonal-curvature model (85) is of parabolic type since the “diffusive” term
lacks the second drivative in the direction of propagation. This term is also proportional
to an anisotropic generalization of the mean curvature [546]. This implies that the prop-
agation is faster when the wavefront is concave. This is in accordance with the diffusion

77
of charge which allows for a faster depolarization in regions close to already depolarized
tissues.
The eikonal equations are unsuitable for recovering the action potential and the ionic
currents. However, they provide accurate results about the activation of cells even in
complex scenarios such as front-front collision, see e.g. [114] for the eikonal-diffusive
model.
The eikonal models are however very appealing from the computational point of view.
First of all, they consist in a single steady PDE. Although being non-linear, it does not
require to manage the coupling with the ODE systems. More importantly, the activation
time, unlike the transmembrane potential, does not feature any internal or boundary layer,
so that no special restriction on the mesh is needed in this case (see Section 6.1).

7.1.4 Boundary conditions and Purkinje network models

We discuss here the initial and boundary conditions of the problems introduced above.
The bidomain and monodomain equations and the ODE systems for the gating variables
and ionic concentrations need to be equipped with suitable initial conditions, i.e.

Vm |t=0 = Vm,0 , w|t=0 = w0 , c|t=0 = c0 in Ωmus ,

for given functions Vm,0 (x), w0 (x), c0 (x).

Regarding boundary conditions for the bidomain, monodomain and eikonal-diffusion
problems, a homogeneous Neumann condition is commonly prescribed at the external
surface Σepi of the epicardium and, in case of a ventricular domain solely, at the basis Σb
(see Figures 15 and 18, left) to prescribe null outgoing current fluxes. In particular, the
following conditions have to be prescribed on Σepi ∪ Σb :

(Σβ ∇φβ ) · n = 0 β = i, e for the PP formulation (75),

¾
(Σi ∇(Vm + φe )) · n = 0
for the PE formulation (76),
((Σi + Σe )∇φe ) · n + (Σi ∇Vm ) · n = 0
(Σ∇Vm ) · n = 0 for the monodomain problem (81),
(M ∇ψ) · n = 0 for the eikonal-diffusion problem (84).

Moreover, for the bidomain problems (75) and (76), they force the following compatibility
conditions on the applied external currents:
Z Z
ext
Ii = Ieext .
Ωmus Ωmus

On the internal surface Σendo of the endocardium, again Neumann conditions are pre-
scribed. In this case, however, they could be non-homogeneous at specific stimulation
points (e.g. the atrio-ventricular node and the points of the His bundle). For the eikonal
problem, Dirichlet data on the activation time could be prescribed at some specific loca-
tions in case they are available thanks to e.g. the NavX system (see Section 6.2) [515].
When redundant (e.g. in presence of the Purkinje network, see below) these data have

78
been used to solve inverse problems, for example to estimate the conduction velocity in
the myocardium [515] or to obtain personalized Purkinje networks [567, 427, 428].
If the mathematical model accounts for the presence of the Purkinje network, interface
conditions on Σendo describing the continuity of the current and of the potential at the PMJ
are implicitly provided for the bidomain and monodomain problems by the solution of the
coupled muscle region/Purkinje network problem [572, 566]. For the sake of exposition,
we do not detail here the bidomain and monodomain models for the Purkinje network,
referring the interested readers to [572, 72, 566]. We only notice that, unlike the muscular
case, in the network the gap junctions connecting two consecutive Purkinje cells are often
explicitly modeled by means of resistances. Specific Purkinje cell models, with the same
structure of those developed for the muscular cells, have also been developed, see e.g.
[140]. However, we describe here the mechanisms of coupling, in particular, we refer to
the coupled problem obtained by considering the monodomain problem both in the muscle
region and in the Purkinje network [566]. We consider N PMJ located at x = sj and we
assume that each of them could be modeled by means of a resistance RP M J . Then, the
monodomain/monodomain coupled problem reads: Find for each t > 0, Vp , Vm , wp , w
and γj , j = 1, . . . , N , such that:
 
N
X 1
Pm Vm , w, I γj + I ext  = 0, (86a)
Ar Br (sj )
j=1

Pp (Vp , wp , γ) = 0, (86b)
R
Vp (sj ) − A1r Br (sj ) Vm dx
γj = j = 1, . . . , N, (86c)
RP M J

where Pm (Vm , w, F ) = 0 represents the monodomain problem in the myocardium with

source term F , Pp (Vp , wp , η) = 0 the monodomain problem in the Purkinje network
with Neumann conditions with data ηj at the PMJ, Vp and wp are the transmembrane
potential and the gating variables in the Purkinje network, γj are the PMJ currents which
are determined by the Ohm’s laws (86c), IY is the characteristic function related to the
region Y ⊂ Ωmus , Br (sj ) is the ball of radius r centered at the point sj and Ar the volume
of this ball. We observe that the two monodomain problems are coupled by means of
the PMJ currents γj : for the 3D problem the latter act as source terms distributed in
balls of radius r, whereas for the network they act as Neumann conditions [72]. A similar
approach could be considered for the bidomain problems as well.
The coupling between eikonal muscular and Purkinje network problems has been ad-
dressed in [567] for normal propagation and in [427] for pathological propagations.

7.1.5 Computing the surface electrocardiogram signals

The computation of the potentials related to the electric propagation in the heart could be
used to numerically compute the surface electrocardiogram (ECG) signals, i.e. the electric
potential on the surface of the body, thus simulating what happens in the common clinical
practice. This could be obtained by coupling the bidomain or monodomain problem with

79
the propagation in the torso, modeled by a simple diffusion problem for the extracellular
potential at each t > 0:
∇ · (ΣT ∇φT ) = 0 in ΩT ,
where ΩT is the torso domain that surrounds the heart domain Ωmus (i.e. all the heart
boundary surface Σepi ∪ Σb represents the interface with the torso), ΩT ∩ Ωmus = ∅,
with φT representing the extracellular potential in the torso. Homogeneous Neumann
conditions have to be applied on the external torso surface ΣT , whereas the following
interface conditions, that replace the boundary conditions for the heart problem, need to
be prescribed at the torso/heart interface:

φe = φT on Σepi ∪ Σb ,
(Σe ∇φe ) · n = (ΣT ∇φT ) · n on Σepi ∪ Σb ,
(Σi ∇(Vm + φe )) · n = 0 on Σepi ∪ Σb .

7.1.6 Numerical discretization

The numerical solution of the bidomain and monodomain problems is very demanding.
Together with the strict constraint on the spatial mesh size due to the propagation of a
very steep front (see Section 6.1), a time step of the order of dozens or even hundredths
of milliseconds has to be used in order to capture the fast dynamics characterizing the
propagation, with time constants of the order of 0.1 ms. Moreover, as discussed below,
the algebraic solution of the linear systems arising at each time step using e.g. Finite
Elements requires a careful treatment due to the coupled nature of the problem, to the
singularity of some of the matrices involved, and to the presence of the non-linear term
due to the ionic currents.
Several time discretizations strategies have been considered so far. We can gather them
into four main categories, i.e. explicit, semi-implicit, implicit and operator splitting-based
methods, which are now briefly presented. We start by introducing the Finite Element
matrices and vectors:
Z
(M )jk = ψk ψj dω,
ZΩmus
(Aβ )jk = Σβ ∇ψk · ∇ψj dω, β = i, e
Ωmus Z
n r s
(I ion (V m , W , C ))j = Iion (Vmn , wr , cs )ψj dω,
Z Ωmus
n r
(G(V m , W ))j = g(Vmn , wr ) · y j dω,
Z
Ωmus
n r s
(S(V m , W , C ))j = g c (Vmn , wr , cs ) · ζ j dω,
Ωmus

with ψi , y j and ζ j denoting the basis functions of the Finite Element spaces, and n, r, s
integers representing the time discrete instants. Vectors V m , Φe , Φi , W and C denote
the unknown coefficients of the Finite Elements solutions associated to the unknowns
Vm , φe , φi , w, c. Notice that the dimension of W and C is M K and SK, respectively,

80
where K is the number of degrees of freedom associated with the mesh and the choice of
Finite Elements (e.g. the number of vertices for linear FE). In all the cases, in order to
simplify the notation, we will set Ieext = Iiext = 0 and we assume that the running temporal
index n+1 is understood.

Explicit methods. In explicit methods, all the problems in (80) are discretized by
means of an explicit scheme, e.g. forward Euler both for the PDEs and for the ODE
systems, [466, 161]. This choice allows to decouple the 4 blocks of the bidomain system
(the two PDEs and the two ODEs system), involving in the PP formulation only the mass
matrix, thus in principle avoiding the need to solve any linear system provided that the
mass-lumping is performed. Instead, for the PE formulation, the absence of time derivative
in the second PDE implies that a linear system in the unknown Φe needs in any case to
be solved (we detail the forward Euler method):

V m − V nm
χm Cm M = −Ai (V nm + Φne ) − χm I ion (V nm , W n , C n ),
∆t
Ai V m + (Ai + Ae )Φe = 0,
W − Wn C − Cn
M = G(V nm , W n ), M = S(V nm , W n , C n ).
∆t ∆t
Explicit methods lead to a severe constraint on the time discretization of the type ∆t <
h2 /Σ, Σ being a representative value of the conductivity [466]. In [507], it has been shown
for a model problem that an explicit method is not absolutely stable for a value of ∆t
much smaller than the one required to capture the front propagation. For this reason and
due to the increased CPU availability, nowadays explicit methods are not so much used.

Semi-implicit methods. ODE systems are usually solved at each time step by means
of explicit or semi-implicit methods (in the latter case the dependence on Vm is treated
explicitly). This suggests to use semi-implicit methods [287, 440, 117] for the whole coupled
PDEs/ODEs problem. These methods are the most used together with operator splitting-
based ones (see below). They are based on treating implicitly the diffusive term and
explicitly the non-linear term. A possible semi-implicit (first order) scheme for the PP
formulation (80) reads as follows (see e.g. [117]):

W − Wn C − Cn
M − G(V nm , W ) = 0, M − S(V nm , W , C) = 0,
∆t ∆t
V m − V nm
χm Cm M + Ai Φi + χm I ion (V nm , W , C) = 0,
∆t
V m − V nm
− χm Cm M + Ae Φe − χm I ion (V nm , W , C) = 0.
∆t
A semi-implicit method like the one reported here features two nice properties. First
of all, the two PDEs are decoupled by the ODE systems, highly simplifying the numerical
solution of the entire problem. This decoupling strategy is justified by noticing that the
Jacobian entries of a fully implicit discretization feature dominant values on the diagonal

81
suggesting a weak coupling between potentials and gating/ionic concentrations variables
[388]. Secondly, it allows for a linearization of the non-linear reaction term given by the
ionic currents, thus requiring the solution of a (2 × 2 block) linear system.
A special attention has been paid to the evaluation of the ionic current in a Finite
Element context. In particular, two strategies have been mainly considered, i.e. ionic
current interpolation (ICI) and state variable interpolation (SVI). In the first case, only
the nodal values are used to build an interpolation of the ionic current, whereas in the
second case the transmembrane potential and the gating/ionic concentration variables
are interpolated within each element and the ionic current is then evaluated by using
these interpolated variables [434]. The ICI approach is of course more efficient from the
computational point of view and allows one to express the reaction term by means of a
mass matrix. However, it features a lower accuracy with respect to SVI, in particular, the
computed conduction velocity is generally larger than the true one, see [434].
From the algebraic point of view, the solution of a semi-implicit discretized problem
is very demanding, since the matrix of the linear system associated to the PDEs written
in terms of the unknowns Φi and Φe is given by
· ¸ · ¸
χm Cm M −M Ai 0
B= + .
∆t −M M 0 Ae

Both terms of this matrix are singular, the first one due to the degenerate parabolic
nature of the PP bidomain formulation, the second one being each block Aβ related to
a pure Neumann diffusive problem. However, matrix B is positive semidefinite, hence
the preconditioned conjugate gradient method is often used for its numerical solution
[440, 117]. Preconditioning is mandatory since matrix B is highly ill-conditioned due to
the block 2 × 2 mass matrix that, unlike usually happens for the classical mass matrix,
worsen the spectrum of the stiffness matrix. Efficient preconditioners are, e.g., block Jacobi
and Gauss-Seidel [361], block SSOR [440], multilevel additive Schwarz [435], multigrid
[456, 573], and a functional block preconditioner obtained by the monodomain solver
[206].
Semi-implicit methods have been proposed also for the PE formulation, see, e.g., [177].
For example, with respect to the unknowns V m and Φe , we have
· ¸ · ¸
χm Cm M 0 Ai Ai
B= + .
∆t 0 0 Ai Ai + Ae

Again the matrix is singular, ill-conditioned, and semidefinite positive. In this context,
we cite [454] for an incomplete block LU factorization precontitioner and [571] where the
two PDEs are decoupled by treating explicitly Vm in (76a) and φe in (76b).
In any case, semi-implicit methods are conditionally stable with a bound on ∆t which
is however independent of the mesh size [177, 118].
A variant of the semi-implicit method reported above arises from treating implicitly
the reaction term Iion . In this case, Newton-Krylov-Schwarz methods are very efficient
[388]. Second order semi-implicit schemes have been successfully considered as well, see
e.g. [161].

82
 Operator splitting-based methods. These methods separate the reaction operator
from the diffusive operator, in a way similar to what is done in fractional step methods
for fluid problems. They were first introduced for the monodomain problem in [467]. We
present here an operator splitting-based method for the PP formulation of the bidomain
problem [534, 118]:

Step 1. Given the quantities at time step tn , solve the reaction problem and the ODE systems
in (tn , tn + θ∆t]:
n+θ
Ve m − V nm
χm Cm M + χm I ion (V ∗m , W ∗ , C ∗ ) = 0,
∆t
f n+θ − W n
W e n+θ − C n
C
M − G(V ∗m , W ∗ ) = 0, M − S(V ∗m , W ∗ , C ∗ ) = 0;
∆t ∆t

n+θ
Step 2. Given Ve m , solve the diffusion problems in (tn , tn+1 ]:
n+1 n+θ
Vb m − Ve m
χm Cm M + Ai Φ∗i = 0,
∆t
n+1 n+θ
Vb − Ve m
− χm Cm M m + Ae Φ∗e = 0;
∆t

n+1
Step 3. Given Vb m , Wf n+θ , C
e n+θ , solve the reaction problem and the ODE systems in
(tn + θ∆t, tn+1 ]:

V n+1 b n+1
m −Vm
χm Cm M + χm I ion (V ∗m , W ∗ , C ∗ ) = 0,
∆t
W n+1 − Wf n+θ C n+1 − Ce n+θ
∗ ∗
M − G(V m , W ) = 0, M − S(V ∗m , W ∗ , C ∗ ) = 0.
∆t ∆t

The superscript ∗ means that the quantity at hand could be treated either explicitly or
implicitly. The variable θ could assume value 1/2 or 1. In the latter case, step 3 is unnec-
essary, we set V n+1
n+1
= Vb m , W n+1 = W f n+1 , C n+1 = C
e n+1 , and a first order method
m
is obtained (Godunov splitting). For θ = 1/2, we have a second order method provided
that all the subproblems are solved with a second order strategy (Strang splitting). Notice
that if step 2 is solved implicitly, then the same preconditioners introduced below for the
semi-implicit schemes could be applied as well since it applies to the same operator.

Implicit methods. Some authors considered a fully implicit discretization of the full
bidomain problem (75) or (76), see e.g. [77, 389]. In this case the whole Jacobian is built
and the Newton method is applied. Due to the small time step required in bidomain
simulations to capture the propagating front and due to the excellent stability properties
of semi-implicit and operator splitting-based methods, fully implicit method are nowadays

83
not so much considered.

In Figures 19 and 20 we report some examples of numerical results related to the so-
lution of the electrical propagation in the myocardium.

Figure 19: Left: Purkinje network generated by the algorithm proposed in [428] in the
case of a real left ventricle. Right: Map of the activation times computed by means of
the Eikonal equation. The time marching scheme and P1 finite elements have been used.
The source term (dark blue) is located within the myocardium as typically happens in
the Wolff-Parkinson-White syndrome. Results obtained by means of a code implemented
in the VMTK envoirment ([Link]). CT images from the Cardiology Division at
Ospedale S. Maria del Carmine, Rovereto (TN), Italy, and from the Radiology Division
of Borgo-Trento (TN), Italy

7.2 Cardiac mechanics and electro-mechanics coupling

7.2.1 The continuous mechanics problems
During a physiological contraction, the cardiac cells change their length up to 20%-30%,
so that finite elasticity models are mandatory to describe heart contraction and relaxation.
In particular, the first Piola-Kirchhoff tensor is written as the sum of two terms [396, 405],
a passive component, T b P , describing the stress required to obtain a given deformation of
s
the passive myocardium (similarly to arteries), and an active component, T b A , representing
s
the tension generated by the depolarization of the propagating electric signal that provides
the internal active forces responsible of the contraction (see Section 5.4):

bP + T
bs = T
T b A, (93)
s s

84
Figure 20: Propagation of the transmembrane potential in the two ventricles at eight
different instants during an heartbeat. Monodomain simulation, semi-implicit method,
P1 Finite Elements. These results have been obtained using LifeV; the computational
mesh has been obtained by an open source biventricular geometry segmented from CT
images, see [499]

where according to the notation of Part 1,b refers to quantities computed in the reference
domain.
The passive component of the stress is obtained as the derivative of a suitable strain
energy function, see (11). The heart mechanics response (just like the electric propagation,
see Section 7.1) highly depends on the presence of fibers and sheets. For this reason, the
passive myocardium is modeled as an orthotropic material, characterized by two princi-
pal directions and with different material responses on three mutually orthogonal planes,
identified by these directions. This is in accordance with the shear tests performed on pig
hearts, that highlighted an elevated resistance to shear deformations producing an exten-
sion along the fiber direction, an intermediate resistance in the direction orthogonal to
fibers and tangential to sheets, and the least resistance in the third orthogonal direction
[146]. By denoting with a bf , a
bs, a
b n the unit vectors along these directions (see Section
7.1) in the reference configuration, the following strain energy function has been proposed
in [258]:
a X ai h 2
i af s h bf s I8,f
2
i
Θ(I1 , I4,f , I4,s , I8,f s ) = eb(I1 −3) + ebi (I4,i −1) − 1 + e s − 1 , (94)
2b 2bi 2bf s
i=f,s

where, referring to the notation introduced in Section 4.2, I1 = tr C, I4,i = a b i · (C a

bi) , i =
b f · (C a
f, s, I8,f s = a b s ) are invariants of C, and a, b, af , bf , as , bs , af s , bf s are material
parameters. I1 is the first isotropic invariant and the related term in Θ accounts for the
isotropic response of the myocardium at small deformations; I4,f and I4,s are the squares
of the stretch of a b f and a b s in the deformed configuration and the related terms in the

85
energy are associated with the increased stiffness of the material along the two principal
directions af and as for large deformations; finally, I8,f s represents the angle spanned
by the two principal directions in the deformed configuration and the related term in
the energy describes the coupling between the two principal directions. Convexity of
this energy is guaranteed for positive parameter values [258]. Other orthotropic models
have been proposed, e.g., in [272, 124], whereas transversally isotropic models with only
one principal direction (that along the fibers) had been previously introduced, e.g., in
[271, 225]. Some authors model the myocardium as incompressible [258], in accordance
with the experiments reported in [579]. In this case the term ps (J − 1) is added to the
strain energy functions, as done in (16).
On the basis of experimental evidences, reported e.g. in [146], highlighting hysteresis
under shear deformations, viscoelastic orthotropic models have recently been proposed
for the passive myocardium, see [234]. The viscoelastic behavior is probably due to the
extracellular fluid that filtrates through the solid part.
Regarding the active contribution of the cardiac cells to the contraction, this is reg-
ulated by the opening of calcium channels as a response to the depolarization, with a
consequent entry of calcium ions in the cells. As detailed in Section 5.4, this process is
responsible for the cardiac contraction. Since the latter occurs along the axial direction
of the cardiac cells, i.e. along the fiber direction, the active part of the stress tensor has
usually the following form [396]

b A = P A af ⊗ a
T bf , (95)
s

where the scalar function of time and space P A represents the pointwise active stress and
should be properly modeled.
In the classical model for the active stress function proposed in [396], P A depends on
the transmembrane potential Vm solely. In particular, for each spatial point, the following
ODE equation is introduced

dP A (t) £ ¤
= ε(Vm (t)) kP (Vm (t) − Vr ) − P A (t) , (96)
dt
where Vr is the resting potential, kP is a suitable parameter controlling the amplitude of
P A , and ε controls the delay of the contraction with respect to the action potential. For
example, in [218] the following expression has been proposed
−ξ(Vm −V̄m )
ε(Vm ) = ε0 + (ε∞ − ε0 )e−e ,

where ε0 , ε∞ , ξ, V̄m are suitable parameters.

Since the cardiac cell contraction is regulated by the calcium ions concentrations, the
active stress function P A can be assumed to directly depend on this specific concentration,
say cca , rather than on the more general variable Vm . For example, in [595] a system of
ODEs of the same structure of (96) has been proposed, however using cca instead of Vm
on the right hand side.

86
 In more sophisticated
q models, the active stress function also depends on the stretch in
dλ
the fiber direction λ = ab Tf C a
b f and on the fiber stretch rate [402, 317]. In compact
dt
form, these models are written as follows:
µ ¶
dy A dλ
= g y P , λ, , c, y ,
dt dt (97)
P A = gP A (λ, y) ,

for suitable functions g y and gP A and where c, as in Section 7.1, is the ionic concentra-
tion variables (in particular the calcium one), whereas y collects other myofilament and
electrophysiology state variables.
Alternatively to the decomposition (93) where the stress tensor is split into a passive
and an active component (active stress approach), a different strategy based on an active
strain approach has been proposed in [98, 7]. In this case, the following multiplicative
decomposition of the deformation tensor is introduced

F = F P F A,

where F A is the active factor acting at the microscales and representing a distortion of the
fibers not preserving in general the compatibility of the body, whereas F P is the factor
describing the passive deformation at the macroscales and the deformation needed to
restore compatibility. With this choice, quite common when analyzing plasticity and the
remodeling of living tissues, the microscale information related to the fiber contraction is
directly incorporated in the body kinematics, allowing for the inclusion of fiber contraction
driven by the depolarization as a prescribed active deformation rather than a further
contribution to stress. In this case, the first Piola-Kirchhoff stress tensor is given by

b s = J A DΘ (F A )−T ,
T
DF P
and, in analogy with (95), a possible expression for F A is

F A = I − βVm a
bf ⊗ a
bf ,

for a suitable parameter β [7]. The active strain approach is probably more satisfactory
from the modeling point of view, since, unlike active stress, it should provide the expected
fiber contraction without any tuning. However, from the numerical point of view it is more
problematic since standard Finite Element finite elasticity solvers cannot be directly used
and should be properly adapted. From now on we refer only to the active stress approach.
For a comparison between the two approaches we refer to [7, 8], and to [450, 496] for
related computational results.

7.2.2 The coupled electro-mechanics problem

The values of the transmembrane potential Vm or calcium ion concentration cca to be
used in (96) or (97) to compute the active stress function P A are sometimes prescribed

87
as given data to the mechanics problem, see e.g. [157]. However, in most of the cases,
they are obtained from the bidomain or monodomain equations. This leads to a coupled
electro-mechanical problem.
The electric propagation problem needs to be solved in a domain that changes in time,
because of the cardiac contraction and relaxation. Under suitable assumptions [119], in
an Eulerian framework, these problems assume the form (75), (76), or (81), provided that
the conductivity tensors (74) or (82) are computed by using the deformed unit directions
af , as , an , see e.g. [289]. However, for computational simplicity, most often the bidomain
and monodomain equations are set in a Lagrangian framework and written in the reference
undeformed configuration. We report in what follows the corresponding monodomain
equation (according to the notation of Part 1, we set Ωmus = Ω0mus ):
à !
∂ Vbm ∂ψ ³ ´
Jχm Cm − F −T ∇Vbm · − ∇ · JF −1 ΣF −T ∇Vbm +
∂t ∂t (98)
³ ´
Jχm Ieion Vbm , w,
b bc, λ = J Ibext in Ωmus ,

coupled with the ODE systems (79e)-(79f) (similar arguments hold true for the bidomain
equations as well). The spatial derivatives have to be intended with respect to the unde-
formed domain, ψ is the deformation map between Ω0mus and Ωtmus and λ is the stretch
in the fiber direction introduced above.
From the previous equation, we observe that there are three sources of mechano-electric
feedbacks, i.e. three ways through which the mechanics problem influences the electric one:
b = JF −1 ΣF −T depends on the deformation gra-
i) the effective conductivity tensor Σ
dient F ;

ii) the ionic current term Iion depends on the stretch in the fiber direction λ;

iii) the relation between the spatial and material time derivatives introduces the advec-
tion term −F −T ∇Vbm · ∂ψ
∂t which depends on the solution of the mechanics problem
[119].
The first and the third terms are due to the pulling back of the monodomain equation
in the undeformed domain (geometrical feedback). The second one is instead due to a
well-known physical process consisting in the opening of ion channels under the action of
deformation (stretch-activated channels, see Sect. 5.4) [299, 300]. Accordingly, the ionic
current term is written as the sum of two contributions,
³ ´ ³ ´
Ieion = Iion Vbm , w,
b bc + I SAC Vbm , λ , (99)

where Iion represents one of the classical independent-stretch models described in Section
7.1.2, and I SAC is the current activated by the deformation. A quite general expression
for the latter term is given by
X
I SAC = Ki (Vbm )(λ − 1)(Vbm − Vi )H(λ − 1), (100)
i

88
where the i − th term of the sum represents the stretch-activated currents related to
the i − th ion (usually sodium and potassium), Ki is a suitable function, Vi the Nerst
potential introduced in Section 7.1.2. The Heaviside function H(·) guarantees that the
stretch-activated ion channels open only under fiber tension (λ > 1) [404]. In [119], it was
shown (computationally) that these terms do not significantly alter the morphology of the
action potential, however they strongly influence the action potential duration.
Stretch-activated ionic currents together with the active stress component of the stress
tensor (or equivalently, to the active strain component of the deformation gradient) make
the electro-mechanics problem a highly coupled system, that can be summarized as follows:
Find the muscle displacement d, the transmembrane potential Vm , the gating variables
w, and the ionic concentration variables c, such that
à à !!
P A d b
d
∇· T b +T
b (d) b b b
c, d, =0 in Ωmus , (101a)
s s
dt
à !
A d b
d
Tb = P A af ⊗ a bf , PA = A b b
c, d, in Ωmus , (101b)
s
dt
³ ´
Vbm = M Vbm ; d,b w,b b
c in Ωmus , (101c)
b b
(w, c) = G(w,
b bc; Vbm ) in Ωmus . (101d)

In the previous system, equation (101a) stands for the mechanical problem. Notice that,
unlike for the vessel case, see (10), the time dependent inertial term has been neglected,
as commonly done for the heart mechanics. Moreover, T b P = ∂Θ , where Θ is given for ex-
s
∂F
A
ample by the orthotropic law (94), and P is given by either (96) or (97). Equation (101c)
represents in compact form the monodoman problem (98) together with the ionic current
expression given by (99)-(79c)-(79d)-(100). Finally, (101d) is a shorthand notation for the
ODE systems (79e)-(79f) for the gating and ionic concentration variables characterizing
the cardiac cell model. Notice also that in writing system (101), we have exploited the
fact that λ and F could be written in terms of d, allowing us to indicate the dependences
of the monodomain problem on λ and F and of T b A on λ in compact form through d.
s
Regarding the well-posedness of the electro-mechanics coupled problem (101), very few
results have been obtained so far. We mention [433], where it has been noticed that for
the general active stress model (97), the equilibrium equation (101a) is not even elliptic
when there is an explicit dependence on the rate of stretch dλ
dt , and [9], where the existence
of a weak solution is proved for the case of a linearized elasticity equation in the active
strain formulation coupled with the bidomain equations including the geometrical feedback
affecting the conductivity tensors.

7.2.3 The issue of boundary conditions for the mechanics problem

As for the mechanical contraction problem (101a), proper boundary conditions should
be prescribed at both the external epicardium and internal endocardium surfaces. For
the former, the presence of both the pericardium and the surrounding tissue has to be

89
accounted for, because of their effect on the heart movement. A sophisticated model of
interaction with the pericardium accounting for a frictionless contact has been proposed
in [198]. More commonly, similarly to the vascular case, a Robin condition like (1) is
prescribed at Σepi .
At the endocardium internal surface Σendo , in presence of a fluid model in the atrial
and ventricular cavities, the usual dynamic and kinematic conditions (22c)-(22d), arising
from the fluid-structure interaction model, implicitly provides both the endocardial dis-
placement and normal stresses (see Section 7.5). When the fluid in the ventricular cavity is
not modeled, suitable strategies to prescribe boundary conditions at Σendo are mandatory.
A common strategy relies on prescribing a Neumann condition of type
b sn
T b = Pn
b, (102)

where P is a suitable estimate of the blood pressure distribution at the endocardium.

During the ventricular filling, condition (102) is prescribed both for atria and ventricles
by means of selected values of blood pressures taken from literature (see Figure 13).
During the isovolumic contraction, the ventricular volume is kept constant by means,
e.g., of fixed point iterations, where the value of the endocardial pressure to be prescribed
in (102) at each time step is updated until satisfaction of the constraint given by the
unchanged volume [556, 157]:

P(k+1) = P(k) + (V − V n )/Ck ,

where V is the cavity volume, k the subiteration index, Ck a penalty parameter, and
where the current temporal index n+1 is as usual understood. Alternatively, a Lagrange
multiplier approach could be employed as well [238]. During this phase the atrial pressure
is kept constant to the values reached when the mitral and tricuspid valves close.
For the ventricular ejection, a reduced 0D model (based on the analogy with electric
circuits, see Section 4.5.1) for the systemic (and in case pulmonary) circulation is usu-
ally coupled with the cardiac mechanics problem, assuming that the ventricular pressure
equals the pressure in the aorta (or in the pulmonary artery) [556, 157]. This is a good
approximation since, although the ventricular and circulatory pressures are different (thus
allowing for the acceleration and deceleration of blood, see Figure 13), they are very simi-
lar. In this case we have a coupled problem between the cardiac mechanics and the lumped
parameters model where the two subproblems exchange suitable interface conditions (e.g.
in an iterative framework). For example, the 0D model could provide the pressure to be
prescribed to the mechanics problem by means of (102), whereas the flow rate Q = ρf dV dt
is prescribed to the 0D model. During the ejection phase the atrial pressure is determined
by the venous pressure that could be obtained again by a 0D model or by literature values.
Finally, during the isovolumic relaxation, the endocardial pressure is decreased accord-
ing to literature values until it reaches the pressure atrial value. Also during this phase
the atrial pressure is determined by the venous pressure.
When only the ventricular mechanics is modeled (no atria), a truncated computational
domain like the one depicted in Figure 18, left, is considered. In this case we have to provide
suitable boundary conditions at the ventricular basis Σb . Often, this surface is kept fixed by

90
enforcing homogeneous Dirichlet conditions. In other cases, the tangential displacement is
allowed by prescribing a homogeneous Neumann condition along the tangential directions.

7.2.4 Numerical approximation

The coupled electro-mechanics problem (101) is composed by four differential blocks,
namely the two PDEs (101a) and (101c) describing the heart mechanics and electric prop-
agation, respectively, the ODE system (101b) providing the active stress function, and the
ODE systems (101d) modeling gating and ionic concentration variables. Moreover, the
algebraic source of coupling (99)-(100) need to be accounted for to determine the stretch-
activated ion currents. Their influence on the electric problem has been included directly
in the right hand side of (101c).
A common numerical solution strategy for problem (101) addresses two PDEs sepa-
rately by relying (when possible) on pre-existing mechanics and electric codes: for instance
the mechanics subproblem is solved in [436] using efficient Newton-Krylov iterations, while
the electric subproblem (monodomain+cell ODE model) is solved by means of one of the
strategies described in Section 7.1.6 (see also [535] for an operator splitting method).
In this context, at each time step the electric subproblem is solved first and the me-
chanics problem later [396, 238, 535]. In particular, after time discretization, the following
scheme is employed (as usual the current temporal index n+1 is understood and ∗ means
that the related term could be treated either implicitly or explicitly, depending on the
temporal scheme used):

EM1 Algorithm:
At each time step

1. solve the monodomain problem together with the cell model:

³ ´
bn , w
Vbm = M Vbm∗ ; d b ∗, b
c∗ in Ωmus , (103a)
b b
(w, c∗ ; Vbm∗ )
b ∗, b
c) = G(w in Ωmus ; (103b)

2. then, solve the mechanics problem:

2a. update the active stress contribution

à !
bn − d
n d bn−1
b A = P A af ⊗ a
T bf , P A
=A b b
c, d , in Ωmus ;
s
∆t

2b. solve the equilibrium equation by means of the following Newton iterations: at
each iteration k solve
à à !!
³ P
³ ´´ P A n dbn − d
bn−1
∇· DF T b b b(k) = −∇· T
δd b(k−1) ) + T
b (d b b b ,
c, d in Ωmus .
s d(k−1) s s
∆t

91
According to the notation introduced in Section 4.6.2, DF is the Gâteaux derivative with
b(k) = d
respect to F , δ d b(k) − db , and, for simplicity, we have considered a Forward
¯ (k−1)
dw ¯
Euler approximation of dt tn . The active stress contribution is treated explicitly, i.e.
it is updated once per time step. This choice may lead to numerical instabilities, as
computationally reported in [405]. In [433] it has been speculated that such instability
could be ascribed
¯ to the (necessarily)
¯ explicit time discretization of the explicit stretch
dλ ¯ dλ ¯
rate term dt tn (instead of dt tn+1 ) in the active stress function solution.
To overcome these instabilities, in [405] it has been proposed to update the active
stress function at each Newton step, i.e. to substite step 2 in the previous scheme with
the following one:

EM1/bis Algorithm:

2bis. solve the mechanics problem by the following Newton iterations: at each iteration k

2bis/a. update the active stress contribution

à !
A d bn
b(k−1) − d
b = P af ⊗ a
T A
bf , A
P =A b b(k−1) ,
c, d in Ωmus ;
s
∆t

2bis/b. solve the equilibrium equation

à à !!
³ ³ ´´ bn
b(k−1) − d
d
∇· bP
DF T dbk−1 δ d
bk = −∇· T P
bk−1 ) + T
b (d b A
b b(k−1) ,
c, d in Ωmus .
s s s
∆t

Notice that step 2bis does not coincide with a fully Newton iterations on (101a)-(101b).
Although at the expence of a higher number of iterations, this scheme provides stable
solutions and allows the use of a standard solver for the mechanics part.
A fully Newton method for the solution of the mechanics problem (101a)-(101b) used
in combination with a different explicit decoupled scheme has been successfully proposed
in [436], where the electric problem (101c)-(101d) is split into the cellular and macroscopic
problems (semi-implicit treatment) and the mechanics problem is solved in between, as
follows:

EM2 Algorithm:
At each time step

1. solve the ODE system for the gating and ionic concentration variables

b b
(w, c∗ ; Vbm∗ )
b ∗, b
c) = G(w in Ωmus ;

2. then, solve the mechanics problem by the following fully Newton method: at each
iteration k solve

92
 2a. update the active stress contribution
à !
A d bn
b(k−1) − d
b = P A af ⊗ a
T bf , PA = A b b(k−1) ,
c, d in Ωmus ,
s
∆t

2b. solve the equilibrium equation

à à !!
P
³ ´ A bn
b(k−1) − d
d
∇ · DF T b b b b b(k−1) , b(k)
s d(k−1) + DF T s c, d δd
∆t
à à !!
bn
b(k−1) − d
d
= −∇ · b P (d
T b(k−1) ) + bA
T b b
c, d(k−1) , in Ωmus ;
s s
∆t

3. solve the monodomain problem

³ ´
b w,
Vbm = M Vbm∗ ; d, b b
c in Ωmus .

A simplified variant of the previous decoupled algorithms has been proposed in [9] for
the active strain formulation. We report here, for coherence with the previous algorithms,
a variant of this scheme for the active stress formulation, where however in the second
equation of (101b) we have P A = A (b c), i.e. the active stress function does not depend on
the stretch and stretch rate:

EM3 Algorithm:
At each time step
1. update the active stress contribution
³ ´
b A = P A af ⊗ a
T bf , PA = A bcn−1 in Ωmus ;
s

2. solve the equilibrium mechanics problem by the following Newton method : at each
iteration k solve
³ ³ ´´ ³ P ³ ´´
∇ · DF T bP d bk−1 δ d
bk = −∇ · T bk−1 ) + T
b (d bA b c n−1
in Ωmus ;
s s s

3. then, solve the electric problem

³ ´
bn−1 , w
Vbm = M Vbm∗ ; d b ∗, b
c∗ in Ωmus ,
b b
(w, c∗ ; Vbm∗ )
b ∗, b
c) = G(w in Ωmus .

In [9] the convergence of the finite element solution of Algorithm EM3 towards the
continuous solution of (101) is proven.
Although segregated algorithms as those reported above are the most widely used to
numerically solve the coupled electro-mechanics problem (101), a monolithic approach has

93
Figure 21: Left: Fiber orientation in a real left ventricle obtained with the method pro-
posed in [496]. CT images from the Cardiology Division of Ospedale S. Maria del Carmine,
Rovereto (TN), Italy, and from the Radiology Division of Borgo-Trento (TN), Italy. Right:
Displacement configuration of a real left ventricle during the contraction phase at three
different instants. Orthotropic model of activation, see [35]. These numerical results
are obtained using LifeV and taken from [35]. CT images from the Cardio-surgery and
Radiology Divisions at Ospedale Sacco, Milan, Italy

Figure 22: Displacements of the two ventricles at six different instants during the systolic
contraction. Electro-mechanic coupled simulation, P1 Finite Elements. These numerical
results are obtained using LifeV; the computational mesh has been obtained by an open
source biventricular geometry segmented from CT images, see [499]

been used in [218], where a Newton method has been successfully applied to the whole

94
coupled problem with a FitzHugh-Nagumo cardiac cell model (i.e. without the ODE
systems (101d)).
In Figure 21 we report an example of fiber configuration in the left ventricle, whereas
in the same figure and in Figure 22, the results of an electro-mechanical simulation.

7.3 The ventricular fluid-dynamics

Blood flow in heart cavities Ωcav , reported in Figure 15, can be regarded as homogeneous,
Newtonian and incompressible [570, 589]. The displacements induced by the interaction
with the myocardium can be larger than 30%, hence the solution of fluid-structure in-
teraction problems is mandatory. Despite the large displacements, the ALE technique
introduced in Section 4.6.3 is suitable in this context, as shown by several authors, see
e.g. [97, 416]. Alternatively, the Immersed Boundary method [446], that does not require
remeshing neither induces mesh deformation, has been often used since the celebrated
paper by C. Peskin [375, 103] ; see Section 7.4 for its mathematical description.
One of the main features of the intraventricular fluid-dynamics is represented by the
formation of a vortex ring during diastole (i.e. the filling phase) just downstream the
mitral valve. This ring is generated during the peak flow due to pressure gradient between
left atrium and ventricle, and it is highly distorted due to the asymmetry of the mitral
valve leaflets [514]. At the end of this phase, the vortex ring is broken up into small-scale
structures that propagate towards the posterior wall and the apex. After the second mitral
peak flow due to the atrium contraction, additional vortex rings pop up. However, due to
the small duration of this phase, they do not propagate far into the chamber [322]. Similar
vortex rings are generated in the left atrium from the flow ejected by the pulmonary vein
[375]. This complex and disturbed fluid-dynamics, in presence of a quite high Reynolds
number (≃ 4000), leads to transitional effect to turbulence even in the healthy case [483].
For this reason, inclusion of turbulence model in heart simulations is common practice; see
e.g. [102] for LES models and [537] for the Variational Multiscale (VMS) formulation, that
includes in a unified framework some stable schemes controlling the numerical instabilities
arising in the convective regimes (e.g. SUPG) and LES turbulence models [268].
Regarding boundary conditions, at the endocardium Σendo (see Figure 15) the kine-
matic and dynamic conditions are implicitly determined by the coupling with the structural
problem holding in the myocardium, leading to a fluid-structure interaction problem (see
Section 7.5). The solution of a coupled fluid-structure interaction (FSI) problem can be
avoided in those cases where the wall velocity can be derived from dynamic MRI or CT
images. This requires however having at disposal several (20-30) wall displacement fields
at different instants of an heartbeat. The wall myocardium velocity (that provides the
Dirichlet boundary condition for the fluid equations in the ventricle) can be generated
by numerically differentiating the displacement field [295, 375]. Alternatively, the wall
myocardium velocity could be provided by the solution of the electromechanics model in
case the latter is segregated from the fluid model [103].
The heart has also four artificial sections that delimit it from the circulatory system,
namely the aortic and pulmonary valve orifices (outlets) and the interfaces with the venae

95
cavae and the pulmonary vein (inlets). In these cases, suitable conditions could be obtained
by coupling the heart fluid-dynamics with the aorta, the pulmonary artery, and the above-
mentioned veins. These can be modeled by means of 3D, 1D or 0D models.
In the case of simulation in the left ventricle solely (see Figure 18, right) we have to
prescribe suitable boundary conditions also at the mitral valve orifice Γmitr . For example,
in [295] it has been proposed to prescribe the flow rates at Γmitr and at the aortic valve
outlet Γao obtained by measuring from MRI images the volume changes of the left ventricle
chamber, noticing that the two valves are never simultaneously open. Alternatively, flow
rates curves taken from literature could be applied as well, see e.g. [416].

7.4 Valves modeling

Heart valves are made of thin leaflets. They do not cause any resistance to the blood
during systole and sustain large pressure gradients during diastole. Moreover, unlike
vascular vessel wall, they are subjected to very large displacements. These features make
the mathematical and numerical modeling of the valves mechanics and of their interaction
with the blood flow very demanding, requiring ad hoc techniques for their description.
Our main focus will be on the aortic and mitral valves of the left ventricle, the most
studied from the mathematical point of view. For recent reviews on valve modeling see,
e.g., [363, 580].

7.4.1 Mechanical models of the leaflets

Due to fibers alignment, which occurs mostly in the circumferential direction a, and to the
presence of collagen, most of the valve mechanical models proposed so far are of non-linear
transversally isotropic type. In particular, they depend on the two invariants I1 = tr C
and I4 = ab · (C ab ). For example, in [369, 463] the following strain energy function has
been used h i
2 2
Θ(I1 , I4 ) = c eb1 (I1 −3) +b4 (I4 −1) − 1 ,

for suitable constants c, b1 , b4 . Variants of the previous energy have been considered e.g.
in [271, 370, 592]. For a comparison of numerical results obtained with these constitutive
laws, see [18]. More sophisticated energy functions include also the microstructure of the
leaflets tissue, see e.g. [152].
Due to their small thickness, heart valves are often modeled as membrane shells by
neglecting the shear bending forces that are one order of magnitude smaller than the in
plane stresses [371, 296, 262].
Some works on the mitral valve consider also the presence of the chordae tendinae
(see Sect. 5.1). They comprise independent rings models, where the chordae are modeled
as non-linear springs with zero elasticity during compression and exponential-like stiff-
ness during traction [310, 353], and one-dimensional models characterized by a non-linear
energy function [152, 581, 424].
Another difficulty arises when modeling the contact among the leaflets during the
closure. A common strategy to handle this process is given by a penalty method [262, 363,

96
353]. This is based on measuring the distance between each vertex and the closest leaflet
and on locating a spring between the vertex and the collision point when the distance is
below a critical value. In [14] a more sophisticated algorithm has been proposed, based on
the introduction of subiterations that guarantee the satisfaction of the contact constraint
by the introduction of Lagrange multipliers that act as a force of repulsion among the
leaflets.
The numerical simulation of valves mechanics has been usually obtained by prescribing
a pressure difference between the two sides of the leaflets, mimicking the effect of the
fluid [217, 580]. However, more realistic models are obtained by considering the FSI
problem arising between the blood flow and the leaflets. This is the topic of the next two
subsections.

7.4.2 Reduced models for fluid-valve interaction

As noticed, the simulation of the fluid-structure interaction between blood and valve
leaflets is computationally demanding. However, if one is interested only in the fluid-
dynamic quantities and not in the internal leaflet stresses, reduced models could be con-
sidered.
The first family does not explicitly represent the leaflets, rather, only the open-
ing/closure mechanism of the valve is modeled through the description of the valve orifice.
The simplest strategy consists in a priori operating the opening and closing of the valve on
the basis of a flow rate or pressure profile obtained from literature or from clinical data,
see e.g. [166]. For an example, see [41, 569, 398] for the flow simulation in the aorta and
[295] for that in the left ventricle. A more realistic situation is obtained by simulating the
opening/closure mechanism of the aortic valve orifice by means of the following conditions
prescribed to the fluid problem:

if Pu > Pd then the valve opens;

(104)
if Qd < 0 then the valve closes,

where Pu is the upstream pressure, Pd and Qd the downstream pressure and flow rate [187].
The quantities Pu , Pd , Qd could be prescribed as data, or else being the results of the
numerical computation of the flow field. For example, in a fluid-dynamic simulation in the
aorta, Pu is the prescribed left ventricular pressure, whereas Pd and Qd are the unknown
aortic pressure and flow rate. For a left ventricular flow simulation, the situation is more
involved as we have to distinguish between the mitral and the aortic valve: for the former,
Pu is the prescribed atrial pressure, whereas Pd and Qd are the unknown left ventricular
pressure and flow rate; for the latter, Pu is the unknown left ventricular pressure, whereas
Pd and Qd the prescribed aortic pressure and flow rate. Many works considered a zero-
dimensional model of the systemic circulation to provide the latter quantities [593]. In
any case, the opening/closure mechanism is not prescribed a priori. This leads to a non-
linear boundary condition for the fluid problem at the valve orifice which, in the electrical
circuit analogy of zero-dimensional models (see Section 4.5.1), is represented by an ideal
diode. Similar models have been proposed in [303, 56], to account for diseased valves, by
introducing a suitable non-ideal law for the diode. More refined reduced models consider

97
the opening/closure mechanism not simply in an on-off mode, rather the open part of the
orifice dynamically changes continuously by projecting the supposed leaflet position on
the orifice plane. For example, in [533] the dynamics of the valve orifice opening/closure
is given by a prescribed law, whereas in [302] it is given by

(1 − cos θ)2
AV = ,
(1 − cos θmax )2

where AV ∈ [0, 1] is the percentage of open orifice area, θ the opening leaflet angle (θ = 0
means closed valve) and θmax the maximum opening angle. The value of θ is determined
by the following ODE
d2 θ dθ
2
+ k1 + k2 (Pd − Pu ) cos θ = 0, (105)
dt dt
for suitable parameters k1 and k2 .
For all these models, that do not explicitly include the leaflets, there could be the need
to switch between Dirichlet to Neumann (and vice versa) boundary conditions along a
single heartbeat in a fluid problem. Indeed, a Dirichlet condition is usually prescribed when
the valve is closed (e.g. homogeneous in the physiological case), whereas a Neumann or a
resistance condition could be preferred when the valve is open. This could be problematic
at the numerical level, in particular in view of the implementation. In [537], a new way to
overcome this problem has been proposed, based on a Robin-like condition implemented
by means of the extended Nitsche’s method proposed in [282]. We detail here for the
sake of simplicity the case of a Neumann-like condition for the open valve and we refer
the reader to [537] for the more physiological case of a resistance condition. Let Γ be the
valve orifice at hand and suppose that one wants to prescribe at each time the following
conditions:

u=g if the valve is closed; (106a)

Tfn = h if the valve is open. (106b)

For example, g could be obtained by a prescribed flow rate (null in the physiological
case) by assuming a priori the shape of the velocity profile (e.g. flat), whereas usually
h = P n, with P a prescribed mean pressure. Thus, the following term is added to the
weak formulation (9) of the fluid problem:
Z Z Z
γh γh 1
− T f (v, p)n · w dω − (v − g) · (T f (w, q)n) dω + (v − g) · w dω
Γ ξ + γh Γ ξ + γh Γ ξ + γh
Z Z
ξγh ξ
− (T f (v, p)n − h) · (T f (w, q)n) dω + h · w dω,
Γ ξ + γh Γ ξ + γh
where ξ ∈ [0, +∞) and γ > 0 are two parameters and h is the mesh size (possibly not
constant). This discrete problem preserves the usual Finite Elements accuracy. We notice
that for ξ = 0 we recover the classical Nitsche method to prescribe the Dirichlet condition
(106a), whereas for ξ → +∞ the formulation is consistent with the Neumann condition
(106b). Thus, the idea is to change the value of ξ (0 or +∞) allowing for the switch

98
between Dirichlet and Neumann conditions. Alternatively, the switch between a flow rate
and mean pressure condition could be prescribed by means of the defective version of the
extended Nitsche method proposed in [565].
A second family of reduced strategies includes a reduced model of the leaflets without
solving the full 3D mechanical problem. These methods are very useful when one wants
to determine with a good accuracy the influence of the leaflets on the direction of the fluid
jet. A first simple model is based on including the position of the open leaflets obtained by
a preliminary mechanic simulation and considering an on-off opening/closure mode, see
e.g. [70]. A similar model has been proposed in [16], where however an immersed resistive
approach has been considered to switch between the open and closed configuration. In
particular, the following term is added to the variational formulation of the fluid problem
Z Z
Ro v · w dγ + Rc v · w dγ,
Γo Γc

where Γo and Γc are the surfaces representing the open and closed leaflets configurations,
respectively, and Ro and Rc the corresponding resistance, which act as penalization terms
to enforce the no-slip condition v = 0 at the leaflet. Thus, when the value of the resistance
is high, the corresponding surface configuration is physically included in the model, the
velocity is zero at the leaflet and, accordingly, a pressure drop is generated across the
leaflet; on the contrary, when the value of the resistance is zero, the leaflet is invisible to
the model, no constraint is prescribed on the velocity and no pressure drop is generated.
The switch between large and vanishing values of the resistances is determined by (104).
More sophisticated models account not only for the open and closed configurations
of the leaflets, but also provide a simplified dynamics of the whole valve opening/closure
mechanism. For example, in [314, 171] a leaflet Γ is represented as a surface described
by two embedded level-set functions which depend on a single scalar function of time, i.e.
the opening angle θ. The latter is determined by a relation very similar to (105). At each
time step, the term Z
R(v − v leaf ) · w dγ,
Γ
is added to the weak formulation of the fluid problem, in order to guarantee a no-slip
condition at the leaflet (v = v leaf , where v leaf is the leaflet velocity determined from θ)
by selecting the resistance R large enough. A similar approach has been recently adopted
also in [19], where however the leaflet is represented in a more simplified way by means of
the opening angle solely (no level set functions involved).
In Figure 23, we report an example of fluid-dynamics across the aortic valve obtained
by means of the method proposed in [171].

7.4.3 Interaction between blood and leaflets

To accurately model the interaction between blood and valve leaflets, aimed, for example
to a detailed computation of the internal leaflet stresses, the solution of a fully FSI problem
is mandatory.

99
 velocity [m/s] pressure [mmHg]
1.68 87.4
1.6
80

1.2
60

0.8
40

0.4
20

0 9.99

systolic peak late diastole

Figure 23: Fluid-dynamics in the ascending aorta with patient-specific aortic valve. These
numerical results are obtained using LifeV. See [171] for a complete overview of the results

Unlike for the blood vessel and ventricular simulations, the Arbitrary Lagrangian-
Eulerian (ALE) formulation, described in Section 4.6, is not suited to handle the FSI
problem arising between blood and heart valve leaflets. Indeed, due to the large displace-
ments of the leaflets, the fluid mesh becomes highly distorted, producing severely stretched
mesh elements and thus requiring a frequent remeshing of the grid. For this reason, spe-
cific FSI techniques have been developed specifically for the numerical solution of this
problem. The most successful are the Immersed Boundary method, the Fictitious Domain
approach, and the Cut-FEM, together with their numerous variants. All these methods
are characterized by a fixed fluid mesh and allow to treat non-conforming interface meshes.

Immersed Boundary method. The celebrated Immersed Boundary (IB) method has
been specifically developed for the fluid-dynamics in the heart in [446]. The leaflets are
represented by membranes in a Lagrangian framework thought as a part of the fluid,
represented in an Eularian framework, and exerting to the latter internal localized forces.
This is achieved by introducing a Dirac delta distribution, δ, on the membrane.
The IB method has been originally developed for finite differences spatial discretization,
see e.g. [446, 374, 448]. In this case, in order to avoid leaks, the Lagrangian grid should
be fine enough so that the distance between two adjacent points is less than the Eulerian
mesh size. Then, the structural forces are interpolated into the fixed nodes of the fluid
mesh. The major issue when finite differences are used is the approximation of δ. Usually,
this is obtained by a function δh which is non-singular for each h and tends to δ for h → 0
[448].
Here, we report the variational formulation of the IB method proposed in [65] which
is useful for a Finite Element approximation and does not require any specific approx-

100
imation of the delta distribution. Given a two dimensional fluid domain Ωf , let Γs be
the immersed (one-dimensional) structure, whose material points are located at each t in
X(s, t), where s ∈ [0, L] is the Lagrangian coordinate. We consider as an example the
case of a massless linear elastic structure with elasticity constant κ. Thus, referring to the
notation introduced in Section 4.1 and assuming homogeneous boundary conditions, the
weak formulation of the IB methods reads for each t > 0 [66]: Find v ∈ V , v = v 0 for
t = 0, p ∈ L2 (Ωf ), and X = X(t), X = X 0 for t = 0, such that
Z Z L 2
∂v ∂ X(s, t)
ρf · w dω + Af (v, v, w) + B(p, w) = κ w(X(s, t))ds, (107a)
Ωf ∂t 0 ∂s2
B(q, v) = 0, (107b)
∂X
(s, t) = v(X(s, t), t), (107c)
∂t
for all w ∈ V and q ∈ L2 (Ωf ). In fact, the right hand side of (107a) guarantees the
satisfaction of the normal stress continuity at the membrane, whereas (107c) is the velocity
continuity.
For the Finite Element approximation of the previous problem, usually the forcing
term is integrated by parts. For example, in the case of linear Finite Elements for the
structure, it becomes for each t > 0 [66]
M −1
à !
X ∂X i+1 ∂X i
κ h
− h
w(X ih ),
∂s ∂s
i=0

where X ih = X h (si ), i = 0, . . . , M , is the Finite Element approximation of X and si are

the nodes of the one-dimensional mesh.
Regarding the time discretization, a first approach consists in considering a fully im-
plicit scheme. For the backward Euler approximation, unconditional stability is guaran-
teed [552]. However, this scheme requires the introduction of subiteratons to handle the
coupling between fluid and structure in (107). A more useful approach relies on treating
the right hand side of (107a) explicitly and the remaining of the equation implicitly. The
position occupied by the structure is then updated in a second step by means of a suit-
able implicit approximation of (107c). This scheme has been proved to be stable under a
CFL-like condition [66].
Several extensions and applications of the IB method have been provided. Among
them, we mention: the variational formulation for the case of a thick Rimmersed struc-
ture proposed in [67], where the right hand side of (107a) becomes Ωs T b t)) :
b s (d(s,
∇w(X(s, t))ds, s being in this case a vectorial Lagrangian coordinate; a mesh adap-
tive approach used in combination with a second order scheme, see [221]; the use of a
curvilinear fixed fluid mesh to improve flexibility and efficiency (CURVIB approach, see
[71]); the application to realistic three-dimensional mitral and aortic valves successfully
addressed in [222, 608].

Fictitious Domain approach. The Fictitious Domain (FD) approach was first intro-
duced in [216, 52] and then applied to the contest of heart valve in [243, 129]. FD can be

101
regarded as the dual of the IB method, in the sense that in the latter a weak enforcement
of the normal stresses continuity at the FS interface is added to the weak formulation of
fluid problem, whereas in the FD approach the velocity continuity is weakly enforced at
the FS interface. This is achieved by means of Lagrange multipliers, as detailed in what
follows (we refer to Sections 4.1, 4.2 and 4.3 for the notation, in particular variables with b
are referred to the reference domain, and we assume homogeneous boundary conditions):
b ∈ D, λ(t) ∈ H −1/2 (Σt ) and p(t) ∈ L2 (Ωt ), such that
Find v(t) ∈ V t , d(t) f
Z Z b Z ³ ´
∂v ∂2d b b : ∇b
ρf · w dω + Af (v, v, w) + B(p, w) + ρs ·b
e dω + Ts d e dω
Ωf ∂t
t Ωs ∂t2 Ωs
Z
= λ · (w − e)dσ,
Σt
B(q, v) µ
Z = 0, ¶
∂d
µ· v− dσ = 0,
Σt ∂t
df = d t
at Σ ,
(108)
for all w ∈ V , be ∈ D, µ ∈ H −1/2 (Σt ) and q ∈ L2 (Ωf ). For the well-posedness of the
discretized-in-time version of problem (108) we refer to [191]. We notice that the use of
the same Lagrange multiplier λ from both the fluid and structure sides guarantees also
the continuity of normal stresses [191].
A validation with experimental measures has been successfully reported in [528], whereas
an application of the FD approach for three-dimensional problems in combination with a
contact model for the leaflets has been reported in [14].

Cut-FEM approach. A natural strategy to handle an internal interface cutting in an

arbitrary way the mesh consists in writing two weak formulations of the problem at hand,
one for each of the two subdomains generated by the presence of the interface, and then
to sum them up. In this case, the meshes of the two subdomains are fixed (background
meshes). Since some of the mesh elements are cut by the interface, their contribution in
the matrices assembling is split into two parts, see Figure 24, left. This method is known
as Cut-FEM [240, 241]. Here, the jump between the normal stresses at the interface is
determined by the physical interaction with the interface, as happens for the blood/valve
interaction.
A common strategy used to guarantee the satisfaction of the interface continuity con-
ditions is the Discontinuous Galerkin (DG) mortaring (often referred to as Nitsche mor-
taring). This is a very effective choice since DG method prescribes interface conditions
in a weak sense, thus allowing a great degree of flexibility of the solution at the interface
[240, 241].
Finally, we notice that the implementation of the Cut-FEM method is not standard
since in the cut elements we may have polygonal elements, see for example Figure 24, left,
where the original triangle is split into a triangle and a trapezoid. A solution is offered by
the Extended Finite Elements method (X-FEM), where the FE space is suitably enriched
in order to make easy the tratment of the cut elements. In particular, the basic idea is to

102
Figure 24: Element mesh K cut by the interface Γ (left) and FS domain (right). For the
latter, we notice that Ω1f and Ω2f are two non-overlapping subdomains, whereas the related
computational meshes feature an overlap (in grey) in view of the X-FEM approach

duplicate the degrees of freedom in the cut elements and to use in both the subdomains the
basis functions of the original triangle either to represent the FE solution and to compute
the integrals [240].
To provide a concrete example of the Cut-FEM method, we report in what follows the
weak formulation introduced and analyzed in [4], related to the case of a valve embedded
in blood and represented by the membrane of equation
∂d
ρs H s + Ld = −[[T f n]] in Σ,
∂t
where Σ is the embedded membrane, L is an elliptic operator, and [[.]] denoted the jump
across the membrane. In particular, referring to Figure 24, right, let Ω1f and Ω2f be two
non-overlapping subdomains separated by Σ, and Ω1f,h and Ω2f,h two meshes containing
Ω1f and Ω2f , respectively, with an overlap region (in gray in the figure) containing the
membrane. Moreover, let V ih and Qih be velocity and pressure Finite Elements spaces
related to Ωif,h and C i one of the bilinear o trilinear forms C introduced in Section 2
restricted to Ωif,h . Thus we have

2
" Z #
X ∂v ih
ρf · w dω + Aif (v ih , v ih , w) + B i (pih , w) − B i (q, v ih )
Ωif ∂t
i=1
Z Z
∂dh
+sh (v h , w) + ρs Hs · e dσ + Ldh · e dσ
Σ ∂t Σ
2 · Z
X
+ − T f (v ih , pih )ni · (w − e)dσ−
i=1 Σ
Z µ ¶ Z µ ¶ ¸
i ∂dh γ ∂dh
T f (w, q)n · v ih − dσ + i
vh − · (w − e)dσ = 0.
Σ ∂t h Σ ∂t

103
The terms in the third and fourth rows of the previous formulation are nothing but the
usual DG terms guaranteeing consistency, symmetry and stability of the method, whereas
sh (·, ·) accounts for the ghost nodes stabilization [79] that guarantees an optimal con-
vergence order. This formulation has been introduced for infinitesimal displacements.
However, the authors treat also the case of a moving interface, for details see [4].
Finally, we notice that the Cut-FEM approach has been also considered for the case
of a thick structure, see [81, 614].

7.5 The total heart integration modeling

We conclude this section by providing some hints on the modeling of the complete heart
function, a challenging endeavor. The latter is very complex and requires the merge of
all the mathematical, numerical and computational issues highlighted previously in this
section, see [325, 88, 417, 470].
Basically, an heart integrated model involves the coupling of the electro-mechanical
problem (101), the blood fluid-dynamics (see Sect. 7.3), and the valve functioning (see
Sect. 7.4). The coupling between the first two subproblems occurs at the endocardium
Σendo , see Figure 15. In particular, this is determined by the classical fluid-structure inter-
action coupling conditions, i.e. the kinematic condition (22c) and the dynamic condition
(22d). In particular, referring to the notation of Sect. 4.1-7.2 and to Figure 15, we have
the following problem: Find the blood velocity v and pressure p, the muscle displacement
d, the transmembrane potential Vm , the gating variables w, and the ionic concentration
variables c, such that
µ ¶
∂v
ρf + ρf (v · ∇)v − ∇ · T f (v, p) = 0 in Ωtcav , (109a)
∂t
∇·v =0 in Ωtcav , (109b)
∂d
v= at Σtendo , (109c)
µ ∂t µ ¶¶
dd
T Ps (d) + T As c, d, n = T f (v, p) n at Σtendo , (109d)
dt
à à !!
P A d b
d
∇· T b (d)b +T b b b
c, d, =0 in Ωmus , (109e)
s s
dt
à !
A d b
d
Tb = P A af ⊗ a bf , PA = A b b
c, d, in Ωmus , (109f)
s
dt
³ ´
Vbm = M Vbm ; d, b w, b b
c in Ωmus , (109g)
b b
(w, c) = G(w,
b bc; Vbm ) in Ωmus . (109h)

After numerical discretization in time and space, the corresponding nonlinear algebraic
system is tremendously stiff and can feature a very high numerical dimension. Devising
efficient numerical strategies for its solution is a very active research area.

104
 An effective strategy to numerically solve the entire heart coupled problem is given by
the iterative solution at each time step of the electro-mechanical and fluid subproblems and
based on the exchange of the interface conditions (109c)-(109d). In particular, one of the
partitioned strategies described in Sect. 4.6.3 can be adapted and used for problem (109)
as well. Of course, at each iteration of the partitioned algorithm, the electro-mechanical
problem could be solved by means of one of the strategies reported in Sect. 7.2.
Another partitioned algorithm for the solution of problem (109) is obtained by con-
sidering the solution at each time step of the electrical problem first and the FSI problem
later [417]. This is nothing but an extension of Algorithms EM1 and EM1/bis reported
in Sect. 7.2, where steps 2b and 2bis/b are replaced by a Newton iteration over the FSI
problem.
A different approach is based on the monolithic solution of the integrated heart problem
(109) by means of a full Newton method. In particular, in [470] fully implicit and semi-
implicit methods are described. The latter is based on updating only a part of the Jacobian
as a consequence of an operator splitting technique. The authors highlighted by means of
numerical experiments that particular choices of the splitting and of the temporal scheme
could lead to numerical instabilities. Whatever strategy is adopted, the tangent problem
at each Newton iteration is solved by means of a preconditioned iterative method, which
is in this case particularly suited due to the block structure of the Jacobian, see [470].
Stability and convergence analysis of the different approaches is a field of current
investigation.

Part III
OPTIMIZING, CONTROLLING,
ADDRESSING UNCERTAINTY,
REDUCING COMPLEXITY

8 Going beyond direct simulation

In all problems considered so far, for given set of inputs (geometry, source terms, boundary
and initial conditions, etc.) the goal was to solve numerically a forward PDE problem and,
possibly, evaluate some physical indices of clinical interest, such as a flow rate across a
vessel section, the wall shear stress over a portion of the lumen boundary, the oscillatory
shear index, etc. On the other hand, efficient numerical algorithms and computational
resources nowadays available allow to tackle several additional kind of problems.
When simulating blood flows one has to face the variability of input data, like geomet-
rical features of the vessel, boundary conditions, physical coefficients (e.g. the Reynolds
number, the Womersley number, the structural model or material parameters related to

105
the vessel wall). Very often, these data vary within a broad range and are almost impos-
sible to be characterized exactly. All these inputs affect the solution of the problem under
investigation, as well as the outcomes of clinical interest. Being able to efficiently char-
acterize input/output relationships is thus important in the clinical context, in order to
investigate both intra-patients and inter-patients variability. In the former case we include
all those effects due to variations affecting a single patient, e.g., over time, or before/after
clinical intervention. Regarding the latter case, we mention for instance the morphological
variability among vessel shapes due to age, size or pathological factors (see, e.g., [599] in
the case of distal coronary anastomoses). Developing mathematical techniques capable to
detect the most relevant parameters, and then address the impact of their variation on the
outputs of interest, is thus of paramount importance. This requires the solution of many
queries to the forward problem. In this context, three classical situations we may face are
those of: (i) optimal control and optimal design, (ii) parameter identification and data
assimilation, and (iii) uncertainty quantification.

1. Optimal control and optimal design problems. When we pursue a suitable objective
through an optimal strategy to be determined, we act on some inputs of the given
partial differential equation, the state problem, such as boundary data, physical
coefficients, or the shape of the domain itself, the control variables, so that the
solution of the corresponding state problem could meet the prescribed goal. This
yields a PDE-constrained optimization problem; its numerical solution usually poses
severe computational challenges, as the state problem needs to be solved several
times while searching the optimal control in an iterative way. Exploiting numerical
optimization procedures for blood flow simulations is meant to (i) improving the
design of prosthetic devices (such as stents, grafts, ventricular assist devices) with the
aim of fulfilling a specific target, or (ii) customizing treatments or surgical procedures
provided patient-specific information can be acquired. Examples include the optimal
placement of electrodes in a defibrillation stage to damp the electrical activity in the
myocardium in a desired way when dealing with the therapeutic treatment of cardiac
arrythmias [393], the optimization of the shape of a cannula in order to maximize
the flow-rate through a ventricular assist device [364], the improvement of the shape
of a coronary bypass graft in order to possibly avoid vessel reocclusion [329, 149],
to mention a few. The first example mentioned above represents an optimal control
problem, where the control variable is a current source for the monodomain (or
bidomain) equation (see Sect. 7.1) – i.e. one of its data. The two other examples
address a shape optimization or optimal design problem to be solved, the control
variable being the shape of the domain where the state problem is set.

2. Parameter identification and data assimilation. In principle, cardiovascular models

necessitate a huge amount of data, such as patient’s radiological images and measure-
ments. In general, however, some of them may be missing – remember for instance
the issue of missing or defective boundary conditions discussed in Section 4.4, or the
lack of biological parameters characterizing the tissue properties. For instance, it is
very hard to estimate electrical conductivity of the myocardium for electrophysiology

106
 models, whereas (pointwise) measures of the electrical potential – whose mathemati-
cal modeling can be characterized by a PDE model requiring electrical conductivities
as inputs – can be more easy to acquire. By solving inverse/identification problems
in cardiovascular modeling, we aim at identifying those inputs which are unknown
or affected by uncertainty. For that, we rely on suitable quantities which are (i) ac-
quired from measurements, such as displacements, pressures, flow rates, etc., and (ii)
obtained as outcome of a numerical simulation, and match these two sets of values.
In other words, identifying those input values yielding the acquired measurements
requires to drive the PDE outcome as near as possible to the measured quantity.
In general, vastly different inputs may have produced the observed outcome: this is
why, instead of finding the most likely input configuration resulting in the performed
observation, we rather rely on statistical inversion theory, in order to incorporate
all possible information about the unknown inputs we may have prior to the mea-
surement. This requires to reformulate inverse problems as problems of statistical
inference, e.g. by means of Bayesian statistics. A second issue is data assimilation:
this rather generic term encompasses a wide range of techniques exploited to merge
measures/images into the mathematical model in order to improve numerical simu-
lations. Not only, a filtering effect induced by the numerical simulation can also be
exploited in view of noise reduction in images/measures acquisition as well.

3. Uncertainty quantification and propagation. The confidence in the outcome of a

cardiovascular simulation directly depends on the level of accuracy and certainty at
which inputs can be determined. In fact, although outcomes are computed from
inputs through a deterministic process, often input data are polluted by some ex-
perimental noise, or cannot be fully ascertained. Common sources of uncertainty in
cardiovascular simulations include (i) boundary conditions, (ii) anatomical models,
where each geometrical model is polluted by image noise, (iii) flow-split, since very
often there is a lack of clinical data to determine flow distribution to multiple distal
branches, and (iv) material properties, related to vessel walls or blood [505]. Due to
uncertainty, computational simulations have to be performed for a set of different
parameter configurations and then merged in order to determine how robust simula-
tion outcomes are with respect to variations of uncertain inputs [506]. This is indeed
strictly related with the task of parametric studies and sensitivity analyses, and can
be seen as a forward uncertainty quantification (UQ) problem. Evaluating the prop-
agation of input uncertainties through computed outputs means to evaluate suitable
statistics of the outputs (such as expected values, moments, confidence bands) that
are functions of the parameters affected by uncertainty. On the other hand, also
the solution of optimal control and inverse identification problems depends on those
experimental noises affecting observations and measures exploited during the identi-
fication process or the setup of a desired target. Evaluating the effect of uncertainties
in this case, providing suitable confidence bands about the estimated quantities –
and not only point estimates – and characterizing the statistical distribution of the
unknown parameters by taking into account a set of measured data represent in-
stances of inverse UQ problems. In this second case, quantifying uncertainties is

107
 even more important because an inverse problem is intrinsically ill-posed.

In the following sections we sharpen our discussion about these three classes of prob-
lems, showing relevant examples in cardiovascular modeling.

9 Control and Optimization

Solving a PDE-constrained optimization problem consists of reaching a target by act-
ing on a system governed by PDEs. The goal is to act on a state problem E(y, u) = 0
given by a PDE system modeling the behavior of a state variable y, through a control
variable u in order to reach a desirable target. This is usually done by minimizing (or
maximizing) a cost functional J = J(y, u) expressing some physical index of interest en-
coding the desired objective – for instance, the squared distance from a target state. For a
comprehensive presentation of the functional setting and well-posedness analysis of PDE-
constrained optimization see the monographs [252, 235]. Other classes of problems of
interest in cardiovascular applications, such as those related with parameter estimation,
can also be casted into a PDE-constrained optimization framework; see, e.g., Sect. 10.1
for further insights.

Generally speaking, a PDE-constrained optimization problem reads

min J(y, u) subject to E(y, u) = 0, y ∈ V, u ∈ Uad ⊆ U. (110)

y,u

Here V and U denote the state and the control space, whereas E : V × U → V ∗ and
J : V × U → R denote the state equation and the cost functional, respectively; V ∗ is
the dual space of V , and h· , ·iV ∗ ,V denotes the duality between two elements in V ∗ and
V ; in the same way, h· , ·iU ∗ ,U indicates the duality between two elements in U ∗ and
U. Additional constraints, depending on the problem at hand, can also be imposed, e.g.
under the form of inequalities – we express this fact by saying that we seek the optimal
control in a closed subset Uad ⊆ U of admissible controls.
We assume that the state equation E(y, u) = 0 has a unique solution y = y(u) ∈ V ,
and that the Fréchet derivative (with respect to y) Ey (y(u), u) has a bounded inverse for
each u ∈ U. Under these assumptions, the solution operator of the state equation is
continuously differentiable – we denote by y ′ (u) its derivative – and the following reduced
formulation
˜
min J(u) = J(y(u), u), u ∈ Uad (111)
u

is equivalent to (110). Note that this is a convenient framework to embrace – upon

defining suitable functional spaces and operators – both stationary and time-dependent
state problems; in the following we will make clear whether or not the proposed methods
can be used to tackle both classes of problems.

108
9.1 Optimality conditions
The solution (ŷ, û) of a PDE-constrained optimization problem like (110) requires a system
of optimality conditions to be fulfilled, including the state equation and an optimality
condition which involves the gradient of the cost functional. Indeed, if û ∈ Uad is a local
minimum of J, ˜ then it satisfies the variational inequality (also referred to as optimality
condition or minimum principle)
hJ˜′ (û), v − ûiU ∗ ,U ≥ 0 ∀v ∈ Uad . (112)
The quantity appearing at the left-hand side is the so-called Gâteaux derivative of J, ˜
˜′
evaluated at û, in the generic, admissible direction v − û; see, e.g., [551]; J (û) denotes the
gradient of J˜ at û. To express this latter quantity in terms of the state solution y, we can
use either (i) the sensitivity approach, or (ii) the adjoint approach. As we will see below,
in both cases at least a second PDE problem has to be solved in order to evaluate J˜′ (u).

9.1.1 Sensitivity approach

Computing sensitivities requires the evaluation of directional derivatives of both the cost
functional and the state solution. For any u, v ∈ U, we can write
hJ˜′ (u), viU ∗ ,U = hJy (y(u), u), y ′ (u)viV ∗ ,V + hJu (y(u), u), viU ∗ ,U . (113)
The quantity δv y = y ′ (u)v represents the sensitivity of the state with respect to the control,
evaluated at u, for a given variation v, and can be obtained by solving a further PDE:
indeed, differentiating the state equation E(y(u), u) = 0 along the direction v we obtain
Ey (y(u), u)y ′ (u)v + Eu (y(u), u)v = 0, (114)
being Ey (y, u), Eu (y, u) the Fréchet derivatives of the state operator with respect to y and
u, respectively; see, e.g., [551]. Then, δv y = y ′ (u)v is given by the solution of the linearized
state equation
Ey (y(u), u)δv y = −Eu (y(u), u)v. (115)
Note that y ′ (u) : U → V is a linear operator so that, for any admissible v ∈ U, a new
problem has to be solved to evaluate δv y. Numerically, this approach is very demanding if
the whole gradient J˜′ (u) is required: in this case, the number of linearized state equations
(115) to be solved is equal to the dimension of the control space U, a task that becomes
out of reach as soon as the control has dimension larger than O(10).

9.1.2 Adjoint approach and Lagrange multipliers

A convenient alternative is based on the so-called adjoint approach. By rewriting (113) as
hJ˜′ (u), viU ∗ ,U = h(y ′ (u))∗ Jy (y(u), u) + Ju (y(u), u), viU ∗ ,U (116)
and exploiting problem (114), we can express the first term at the right-hand side of (116)
as
(y ′ (u))∗ Jy (y(u), u) = −Eu∗ (y(u), u)((E ∗ y (y(u), u)))−1 Jy (y(u), u),

109
where Eu∗ (y(u), u) is the adjoint operator2 of Eu (y(u), u). Let us introduce an additional
variable λ = λ(u) ∈ V , called adjoint state, the solution of

Ey (y(u), u))∗ λ = −Jy (y(u), u). (117)

Then (y ′ (u))∗ Jy (y(u), u) = Eu∗ (y(u), u)λ; moreover, owing to (116), the evaluation of

J˜′ (u) = Eu∗ (y(u), u)λ(u) + Ju (y(u), u), (118)

simply requires solving the state problem and a further PDE problem, no matter which the
dimension of the control space U is. Note that the adjoint problem is always a linear PDE.

The adjoint-based expression of the gradient of the cost functional (and, more in
general, a system of first-order optimality conditions) can also be obtained, in a more
straightforward way, relying on the Lagrange multipliers’ method [252]. Let us define the
Lagrangian functional L : V × U × V → R,

L(y, u, λ) = J(y, u) + hλ, E(y, u)iV,V ∗ ,

where λ ∈ V denotes a Lagrange multiplier enforcing the PDE constraint (playing the role
of adjoint variable); note that in this case the three variables are independent.
By deriving L with respect to λ and imposing that the derivative is equal to zero, we
recover the state problem,

hLλ (y, u, p), ϕiV ∗ ,V = 0 ∀ϕ ∈ V ⇔ E(y, u) = 0.

Similarly, by deriving L with respect to y, we obtain the expression of the adjoint problem
(117),

hLy (y, u, p), ψiV ∗ ,V = 0 ∀ψ ∈ V ⇔ Jy (y, u) + Ey∗ (y, u) = 0. (119)

Finally, we impose that at the optimum,

hLu (y, u, λ), v−uiU ∗ ,U = 0 ∀v ∈ Uad ⇔ hJu (y, u)+Eu (y, u)∗ λ, v−uiU ∗ ,U ≥ 0 ∀v ∈ Uad .

In this way, a system of first-order necessary optimality conditions to be fulfilled by the

optimal solution (y, u) and the corresponding adjoint state λ is given by the following
Karush-Kuhn-Tucker (KKT) system:

 E(y, u) = 0,

Ey (y, u)∗ λ̂ = −Jy (y, u), (120)

 ∗
hJu (y, u) + Eu (y, u) λ, v − uiU ∗ ,U ≥ 0 ∀v ∈ Uad .
2
The adjoint operator of E, denoted by E ∗ , is given by the relation (E ∗ ϕ, ψ)L2 (Ω) = (ϕ, Eψ)L2 (Ω) for
any ϕ, ψ ∈ C0∞ (Ω); note that no boundary condition is involved in its definition. For this reason, E ∗ is
also referred to as the formal adjoint operator.

110
In the unconstrained case Uad ≡ U, the variational inequality reduces to an equation,

hJu (y, u) + Eu (y, u)∗ λ, viU ∗ ,U = 0 ∀v ∈ U,

so that (120) can be viewed as the Euler-Lagrange system for the Lagrangian functional,
that is, the solutions of (120) are the stationary points of L(·, ·, ·), i.e.

∇L(y, u, λ)[w, v, η] = 0 ∀(φ, v, ψ) ∈ Y × U × V.

The third inequality of system (120) allows to recover the expression of the gradient
J˜′ (u). Indeed, since E(y(u), u) = 0, it holds L(y(u), u, λ) = J(u)
˜ for any arbitrary λ ∈ V ,
so that, by differentiating this latter equality with respect to u, we obtain

hJ˜′ (u), viU ∗ ,U = hLy (y(u), u, λ), y ′ (u)viV ∗ ,V +hLu (y(u), u, λ), viU ∗ ,U = hLu (y(u), u, λ), viU ∗ ,U .

Hence, J˜′ (u) = Lu (y(u), u, λ(u)), since hLy (y(u), u, λ), ψiV ∗ ,V = 0 for any ψ ∈ V , thanks
to (119).

For the sake of illustration, we discuss a specific example, by considering the problem
of minimizing the energy
Z Z
1 2 α
J(v, u) = |v − v d | dΩ + |∇u|2 dΓ,
2 Ωobs 2 Γcon

of a fluid flow in a tract of blood vessel where a device (e.g. a bypass or a cannula) is
implanted. The function of the device is represented by a velocity control u acting on
the boundary Γcon ⊂ ∂Ωf that represents the crossing section between the device and the
vessel, see Fig. 25. The goal is to regularize the velocity pattern in a suitable observation
region Ωobs ⊆ Ωf by requiring v to be as close as possible to a desired distribution v d ; see
the related discussion in the next section. Referring to the notation in Section 4, for the
sake of simplicity, we consider a steady version of the Navier-Stokes equations (4a)–(4b),
for which the velocity-pressure couple y = (v, p) ∈ V = (H0,Γ 1 (Ωf ))d solves the state
D
problem
−∇ · Tf (v, p) + ρf (v · ∇)v = 0 in Ωf
∇·v =0 in Ωf
v = v in on Γin
v=0 on Γw
v=u on Γcon
T f (v, p)n = 0 on Γout
where the control variable is the velocity u ∈ U = (H01 (Γcon ))d imposed on the boundary
Γcon . α > 0 is a parameter penalizing the control magnitude (or cost); this can also be
seen as a regularization term, ensuring the convexity of the cost functional. The fluid
Cauchy stress tensor Tf (v, p) has been defined in (5).

111
Figure 25: Left: schematic representation of a bypass graft. Right: domain, boundary
portion and observation region for the bypass model problem

Following the Lagrangian approach, we can derive a system of first-order optimality

conditions, where the adjoint problem for the adjoint variables λ = (z, q) ∈ V reads

−∇ · Tf (z, q) + ρf (∇T v)z − ρf (v · ∇)z = (v − v d )IΩobs in Ωf

∇·z =0 in Ωf
z=0 on Γin ∪ Γw ∪ Γcon
T f (z, q)n = 0 on Γout

being IΩobs = IΩobs (x) the characteristic function of the region Ωobs . Note that the adjoint
problem is linear in (z, q), and comes from the linearization of the Navier-Stokes equations
around the state solution; the optimality condition reads instead

αu + z = 0 on Γcon .

Remark 3. In the case of time-dependent state problems, the adjoint problem is a back-
ward in time time-dependent problem. Depending on the observation appearing in the cost
functional – which can be either on the whole time interval (0, T ), or at the final time
T only – the dependency of the adjoint problem on the state is only at t = T (thus, as
initial condition) or on the whole interval (0, T ). The approach followed so far can still
be employed to derive a system of optimality conditions, see, e.g., Sect. 9.3.2 for further
details on a case of interest.

9.2 Numerical approximation

Solving a PDE-constrained optimization problem entails a computational effort bigger
than the one required for the solution of a forward (state) problem. The two formulations
(110) and (111) yield two different paradigms for the approximation of such a problem.
In the former case, both state and control variables are optimization variables and PDE
constraints are explicitly specified; in the latter, only the control variable u is an opti-
mization variable, whereas the state variable y is considered to be an implicit function of

112
u through the PDE constraint3 . In this latter case, the solution of the state problem is
nested in the evaluation of the gradient J˜′ (u) of the reduced cost functional.
Algorithms for solving PDE-constrained optimization problems can be sorted following
several criteria. a first classification criterion is between iterative methods, rooted on
iterative minimization algorithms for the reduced cost functional, and all-at-once methods,
where the PDE constraint is kept explicitly and the three equations forming (120) are
solved simultaneously. Another criterion is concerned with the highest order of derivatives
exploited by the algorithm, yielding derivative-free methods, gradient-based methods and
Hessian-based methods.
A different perspective is taken when addressing the interplay between optimization and
discretization: numerical discretization can be performed prior or after the derivation of a
system of optimality conditions. More precisely, in the so-called optimize, then discretize
approach optimization is carried out at the continuous level (e.g. to find system (120)) and
then the discretization is operated on the resulting optimality system; alternatively, using
the discretize, then optimize approach we first approximate the state equation (and the
cost functional) and then carry out the optimization at the discrete level. Here we address
the former approach; further details are given at the end of this section. For the sake
of brevity, we recall the main features of iterative and all-at-once methods in the case of
unconstrained problems, that is, problems without further equality/inequality constraints
or, equivalently, for which Uad ≡ U. In particular, Y ∈ Rny , U ∈ Rnu denote the discrete
representation of the state and the control variable, respectively, whereas λ ∈ Rnλ is the
discrete adjoint variable.

9.2.1 Iterative methods

Iterative (also referred to as black-box) methods treat the reduced problem
˜ ) = J(Y (U ), U ),
min J(U
U

once U is known; Y (U ) is obtained as the solution of the state equation. An existing

algorithm for the solution of the state equation has therefore to be embedded into an opti-
mization loop, and any available PDE and optimization routines can be freely combined.
In particular, iterative methods represent a popular choice when dealing with OCPs by
extending an existing code for the state problem. Within this class, a notable difference
exists between gradient and non-gradient based algorithms.

Non-gradient based (or derivative-free) algorithms – such as the popular Nelder & Mead
algorithm – exploit either the comparison among functional evaluations in different direc-
tions at each step, or low-order local approximants of J˜ in order to assess its local behavior
and localize the minimizer; see for instance [365]. Employing a finite difference approxi-
mation of the gradient is an attractive alternative due to its ease of implementation, but it
may suffer from limited accuracy and large costs in the presence of many design variables.
3
From the numerical standpoint, the former approach is often given the name of Simultaneous Analysis
and Design (SAND), whereas the latter is referred to as Nested Analysis and Design (NAND).

113
Hence, these methods are feasible only in the case where the control space has very small
dimension – e.g., if the control is expressed in terms of a vector of nu = O(10) design
variables.

Gradient-based algorithms exploit the gradient J˜′ to iteratively update the control until
a suitable convergence criterion is fulfilled. Remarkable instances are descent methods,
such as the gradient, (nonlinear) conjugate gradient, quasi-Newton or Newton methods.
In the simplest case of a gradient method, starting from an initial guess U (0) we iteratively
generate a sequence

U (k+1) = U (k) − τk J˜′ (U (k) ), k = 0, 1, . . .

being τk > 0 a step size, possibly varying at each step, until e.g. kJ˜′ (U (k) )k < ε, for
a given tolerance ε > 0. Further details can be found, e.g. in [415, 290]. The solver
for the state equation has to be augmented with a routine which provides the gradient
of the state with respect to the optimization variables, and hence the solution of the
adjoint problem. Relying on optimization procedures showing a faster convergence rate
is mandatory to speedup the execution of the whole algorithm. Although straightforward
to implement, the gradient method suffers from a poor rate of convergence when dealing
with the numerical solutions of PDE-constrained optimization problems; more efficient
methods, such as quasi-Newton methods, are more typically employed; see e.g. [73] for
further details.

9.2.2 All-at-once methods

By treating both the control and the state variables as independent optimization variables,
coupled through the PDE constraint, we deal with an equality constrained nonlinear opti-
mization problem, the state equation playing now the role of equality constraint. The goal
of all-at-once (also referred to as one-shot) methods is to tackle the (possibly) nonlinear
optimality system (120) as a whole problem to be solved. After numerical discretization,
in the unconstrained case we land on the algebraic system
 T
 JY (Y , U ) + eY (Y , U )λ = 0
J (Y , U ) + eTU (Y , U )λ = 0 (121)
 U
e(Y , U ) = 0.

Here e(Y , U ) denotes the discrete state operator, JY and JU are the gradients of J with
respect to state and control variables, respectively, whereas eY and eU are the Jaco-
bians of the state equations with respect to state and control variables, respectively. The
three equations of system (121) can also be seen as the conditions obtained by requiring
that the gradient of the discrete Lagrangian L(Y , U , λ) = J(Y , U )−λT e(Y , U ) vanishes.

The strategy above is well-suited for PDE-constrained optimization problems involv-

ing stationary state systems, however it is more computationally involved in the time-
dependent case. If J(Y , U ) is quadratic and e(Y , U ) is linear in Y and U , (121) is a

114
linear system of equations in saddle-point form, as those arising from quadratic program-
ming. In this case, preconditioned iterative methods for linear systems such as those based
on Krylov subspaces could be employed; in this respect, several preconditioners have been
proposed in the last decade, in which multigrid schemes are exploited as inner solvers (or
preconditioners) for some blocks of the KKT matrix within an outer iterative solver; see,
e.g., [47, 488]. More recent extensions to constrained problems have been addressed, e.g.,
in [73] and the references therein.
If the state problem is nonlinear, the optimality system has to be solved through
appropriate linearization procedures (like sequential quadratic programming methods) or
modern penalty methods (like augmented Lagrangian methods). Indeed, the equations
of system (121) are still linear in λ but are nonlinear in (Y , U ). When a Newton-type
method is applied to (121), each iteration on the KKT system entails the solution of the
following linear system
    
LY Y LY U eTY δY JY
 LUY LUU eTU   δU  = − JU  (122)
eY eU 0 λ(k+1) e

for the update δX T = (δY , δU )T of the optimization variable and the new value λ(k+1)
of the adjoint variable; then, we set (Y (k+1) , U (k+1) )T = (Y (k) , U (k) )T + (δY , δU )T . Here
µ ¶
LY Y LY U
H=
LUY LUU

denotes the Hessian of the Lagrangian functional, to be evaluated at (Y (k) , U (k) , λ(k) )T .
Note the saddle-point structure of system (122), where the adjoint variable indeed plays
the role of multiplier for the constraint expressed by the (linearized) state equation. The
system (122) can be equivalently obtained as the KKT system for the following linear-
quadratic optimization problem, see, e.g., [252, 73],
" µ ¶T # µ ¶
1 T JY eY
min δX HδX + δX subject to δX + e = 0,
δX 2 JU eU

whence the name of sequential quadratic programming method which is commonly used
to refer to the Newton iterations on system (122). Indeed, a quadratic programming
problem has to be solved at each step, e.g. by means of a preconditioned Krylov mehod
like MINRES, until a suitable convergence criterion is fulfilled. Suitable approximation of
the Hessian, based e.g. on quasi-Newton methods, are required to make this algorithm
more computationally attractive, see, e.g., [73].

Remark 4. Constraints on the control and/or state variables add nonlinearity to the
optimization problem. A first option is to treat inequality constraints in an outer loop,
through penalty methods which allow to convert them into additional terms in the cost
functional. For instance, if u ≤ b is a pointwise control constraint, the term 2c k max{0, u −
b}k2U can be added, being c > 0 a penalty parameter to be properly selected, and then an

115
iterative method can be used in the inner loop. Another option in iterative methods is
to perform a projection over the space of admissible controls at each step. More efficient
strategies to tackle constrained problems are usually obtained when dealing with constraints
in the main optimization loop, such as in the case of primal-dual active set strategies, see,
e.g., the monographs [73, 252].

We finally point out that in this section we have opted for the optimize, then discretize
approach; that is, we have shown how to recover a system of optimality conditions and then
proceed to its numerical discretization. The opposite strategy (discretize, then optimize)
would have led to substantially similar numerical methods, by setting either an iterative
or an all-at-once method on the system of optimality conditions derived once the original
state system had been discretized. The two approaches do not yield identical solutions in
general; see, e.g., [252] for a discussion.

9.3 Examples of applications in cardiovascular modeling

9.3.1 Optimal design of bypass grafts
PDE-constrained optimization problems are obtained when looking for the optimal design
of prosthetic devices, such as ventricular assist devices or bypass grafts. For instance,
coronary artery bypass grafting (CABG) is a standard surgical procedure to restore blood
perfusion to the cardiac muscle by redirecting blood from the aorta through a graft vessel
to the downstream of a coronary artery affected by stenosis or occlusion. Being able to
design the graft-vessel connection (the so-called anastomosis, which we refer to as the
domain Ω) in an efficient way is a critical factor to prevent post-operative recurrence of
restenosis.
It is nowadays accepted that intimal wall thickening, caused by the accelerated growth
of smooth muscle cells and surrounding matrix, represents one of the leading causes of long-
term failure of end-to-side vascular grafts [244]. Low and/or highly oscillatory patterns of
WSS, as well as strong vorticity and recirculations, cause intimal wall thickening [160, 294]
at sites where curvatures, bifurcations, tortuosity and branching occur, and, more in
general, where flow departs from unidirectional patterns [213, 342]. In mathematical
terms, an optimal graft is one that minimizes suitable cost functionals involving the area
of low WSS, spatial WSS gradient [329] or the vorticity [473, 358], see e.g. [298] for a
detailed review. Taking as state variables blood velocity v and pressure p, and as control
variable u = Ω, i.e. the shape of the domain itself, the goal is thus to find the optimal
shape Ω̂ of the graft by minimizing, e.g.
Z T Z
J(v, Ω) = |∇ × v|2 dΩ dt,
0 Ωobs

where Ωobs ⊂ Ω is a given observation region, in the artery portion right after the anasto-
mosis. Indeed, high downstream vorticity may lead to strong flow recirculation, yielding
similar effects in terms of intimal thickening.
Other cost functionals that can be employed are, for instance:

116
 • a tracking-type functional, in order to drive the blood velocity (and pressure, if
δ > 0) towards a specified velocity (and pressure) target state v d , pd , featuring a
regular pattern
Z Z Z Z
1 T δ T
J(v, Ω) = |v − v d |2 dΩdt + |p − pd |2 dΩ dt;
2 0 Ωobs 2 0 Ωobs

a typical choice of (v d , pd ) is provided by the Stokes (or a low-Reynolds Navier-

Stokes) flow in the same domain;

• in dimension two, a Galilean invariant vortex measure identifies a region to be a

vortex if ∇v has complex eigenvalues, that is, if det(∇v) > 0; following [251, 309,
318], the functional
Z TZ
J(v, Ω) = max(0, det(∇v)) dΩ dt
0 Ωobs

can be used as a cost functional when dealing with vortex suppression;

• a WSS gradient-based functional is given by

Z Z õ ¶2 µ ¶2 !
T
∂wp ∂wn
J(v, Ω) = W SSG(t) dΓ dt, W SSG(t) = +
0 Γobs ∂τ p ∂τ n

that is, by a time-average of the WSS gradient, measured over the portion Γobs ⊂ ∂Ω
of the boundary in the anastomosis region. Here we refer to WSSG as to the WSS
gradient, being w = w(t, x) the WSS vector of components w(j) = ∇vn(t, x) · τ (j) ,
τ p and τ n the unit vectors parallel and normal to the direction of the time-averaged
WSS vector w, respectively. Only a linear combination of the normal components
∂xj /∂τ j , j = p, n is considered as index to quantify the tension yielding to intimal
thickening. Moreover, the time-averaged WSS is relatively insensitive to changes
in the anastomosis configuration, while the time-averaged WSS gradient is highly
sensitive, and has been linked to localized mechanobiological responses in tissues (see
[329] and discussion therein). Using WSSG rather than WSS can be understood as
filtering the WSS by removing its component induced by the steady mean flow and
considering only the spatially fluctuating term as part of the indicator. For the sake
of numerical efficiency of the whole optimization process, a steady flow simulation
(e.g., corresponding to the systolic peak) can be considered, and the WSS gradient
of the steady flow can be taken as an approximation of the time-averaged WSS
gradient. The interested reader can refer, e.g., to [298] for further details.

Many works have focused in the last decades on the optimal shape design of end-to-side
anastomoses, typically by acting on the wall shape near the anastomosis by local shape
variations. If a simpler parametric optimization approach is considered, the three most
significant design variables are [342]: the anastomosis angle, the graft-to-host diameter
ratio [294, 144], and the toe shape, that is, the profile of the bypass junction. Optimizing

117
a complex configuration in terms of few design variables is more advantageous since it is
known that grafts usually feature an extended patch or cuff that effectively provides a
hood to ease the flow transition from the graft to the artery. Not only, improved anas-
tomosis design must comply with the constraints set by surgical realization. For a more
comprehensive review of bypass graft design results, we refer to [373, 342, 423, 364].

Concerning the objective to be minimized, each cost functional involves suitable ob-
servations, obtained as functions of the state variables over the entire domain or some
boundary portions, and over a given time interval and, possibly, data acquired from mea-
surements, like in the case of a tracking-type functional. There are of course other param-
eters affecting the distribution of physical indices like the WSS or the vorticity: this is e.g.
the case of the Reynolds number, or of the flow split between the proximal host artery
and the graft [213]. All these parameters should be considered as uncertainties within a
robust optimization framework, in order to characterize the optimal shape of the graft in
a range of possible operating conditions. First examples of optimal design in presence of
uncertainty in cardiovascular applications have been considered e.g. in [505].
As a concluding remark, we point out that the coupling of optimization algorithms to
blood flow simulations is computationally challenging, since each evaluation of the cost
functional requires the solution of an unsteady, three-dimensional Navier-Stokes problem.
If relying on the adjoint approach to characterize the gradient of the cost functional,
a further linearized Navier-Stokes (Oseen) problem has to be solved at every iteration.
For this reason, the most part of works related to applications in surgery and device
optimization have focused, so far, on small-scale and/or two-dimensional problems, or on
steady-flow problems, usually relying on gradient-free methods.
For the sake of illustration, we report some numerical results for the optimal design
of a femoro-popliteal bypass graft. This surgery is used to bypass diseased blood vessels
above or below the knee, and represents one of the most common surgical treatments of
chronic lower-extremity ischemia. We model the blood flow across a bypass graft by a
steady Navier-Stokes model and consider a vorticity cost functional
Z
J(v, Ω) = |∇ × v|2 dΩ
Ωobs

where Ωobs ⊂ Ω is the observation region. Initial and optimal shapes are shown in Fig. 26
for the case of a stenosed host artery; its occlusion, either total or partial, is expressed
through a Dirichlet boundary condition on the incoming velocity field on Γres , homoge-
neous for the completely occluded case, non-homogeneous otherwise; see Fig. 26 (b) for
the definition of Ωobs and Γref . A shape parametrization technique based on Free-Form
Deformations (FFD) is very suitable to describe admissible shapes through a deformation
of a reference configuration by acting on a small set of control points; see, e.g., [354, 358].

9.3.2 Optimal control of electrical defibrillation

While in healthy conditions the electrical activation of the heart is an extremely organized
(and efficient) process, some disturbances in the formation and/or propagation of electri-

118
Figure 26: Optimal design of bypass grafts. (a) A tract of femoral artery with the by-
pass graft; (b) computational domain, boundaries and observation region; (c) FFD shape
parametrization used to generate admissible shapes. Global shape deformations are in-
duced by the displacement of few, selected control points (shown in red) in the 6 × 4 FFD
lattice. These control points are selected by a preliminary screening procedure based on
sensitivity analysis; (d) initial and (e) optimal bypass configurations in the case of to-
tal (above) or partial (below) occlusion. These numerical results are obtained using the
Matlab Finite Element library MLife

cal signals may induce reentrant activation patterns which lead to tachycardia, that is, a
noticeable increase in the hearts activation rate. In the worst cases, this may turn to an
even less organized activation pattern, called fibrillation. A common therapy to terminate
a fibrillation and restore a regular cardiac rhythm is electrical defibrillation, consisting in
the delivery of a strong electrical shock by injecting external currents through a set of
electrodes. This restores a spatially uniform activation pattern, recovering an extracel-
lular potential distribution showing damped voltage gradients. Electrical defibrillation is
nowadays operated by implanting some devices (the so-called cardioverters defibrillators)
able to monitor the heart rhythm and then to deliver electrical discharges when needed.
This process can be modeled e.g. by considering as state system the monodomain
equation (81), for which the extracellular stimulation current Ie = Ie (t) to be applied plays
the role of distributed control function, and has to be determined in order to minimize
e.g. the cost functional
Z Z Z Z
1 T 2 α T
J(Vm , Ie ) = |Vm − vd | dΩdt + |Ie |2 dΩdt. (123)
2 0 Ωobs 2 0 Ωcon

Here Vm is the transmembrane potential, Ωobs ⊆ Ωmus is the observation domain, Ωcon ⊆

119
Ωmus is the control domain, whereas vd is the target potential distribution. For instance,
if vd = 0, the minimum of J(Vm , Ie ) corresponds to the case of an excitation wave which
is suppressed in the region Ωobs , see e.g. [391]. The transmembrane potential Vm can be
obtained by solving the following monodomain equations (see Sect. 7.1.3)
∂Vm
χm Cm − ∇ · (Σ∇Vm ) + χm Iion = Ie in Ωmus × (0, T )
∂t
∂w
= g(Vm , w) in Ωmus × (0, T )
∂t (124)
Vm |t=0 = Vm,0 in Ωmus
w|t=0 = w0 in Ωmus
Σ∇Vm · n = 0 on Σepi ∪ Σendo
with Iion provided e.g. by the Fitzhugh-Nagumo model (see Sect. 7.1.2),
Iion = f (Vm , w) = −kVm (Vm − a)(Vm − 1) − w, g(Vm , w) = ǫ(Vm − γw).
An analysis of this OC problem can be found, e.g., in [391]. We can exploit the La-
grangian approach to derive a system of first-order optimality conditions, by introducing
the Lagrangian
L(Vm , w, Ie , z, q) = J(Vm , Ie ) + hE(Vm , w, Ie ), (z, q)iV,V ∗
= J(Vm , Ie )
Z TZ µ ¶
∂Vm
+ χm Cm − ∇ · (Σ∇Vm ) + χm Iion − Ie zdΩdt
0 ΩmusZ ∂t
µ ¶
T Z
∂w
+ − g(Vm , w) qdΩdt
0 Ωmus ∂t
where (z, q) denote the dual variables of Vm , w, respectively; here (Vm , w) ∈ V =
L2 (0, T ; V) × W 1,2 (0, T ; H), being V = H 1 (Ωmus ) and H = L2 (Ωmus ); the control space
can be chosen as U = L2 (0, T ; L2 (Ω)), and the initial conditions can be kept as explicit
constraints. By setting the partial derivatives of L equal to zero, we find the following
expression for the adjoint problem:
∂z ∂Iion ∂g
−χm Cm − ∇ · (Σ∇z) + χm z− q = vd − V m in Ωmus × (0, T )
∂t ∂Vm ∂Vm
∂q ∂g ∂Iion
− − q+ z=0 in Ωmus × (0, T )
∂t ∂w ∂w (125)
z|t=T = 0 in Ωmus
q|t=T = 0 in Ωmus
Σ∇z · n = 0 on Σepi ∪ Σendo
and the following optimality condition
z + αIe = 0 on Ωcon . (126)
The OC problem (123)–(124) can be solved by an optimize-then-discretize strategy, where
both spatial and temporal discretizations are required because of the time-dependent na-
ture of the problem. After discretization, we can employ an iterative method, by com-
puting at each step the solution of the (coupled) state problem (124) over (0, T ), and

120
the solution of the adjoint problem (125), which is a linear problem, backward in time,
where the adjoint variables are coupled similarly to (Vm , w) in the state problem. Note
that the data of the adjoint problem are related with the Fréchet derivative of the cost
functional with respect to the state variables, and that the adjoint problem depends on
the control function only through the state variable. Moreover, the adjoint operator calls
into play the linearization of the state operator around the computed state solution – that
is, the derivatives ∂Iion /∂Vm , ∂g/∂Vm , ∂Iion /∂w, ∂g/∂w have to be evaluated, at each
step, around the computed solution of the state system. The optimality condition (126)
then allows to determine the gradient J˜′ (Ie ) of the cost functional J(I ˜ e ) = J(Vm (Ie ), Ie ),
required to update the control function at each step.
To simplify the OC problem, we can for instance look for control functions under the
form
Nel
X
Ie (t, x) = uk (t)IΩcon,k (x)I(0,Tdef ) (t)
k=1

where the location of Nel electrodes through which the current is delivered is prescribed,
and only its time intensity has to be controlled over time; here IΩcon,k (x) denotes the
indicator function of the region Ωcon,k where the k-th electrode is located. Additional
inequality constraints under the form umin ≤ uk (t) ≤ umax can also be taken into account
in order to limit currents amplitude.
Further extensions of this framework consider e.g. the bidomain model [393], the
case of boundary controls [392] and a different Mitchell-Schaeffer model to describe ionic
currents, together with the presence of a conductive bath medium outside the heart effects
and experimentally-based fiber and sheet orientations [394]. Moreover, the case where the
shock duration itself is also optimized has been considered in [308]. The reason to consider
this further effect is that after applying a defibrillation shock, the muscle tissue needs a
suitable amount of time to reach a non-fibrillatory state, so that a successful defibrillation
can only be assessed at a time T ≫ Tdef , being Tdef the end time of the defibrillation shock
and T the final observation time, at which the tissue should have reached a non-excited
state. In this context, the following cost functional
Z N Z
µ 2 αXel T
J(Vm , u, Tdef ) = Tdef + |Vm (·, T )| dΩ + |uk |2 dt
2 Ωobs 2 0
k=1

can be minimized, accounting for (i) a minimum time term, for (ii) a final time obser-
vation at t = T and (iii) for the amplitude of the applied currents, subject to the state
system (124) to be solved over (0, T ). Indeed, the shock duration and its energy have
to be minimized to avoid negative side effects of the applied shock, while the final time
observation term is taken into account to quantify defibrillation, thus requiring that the
tissue reaches a non-excited state at T ≫ Tdef .

For the sake of illustration, we consider the case of an axially symmetric domain
Ωmus = (0, 2) × (0, 0.8) where a constant pulse of intensity u is delivered in the control
domain Ωcon = (0, 0.25) × (0.3, 0.55) ∪ (1.75, 2) × (0.3, 0.55), until time Tdef ; here U =

121
(u, Tdef )T is the control variable, with 1 ≤ u ≤ 100 mV and 0.1 ≤ Tdef ≤ 4 ms; the final
time is T = 64 ms. The initial condition (Vm,0 , w0 )T describe a reentry wave of the “figure
of eight” type, obtained following the procedure described in [308], where more general
optimal control problems are also taken into account. For the case at hand, a planar
wave front travelling from the bottom to the top can be damped by acting an optimal
control of intensity u ≈ 95 mV until Tdef ≈ 1.2 ms on the control region. The successful
defibrillation – resulting from a trade-off between a large intensity and a short duration
of the pulse – is clearly visible in Fig. 27, where in the controlled case at the final time
the tissue is almost completely unexcited. Indeed, the pulse acts on the excitable region
of the tissue adjacent to the wave front, bringing it to a non-excitable state.

Figure 27: Electrical potential at times t = 0, 4, 12, 20, 40, 52, 64 ms in the uncontrolled case
(top) and in the controlled case (bottom). The reentry wave appearing in the uncontrolled
case is damped by the control acting on Ωcon

122
Remark 5. We point out that choosing the cost functional and, if needed, imposing suit-
able constraints on the control and/or the state variable, are two extremely hard, problem-
dependent tasks. Moreover, very often control functions are described in terms of (possibly
few) relevant parameters, which play the role of design variables; their automatic selection
however, in the case e.g. of complex, patient-dependent geometries, can result in a quite
involved procedure. Last, but not least, in the case where a target state to be reached de-
pends on acquired data, the effect of uncertainty has to be considered in the formulation
(and then, solution) of the problem. This feature may be accommodated by using, e.g., ro-
bust optimization problems, or PDE-constrained optimization problems under uncertainty.
This is indeed an active research field, where very few applications to cardiovascular mod-
eling have been considered so far.

10 Parameter estimation from clinical data

When dealing with the mathematical and numerical modeling of the circulatory system,
initial conditions, boundary conditions or physical coefficients, e.g., tissue properties,
might be (partially) unwknown [563], see Sects. 3 and 6. Hereon we refer to any of
these quantities as input parameters, independently of their mathematical nature – they
could be scalar or vectors, or even parametric fields varying in space – and to the issue
of making inferences about unknown parameters from data as to parameter estimation
(or identification). To perform parameter estimation for a system of PDEs we need to
combine state observations and additional data which are not strictly required for solving
the PDE problem, and have to be acquired from measurements.
Parameter estimation for PDEs can be done according to several techniques, which
can be coarsely classified depending on their outcome [539, 143]. Point estimates rely on
either variational or sequential methods [274]. Both methods provide optimal least-squares
estimates by minimizing a cost functional accounting for the misfit between measured data
and state observations. A second class of techniques yields instead confidence regions or,
more generally speaking, the possibility to characterize the probability distribution of the
unknown parameters provided they are described in terms of random variables; this is the
goal, e.g., of statistical inversion theory relying on Bayesian inference, which Sect. 11.2 is
largely devoted to.
In the context of cardiovascular modeling, parameter estimation is necessary for the
sake of model calibration/personalization, e.g. in view of diagnosis or treatment purposes;
indeed, parameters that are not directly measurable – this is e.g. the case of tissue
conductivity or vessel compliance – are tuned so that the outcome of the numerical model
is able to reproduce patient-specific data [305]. Difficulties arise because of data sparsity
[301]: indeed, spatial resolution and temporal frequency are typically undersampled; on
the other hand, experimental data are always polluted by measurement noise.

Remark 6. Data assimilation (DA) is the process by which a numerical model of a given
system, usually affected by noise or model uncertainties, is improved by incorporating

123
system observations. Although DA relies on the same variational or filtering approaches
addressed in this section, in the case of, e.g., geophysical fluids, its main goal often goes
beyond parameter estimation; indeed, it is more often related to state estimation, namely,
to improving the outcome of the numerical model and of its initial state to correctly ini-
tialize forecasts, by assimilating available measurements into the numerical model itself.
DA is intrinsically related to time-varying phenomena and deals with highly nonlinear dy-
namical systems, very often far from being periodical (such as in the case of meteorological
models) and ill-posed; see, e.g., [61, 577, 143] for a detailed discussion. In the last decade
several works dealing with cardiovascular applications have focused on data assimilation
[516, 135, 49, 315] which has been considered, in many cases, as synonymous of parameter
estimation.

In this section we provide an overview of both variational and sequential approaches for
parameter estimation in time-dependent systems. Parameter estimation problems dealing
with stationary systems in cardiovascular applications have also been solved – see, e.g.,
[133, 49, 347, 406, 319, 355] for the sake of space and relevance here we focus on time-
dependent problems.

10.1 Variational approach: PDE-constrained optimization

The variational approach recasts parameter estimation in the framework of PDE-constrained
optimization, by considering the equations governing the problem at hand as state system
and the discrepancy between the observation of the state and the measured data as cost
functional to be minimized [33, 92]. The parameters to be estimated (often involving the
initial condition) play the role of optimization variables, just like control variables in the
case of optimal control (OC) problems; however, differently than the OC case, parame-
ters to be estimated are quantities which no one can actually control. Often, they are
coefficients of the operator appearing in the PDE problem.
Here we provide an abstract formulation of the variational approach, following [51, 61];
for more see, e.g., [400, 89, 516]. We assume that the state problem has already been
discretized in space. We denote by X(t) ∈ Rnx the semi-discrete state, by θ ∈ Rp the
parameters of the model to be estimated, and by A(t, X(t), θ) the (semi-discretized in
space) state operator; note that usually p ≪ nx . The state variable then solves the
dynamical system ½
Ẋ(t) = A(t, X(t), θ), t ∈ (0, T ),
(127)
X(0) = G.
For instance, in the case of the FSI system (22), X = (v, p, d) contains fluid velocity and
pressure, and structure displacement, θ may contain, e.g., the value of the Young modulus
(here assumed to be piecewise constant) in different patches of the arterial wall.
We consider the case where the parameter vector θ is unknown, and for which the
b such that the discrepancy between the observation
estimation problem consists in finding θ
n
and a set of measurements Z(t) ∈ R z , t ∈ (0, T ), is minimized, e.g. in a least-squares
sense. The case of unknown initial data G to be estimated can be treated essentially in
the same way. Usually, we assume that measurements Z(t) are related to observations of

124
the true state through an additive noise model, that is,
Z(t) = H(t)X(t) + ε(t), t ∈ (0, T ),
where H = H(t) ∈ Rnz ×nx is an observation operator which maps the state space into the
observation space Rnz , and ε = ε(t) is a noise term accounting for measurement errors.
The following minimization problem is then solved
Z
1 T αθ
J(X, θ) = kZ(t) − H(t)X(t)k2M dt + kθ − θ 0 k2P −1 → min (128)
2 0 2 θ θ∈P

being X = X(t) the solution of (127). Here P ⊂ Rp × Rnx denotes the set of admis-
sible parameters. Additional information is usually added to the least-squares objective
expressed by the first term in (128), through a background estimate θ 0 of θ; M and
Pθ−1 are suitable symmetric positive definite matrices (the reason why we consider an
inverse matrix to define this latter norm will be clarified in the following). This proce-
dure goes under the name of Levenberg-Marquardt-Tykhonov regularization; see, e.g., [283]
for a discussion on classical regularization methods for inverse problems; note that usual
penalization coefficients are embedded in the definition of the matrix Pθ−1 .
The minimization problem (128) can be solved by an optimization algorithm based
e.g. on the evaluation of the gradient of J with respect to θ; as shown in Sect. 9.1, this
latter can be computed by relying on the solution of a suitable adjoint problem. This
is a four-dimensional variational (4D-Var) assimilation (a three-dimensional variational
(3D-Var) assimilation would arise in the case of steady state systems). See Sect. 10.3.1
for further details on a relevant example in cardiovascular modeling and, e.g., [61, 89] for
more on 4D-Var assimilation problems.

Since measurements Z are only available at a discrete number of time instants τ 1 , . . . , τ K ,

we formulate the identification problem by replacing the dynamical system (127) with its
discretized-in-time version:
½ k+1
X = Ak|k+1 (X k , θ), k = 0, . . . , K − 1
0 (129)
X =G

where Ak|k+1 is a nonlinear function describing the evolution of the state from time τ k to
time τ k+1 and X k ≈ X(k∆τ ) denotes the state vector at time τ k . Note that the length
∆τ = τ k+1 − τ k of the time windows between two subsequent measurements is usually
larger than the time step ∆t used for the sake of time discretization, and that θ does
not depend on k. We notice that hereon k will denote the temporal index of the system
evolution, thus using a different notation from the one introduced in Parts 1 and 2 (the
temporal index was denoted by n therein).
We then formulate an optimal discrete time minimization criterion and, finally, de-
termine the corresponding adjoint problem, rather than discretizing in time the adjoint
problem. This yields the minimization problem
K
1X 1
JK (X, θ) = kZ k − HX k k2Mk + kθ − θ 0 k2P −1 → min (130)
2 2 θ θ∈P
k=1

125
where X = (X 1 , . . . , X k ) and we set
Z k = Hk X k + εk . (131)
Here εk denotes the noise of the measurement device at τ k = k∆τ ; a possible choice for Mk
is Mk = ∆τ M , whereas Hk ≈ H(k∆τ ). Also in this case a gradient-based optimization
procedure can be used to solve the constrained optimization problem (129)–(130) with
gradients evaluated by introducing a suitable adjoint problem.

10.2 Sequential approach: Kalman filter and extensions

A drawback of the variational approach is the need to wait until the whole set of measure-
ments has been acquired in order to perform an optimization step. A sequential approach,
instead, performs the assimilation of acquired measurements on the fly and updates the
estimate of the unknown quantities accordingly.
A numerical milestone for the solution of sequential estimation problems, the Kalman
Filter (KF) [284], has been introduced as a recursive filter for the estimation of the state of
a noisy dynamical system from a set of measurements, that is, to improve the prediction
of the state dynamics by taking into account additional data; an augmented form of the
KF can be easily adapted, as we will see, also to the problem of estimating unknown
parameters. Originally designed for linear dynamical systems, the KF has severe memory
requirements. To mitigate these two limitations, several improvements and extensions
have been proposed in the past decades, most notably the Extended KF (EKF), the
Unscented KF (UKF) and the Ensemble KF (EnKF); detailed reviews can be found, e.g.,
in [283, 520, 270]. In this section we recall the formulation of the basic KF and provide
some hints about its extensions, with special focus on the field of cardiovascular modeling
where these techniques have been applied.
The literature offers many possible derivations of the KF; following [270, 89], here we
exploit the analogy with the solution of a recursive least-squares problem yielding the best
linear unbiased estimator for a linear model; alternative derivations can be obtained, e.g.,
relying on the so-called push-forward and subsequent conditioning of Gaussian measures
[531] or a sequential Bayesian estimation framework [321].

10.2.1 The Kalman filter algorithm

The KF algorithm sequentially generates an estimate of the unknown quantity through
a linear combination of the current estimate and the acquired measurement. Let us first
consider the case where there is no dynamics and data are generated by the linear model

Z = HX + ε (132)

where H is a given nz × nx matrix of rank nx , ε is an nz -dimensional random variable

with zero mean and known positive-definite covariance Q = E[εεT ] > 0 and Z represents
known, but inexact, measurements with errors given by ε. The vector X ∈ Rnx is the
quantity to be estimated from the observation Z; E[·] denotes the expected value.

126
 c = KZ for some
Among all linear estimators of X, that is estimators of the form X
n ×n
matrix K ∈ R x z , which are unbiased (i.e. E[X]c = X), the best choice is the one which
c
minimizes the mean-squared error E[(X − X) (X T c − X)]; by the Gauss-Markov theorem
(see, e.g., [531, Chap. 7]), the best or minimum variance linear unbiased estimator for
(132) is given by
c = (H T Q−1 H)−1 H T Q−1 Z.
X (133)
c − X)(X
In that case, E[(X c − X)T ] = (H T Q−1 H)−1 . Equivalently, (133) can also be
obtained by solving the weighted least-squares problem

c = 1 kHX − Zk2 −1 → min

X (134)
Q
2 X

again resorting to a variational argument. In the slightly different case where we want to
combine the observation and a background estimate X 0 of X with covariance P − , (134)
becomes
c = 1 kHX − Zk2 −1 + 1 kX − X 0 k2 − −1 → min
X Q (P )
2 2 X
and, instead of (133) we find
c = X 0 + K(Z − HX 0 ),
X K = P + H T Q−1 , (135)

upon defining the matrix playing the role of updated covariance as

P + = ((P − )−1 + H T Q−1 H)−1 . (136)

Note that this estimate is given by a linear combination of the background estimate X 0
and the so-called innovation Z − HX 0 . K is usually referred to as Kalman gain matrix;
it can also be evaluated exploiting the prior covariance P − instead than the updated
covariance P + , according to the equivalence

P + H T Q−1 = P − H T (HP − H T + Q)−1 . (137)

We consider now the case of a discrete-time linear system, with data acquired over a
time interval. Then, the model (132) is replaced by

X k = Ak−1|k X k−1 + v k , k = 1, . . . , K, X 0 = G,
(138)
Z k = Hk X k + w k

where X k ∈ Rnx denotes the state and Z k ∈ Rnz are the measurements; v k and wk
are uncorrelated zero-mean random noise processes with positive-definite covariances Qk
and Rk , modeling the uncertainty of the model and the additive noise in the observation,
respectively. The state estimation problem is the problem to find the state X k given k
known observations Z 1 , . . . , Z k .

The Kalman filter is a recursive algorithm that provides the best linear unbiased esti-
mate X ka of X k in terms of both the previous estimate X k−1a and the latest data Z k up
to that point in time. It is based on a predictor-corrector strategy, made of:

127
 1. a prediction step (called forecast, or time update) consists in letting the system
dynamics evolve from X k−1
a without taking into account the observations, yielding
k
the forecast state X f ;

2. a correction step (called analysis, or measurement update) updates the forecast state
X kf by assimilating the measurements into the model, yielding the assimilated state
X ka .

To derive the expression of the correction step, let us suppose that the current prediction
based on observations Z 1 , . . . , Z k−1 is X kf , with covariance matrix Pkf . If the true state
is X k , model (132) becomes · k¸ · ¸
Xf I
k = H Xk + ε
Z k

being ε an (nx + nz )-dimensional random variable with zero mean and covariance Qk =
diag(Pkf , Rk ). The best linear unbiased estimator of this system (see (133)) is given by
· ¸ · k¸ ³ ´
£ ¤ (P f )−1 0 Xf f −1 k T −1 k
X ka = Pka I HkT k
−1
a
k = Pk (Pk ) X f + Hk Rk Z (139)
0 Rk Z

and results from the linear combination of the current estimate and the last observation
Z k , where
µ · ¸ · ¸¶−1
£ ¤ (P f )−1 0 I
Pka = I HkT k
−1 = ((Pkf )−1 + HkT Rk−1 Hk )−1 ;
0 Rk Hk

note the formal analogy with (136), where now Pkf and Pka play the role of P − and P + ,
respectively. (139) can be written in the more convenient form

X ka = Pka [(Pka )−1 − HkT Rk−1 Hk ))X kf + HkT Rk−1 Z k ] = X kf + Pka HkT Rk−1 (Z k − Hk X kf )
= X kf + Kk (Z k − Hk X kf )

as a function of the innovation Z k − Hk X kf . Similarly to (137), the following relation

holds
Kk = Pka HkT Rk−1 = Pkf HkT (Hk Pkf H T + Rk )−1 , (140)

so that it is possible to evaluate the Kalman gain matrix Kk as a function of Pkf , and then
correct the covariance,

Pka = ((Pkf )−1 + HkT Rk−1 Hk )−1 = (I − Kk Hk )Pkf (I − Kk Hk )T + Kk Rk KkT = (I − Kk Hk )Pkf .

The prediction step instead exploits the dynamical system to propagate the state,
yielding
X k+1
f = Ak|k+1 X ka

128
for the time update of the state, and
f
Pk+1 = E[(X k+1
f − X k+1 )(X k+1
f − X k+1 )T ]
ck − Ak|k+1 X k − wk )T ]
= E[(Ak|k+1 X ka − Ak|k+1 X k − wk )(Ak|k+1 X a
= Ak|k+1 Pka ATk|k+1 + Qk+1

for the time update of the covariance. Grouping together the prediction and the correction
steps, we finally obtain the k-th step of the KF algorithm:

X kf =Ak−1|k X k−1
a state prediction (141a)
Pkf a
=Ak−1|k Pk−1 ATk−1|k + Qk error covariance prediction (141b)
Kk =Pfk HkT (Hk Pkf HkT + Rk )−1 Kalman gain evaluation (141c)
X ka =X kf + Kk (Z k − Hk X kf ) state correction (141d)
Pka =(I − Kk Hk )Pkf error covariance correction. (141e)

Note that from (141d) only the estimated state from the previous step and the current
measurement are needed to compute the estimate of the current state. The two pre-
diction and correction steps alternate: the prediction advances the state until the next
measurement is acquired, then the correction incorporates this latter.
X Z
Remark 7. The Kalman gain Kk defined in (140) can also be expressed as Kk = Pk f (PkZ )−1 ;
X Z
here Pk f = E[X kf (Z k − Hk X kf )T ] = Pkf HkT is the the cross-covariance between X kf
and the innovation Z k − Hk X kf , whereas PkZ = E[(Z k − Hk X kf )(Z k − Hk X kf )T ] =
Hk Pkf Hk + Rk is the innovation covariance. Similarly, (141e) can be rewritten as Pka =
Pkf − Kk PkZ KkT . This interpretation is useful when dealing with the unscented Kalman
filter (see Sect. 10.2.3).

Remark 8. In the linear case, the variational and the sequential approaches yield the same
result at the end of a time window, provided the following assumptions are made: the same
background estimation and the same covariance matrices are used, the same measurements
are acquired – that is, both algorithms are optimal in a least-squares or minimum variance
standpoint.

Let us now return to our problem of estimating the parameter θ. For that, we apply
the KF algorithm to the system:

X k = Ak−1|k X k−1 + Bk θ k + v k , k = 1, . . . , K, X 0 = G,
θ k = θ k−1

with observations
Z k = Hk X k + w k , k = 1, . . . , K;
here θ k ∈ Rp denotes the parameter vector and (under the linearity assumption), Bk ∈
Rnx ×p and we assume that no random error is associated with model parameters. This

129
is the so-called state augmentation technique. In order to exploit the KF algorithm,
fk = (X k , θ k )T instead of X k , thus yielding the following
we consider as state vector X
augmented KF algorithm:

fk =A
X fk−1
ek−1|k X state prediction (142a)
f a
e f e e a eT e
Pk =Ak−1|k Pk−1 Ak−1|k + Qk error covariance prediction (142b)
Kk =Pefk H e k Pef H T + R
e T (H ek )−1 Kalman gain evaluation (142c)
k k k
fk =X
X fk + Kk (Z ek − H fk )
ek X state correction (142d)
a f f
f
Pek =(I − Kk H
a e k )Pe
k error covariance correction (142e)

where
· ¸ · k¸ · ¸ · ¸ · ¸
ek−1|k = Ak−1|k Bk ,
A ek = Z ,
Z e k = Hk 0 ,
H e k = Qk ,
Q e k = Rk 0 .
R
0 I 0 0 0 0 0 0

We point out that, by construction of the filtering procedure, the estimated parameter
values evolve along the simulation period and the actual estimation is achieved with the
final values, that is, the estimated parameter vector is θb = θ K . Hence, we expect these
a
estimation trajectories to become less and less fluctuating more during the simulation (see,
e.g., Fig. 32 in Sect. 11.3.2); non-converging case would therefore denote the presence of
persistent modeling errors.
We close this section by pointing out that when a sequential approach like the Kalman
filter is used for the sake of parameter estimation, the dynamical system has to be solved
only once, by updating the parameter value after each assimilation of new measurements.
On the other hand, a variational approach would require to solve the dynamical system
on the whole time interval several times, assuming that an iterative approach is used for
performing the optimization; see the sketch in Fig. 28.

Figure 28: Variational approach (left) versus KF (right) approach: in the former, at each
optimization stage the whole state dynamics has to be computed, whereas in the latter
each measurement is sequentially used for the state (and parameter) correction

130
10.2.2 Extended Kalman Filter
The classical KF formulation is well-suited for low-dimensional, linear dynamical systems,
although in real applications these assumptions are seldom verified. The extended Kalman
filter (EKF) has been introduced for nonlinear dynamical systems, where (138) is replaced
by
X k = f (X k−1 , τ k ) + εk , Z k = h(X k , τ k ) + η k , (143)
being f and h two nonlinear functions. Nonlinearity here involves both the system dy-
namics and the observation model. The EKF consists in applying the KF algorithm
(141a)–(141e) to a linearized version of (143) around the previous state, so that at each
step we set ¯ ¯
∂f ¯¯ ∂h ¯¯
Ak−1|k = , Hk = ¯ k.
∂X ¯Xk−1
a ,θk ∂X X f

At each step two Jacobian matrices have to be evaluated at the current predicted state/pa-
rameters. A similar extension of the algorithm (142a)–(142e) provides the EKF for pa-
rameter estimation. Although feasible in principle, EKF suffers from several drawbacks:
for instance, it entails prohibitive computational costs to invert large matrices and to
propagate the covariance matrix in time. Even more importantly, the EKF may lack of
stability, meaning that as the estimated state deviates from the true state, the linearized
model becomes inaccurate, which may lead to an even larger error in state estimation.
To mitigate these shortcomings, several strategies have been put in place: low rank ap-
proximation of the covariance matrices have been considered, and other extensions of the
original Kalman filter like UKF and EnKF have been introduced.

10.2.3 Unscented Kalman filter

While the EKF exploits the differentiation of nonlinear operators modeling both the state
dynamics and the observation process to evaluate the propagation of means and covari-
ances, the unscented Kalman Filter (UKF), introduced in [281], relies on a set of well-
chosen deterministic points (or sigma points) whose propagation through the nonlinear
operators yields the empirical means and covariances required in the Kalman predic-
tion/correction formulas. The rationale behind goes under the name of unscented transfor-
mation, whose goal is to map a set of points so that their sample distribution approximates
the true distribution [280].
Supposing that we know the mean E[X] and the covariance Σ of a random vector
X ∈ Rnx , the simplest choice is to select 2nx (symmetric) sigma points x(i) as
³p ´ ³p ´
x(i) = E[X] + nx Σ , x(nx +i) = E[X] − nx Σ , i = 1, . . . , nx
i i
¡√ ¢
being nx Σ i the i − th column of the Cholesky factor of nx Σ; different options for the
sigma points selection are however possible, see, e.g., [279].
The prediction-correction strategy of the KF is then performed. Referring to the
same notation used in Sect. 10.2.1 and supposing that the current forecast based on
ck , with covariance matrix P f , the following correction step is performed:
Z 1 , . . . , Z k−1 is X
f k

131
 • from the estimated mean Xck and covariance P f at time τ k , select a set of 2nx sigma
f k
points
µq ¶ µq ¶
ck +
xkf,(i) = X n P
x k
f
, xk
= c
X
k
− n P
x k
f
, i = 1, . . . , nx
f f,(nx +i) f
i i

centered around ck
X at a distance given by the standard deviation extracted from
f
the covariance matrix and obtain the predicted measurement
2nx
X
bk = 1
Z bk ,
Z b k = h(xk , τ k ), i = 1, . . . , 2nx ;
being Z
(i) (i) f,(i)
2nx
i=1

• estimate the covariance of the predicted measurement

2nx ³ ´³ k ´T
1 X bk − Zbk Zb −Z b k + Rk
PkZ = Z (i) (i)
2nx
i=1

bk
and the cross-covariance between X kf and Z
2nx ³ ´³ k ´T
Xf Z 1 X ck − xk bk
b −Z
Pk = X f f,(i) Z (i) ;
2nx
i=1

• perform the Kalman gain evaluation, state correction and the error covariance cor-
rection similarly to (142c)–(142e) (recall Remark 7),
X Z
Kk = Pk f (PkZ )−1
ck = X
X ck + Kk (Z k − Z b k)
a f
Pka = Pkf − Kk PkZ KkT .

In the prediction step, a set of sigma points is selected as

¡p ¢ ¡p ¢
ck +
xka,(i) = X ck −
nx Pka i , xka,(nx +i) = X nx Pka i , i = 1, . . . , nx
a a

and a forward computation of one time window is performed to get the state prediction
at step k + 1
2nx
X
ck+1 = 1 xk+1 being xk+1 k k+1
X f
2nx f,(i) , f,(i) = f (xa,(i) , τ )
i=1
(note that each sigma point is treated independently); finally, the covariance is updated
as 2nx ³
1 X ´ ³ k+1 ´T
f
Pk+1 = ck+1 − xk
X c
X − xk
+ Qk+1 .
f f,(i+1) f f,(i+1)
2nx
i=1
Provided the analysis step is performed in parallel, the cost of the whole estimation pro-
cedure is comparable to that of a simple forward simulation; by contrast, a variational
estimation would require a significant number of successive iterations (typically, order of
hundreds) of both forward and adjoint simulations.

132
 An augmented formulation of the UKF can be easily obtained, similarly to (142), in
order to treat the case where the system depends on a set of p parameters affected by
uncertainty, which need to be estimated; nevertheless, the very large dimension of the
state vector (depending on the spatial discretization of the state variable) makes this filter
intractable in practice. In those cases where uncertainty only affects the parameters, and
p ≪ nx , a much more feasible version yielding a reduced-order UKF limits the computa-
tions of the filter operator to a subspace of small dimension and is much more efficient.
The reduction to the parametric space as regarding the choice of the sigma-points for the
sake of parameter estimation was originally introduced in [452, 261]; a first application to
the estimation of electrophysiology parameters can be found instead in [588] and in [587].
A detailed analysis of the reduced-order UKF can be found, e.g., in [377]. A possible
alternative, recently explored in [426, 425], is to perform instead a state reduction relying
on a reduced-order model, and then consider the problem of simultaneous state-parameter
estimation.

10.3 Examples of applications in cardiovascular modeling

In the past decade, parameter identification problems have been considered in several ap-
plications to cardiovascular modeling. In this section we report a list of relevant contribu-
tions; two meaningful examples will be more specifically discussed in Sects. 10.3.1–10.3.2.
The problem of identifying a set of parameters of one-dimensional models for the arte-
rial circulation has been first considered in [366], where a nonlinear least squares approach
based on the optimization of a cost function and the introduction of a suitable adjoint
problem has been considered. A relevant application in this context is the estimation of
the elastic coefficient of a subject-specific vessel based on measurements of its displace-
ment recovered from medical images. A variational approach based on the minimization of
suitable functionals has been proposed in [442] and, more recently, in [50]; we present this
technique in Sect. 10.3.1. A different approach based on an unscented Kalman filter has
been proposed in [51] and further explored in [376] for estimating modeling parameters
pertaining to vessel wall boundary conditions. Similar strategies based on the unscented
Kalman filter have been considered e.g. in [340] to solve inverse problems in large one-
dymensional arterial networks, and in [430] for the sake of estimation of lumped (e.g.,
windkessel) model parameters by using the pressure curve in ascending aorta.
Other approaches have been considered e.g. in [59] to characterize the terminal periph-
eral resistances in a network describing one-dimensional arterial blood flow, or in [522] to
adjust the parameters of a windkessel outflow boundary conditions of three-dimensional
blood flow models, in order to match some desired features of pressure and flow waveforms.

Several works have been recently focused on parameter identification and, more gener-
ally speaking, on the solution of inverse problems in heart modeling. The classical inverse
problem in electrocardiology has been considered by many authors in the last decades; it
consists in recovering the electrical potential at the epicardial surface by using a number
of remote, non-invasive or minimally-invasive potential recordings, e.g. acquired along the
body surface [500, 464, 96] which are usually referred to as body surface-potential data.

133
In its original formulation, this problem involves the pure (linear) diffusion model for the
torso as direct problem [122, 116, 604], see Sect. 7.1.5. Such inverse problem is mathe-
matically ill-posed, and several regularization strategies have been proposed; see, e.g., the
reviews [465, 118]. Another method to solve the inverse electrocardiology problem relies
on measurements of the electrical potential within one of the heart chambers (ventricles
and atria) by means of non-contact multi electrode catether arrays, and tries to estimate
the endocardial surface potential from them. Since the measurements are recorded closer
to the endocardial surface than the ones acquired on the body surface, this problem is less
ill-conditioned than the former. If the problem is recast under the form of a parameter
estimation problem, i.e. where the epicardial potential distribution is described in terms
of a set of parameters, the inverse problem can be more easily tackled.

More generally speaking, inverse and parameter estimation problems have been con-
sidered in several works in order to reconstruct, e.g., the cardiac electrical activity in
the myocardium, or to locate ischemic or infarcted zones by estimating conductibility pa-
rameters. Concerning variational approaches, we can mention a level set framework for
identifying heart infarctions [346] by relying on a least-squares formulation and an adjoint
problem to determine the gradient of the cost functional. In this case, the infarcted region
in a simplified two-dimensional domain has been described in terms of a discrete level-set
function, involving a set of parameters to be identified from synthetic ECG boundary
measurements, and a Tikhonov regularization procedure. The solution of a parameter
identification problem to locate (in terms of size and position) ischemic regions, where a
simplified state elliptic system modeling the electrical potential in both the heart and the
torso in the resting phase is considered also in [347, 406]. In all these papers, the PDE-
constrained optimization problem has been solved using an iterative method, following an
optimize, then discretize approach. An all-at-once approach has been instead considered
in [407] for a similar problem, taking also into account anisotropic cardiac conductivities
and fiber orientation. A more involved version of this problem, still focusing on the inverse
electrocardiographic source localization of ischemias, has been more recently considered
in [585], taking into account physically-based constraints (in both equality and inequality
forms), a total variation regularization and a primal-dual interior point method.
The personalization of a cardiac electrophysiology model involving e.g. the identi-
fication of conductivities, local conduction velocities and parameters related to action
potential duration restitution curve has been addressed in [100, 489] – although relying
on a simplified Eikonal model – and applied to a clinical dataset derived from a hybrid
X-ray/magnetic resonance imaging and non-contact mapping procedure on patients with
heart failure. More recently, a variational approach for the estimation of cardiac conductiv-
ities entering in a bidomain model, from measures of the transmembrane and extracellular
potentials available at some sites of the tissue, has been proposed in [606]. It relies on a
derivative-based optimization method where the gradient of the functional to minimize is
computed by resorting to the adjoint equations of the bidomain model. Parameter esti-
mation in a problem where the bidomain model for the heart is coupled with the Laplace
equation for the passive conduction in the torso has been considered in [74] for the sake
of estimating the torso conductivity parameters.

134
 The unscented Kalman filter has been exploited in [587] and more recently in [360, 538]
for the identification of scar locations and their size from body surface-potential and MRI
data, using a monodomain two-variables Allev-Panfilov model. A reduced-order UKF has
been recently exploited in [123] for the estimation of electrical parameters within an elec-
tromechanical model, by taking advantage of observations from both electrocardiograms
and myocardium displacements.
Another inverse problem in electro-physiology has been studied in [567, 428]. The mea-
sures of the electrical activation time on the endocardium, acquired by the NavX system
(see Sect. 6.2), were used to find the optimal Purkinje network configuration. For this
problem, the parameters to be estimated are the coordinates of the network. In partic-
ular, starting from an initial network with fractal shape, a functional accounting for the
discrepancy between measured and computed activation times is minimized in order to
find the patient specific location of the PMJ (see Sect. 5.3). This methodology has been
successfully applied to real pathological scenarios, see [427].

Regarding instead cardiac biomechanics, the estimation of contractility parameters has

been addressed in [379, 377] using the unscented Kalman filter, where the degree of damage
in cardiac tissues caused by an infarct is estimated using velocity measurements coming e.g.
from tagged MRI; see also [378]. A reduced-order UKF has been applied in [597] for the
sake of identification of material parameters in a transversally isotropic mechanical model;
further details concerning this applications will be provided in Sect. 10.3.2. Later on, the
same technique has been exploited in [87] for estimating contractility values in a more
complex cardiac model, where the tissue is described by combining an active constitutive
law in the muscle fiber direction and a visco-hyperelastic material, using actual clinical
data consisting of in-vivo Cine-MR image sequences and pressure measurements.
Finally, concerning the coupled electromechanical problem, very few numerical results
are nowadays available. The personalization of a 3D electromechanical model has been
tackled by variational approaches e.g. in [516], where local ventricular myocardium con-
tractility has been estimated using MRI in an electromechanical model. Similarly, in [135]
both the parameters of a Mitchell-Schaeffer model and cardiac contractilities are esti-
mated from catheterized electrophysiology data and cine-MRI images, respectively. The
calibration of mechanical parameters of a complete electro-mechanical model of the heart
involving the Eikonal model for electrophysiology and an isotropic Mooney-Rivlin material
for cardiac mechanics has been performed e.g. by relying on the unscented Kalman filter
in [359].

Although several works have made a big step ahead, the solution of parameter esti-
mation problems remains an open computational challenge if complex, coupled models
and patient-specific data are taken into account. Also in the case where UKF and EnKF
techniques are exploited – which features natural parallelism to a high degree – the need to
evaluate the state dynamics for several different scenarios (given e.g. by the elements of the
ensemble) makes the computational effort exorbitant. For this reason, simplified physical
models have been considered in many contexts so far; on the other hand, reduced-order
strategies, like the reduced-order unscented Kalman filter, have been recently proposed

135
as a possible way to overcome the computational complexity arising from these problems.
Dramatic progress is expected in the following years in this field.

10.3.1 A variational approach for estimating the tissue compliance

In this section we illustrate a variational approach introduced in [442] for estimating the
Young modulus E of a cardiovascular tissue from displacement data. Indeed, this param-
eter – and, more generally speaking, the deformability of a soft tissue – is an important
index for detecting diseases: low compliance may indicate atherosclerosis or hypertension
in the case of an artery, or a marker of diastolic dysfunction in the case of the left ventricle
wall. The basic steps consist in (i) retrieving the vessel displacement dmeas by image
registration procedures on time frames of the vessel of interest, and then (ii) minimiz-
ing the difference between dmeas and the displacement computed by solving the coupled
3D blood-vessel problem, in order to estimate the Young modulus. Here the structure
is assumed to be linearly elastic and the only parameter to be estimated is E, which is
in general a function of space (e.g. an atherosclerotic plaque has different modulus than
healthy tissue) but not of time.
The former step consists of data acquisition, image segmentation and reconstruction,
and finally registration, see [442] for further details. The latter, which we focus on, is
an example of an inverse fluid-structure interaction (IFSI) problem, and results in a con-
strained optimization problem, where the constraint is the FSI problem, and the functional
to be minimized is a measure of the mismatch between the data and the computed solution.

A first formulation of the IFSI problem is obtained by assuming that the displacement
data dmeas (x, τk ) for x ∈ Στk retrieved from the image registration process are available
within the interval [0, T ] in some instants denoted by τ k , k = 1, . . . , K. K is the number
of instants when image registration is performed, ∆τ the time step between two measure-
ments, driven by the sampling frequency of the image devices, Στk denotes the interface
between the fluid and the structure domain. Following the formulation of problem (130),
we consider the functional
N Z ³ ´2 N Z ³ ´2
1X k k αE X
JK (d, E) = dmeas (x, τ ) − d(x, τ ) dσ + E(x, τ k ) − Eref dx,
2 2
k=1 Στk k=1 Ω
s
(144)
where d(x, τ k ) denotes the solution of the FSI system (47) in the ALE formulation, at
t = τ k and the second term is a non-negative Tikonov regularization term. Here Eref
is a prior estimate (available e.g. from ex vivo specimens), so that the regularization
forces E to be close to its reference value. Given an admissible set Ead where we seek the
parameter E, a possible formulation of the IFSI problem reads: for t > 0, x ∈ Ωs , find
E = E(x, t) ∈ P that minimizes (144) under the constraint (47). A possible choice for P
is
P = {E : E ∈ L∞ (Ωs ), 0 < Emin ≤ E ≤ Emax , with Emin , Emax ∈ R}. (145)
Such a problem entails the solution of a time-dependent minimization problem, for which
a classical KKT system can be obtained using the Lagrange multipliers method. In this

136
setting the parameter E plays the role of control variable. However, this approach is rather
involved, since the adjoint problem results in a final value problem which would involve
differentiation with respect to the deformable domain, too – the so-called shape derivative.
Moreover, it requires a massive memory occupancy as the solution at all the time steps
needs to be stored due to the back-in-time nature of the adjoint problem.

A more convenient alternative numerical approach proposed in [442] considers first

the time-discretization of the forward problem, and formulates a minimization problem at
each time step. For the sake of simplicity a constant time step ∆t = ∆τ s is assumed, for
a suitable s ∈ N, s ≥ 1, that is, the instants τ k in which measurements are acquired are
a subset of the time discretization of (47); for the sake of simplicity, hereon we consider
s = 1. For the time discretization of (47) implicit methods with a semi-implicit treatment
of the convective term and of the fluid domain are considered, whereas the fluid viscous
term is treated implicitly. Exploiting the notation already introduced in Sect. 4.6.1; we
denote by Ω⋆f , u⋆ and w⋆ appropriate extrapolations of the fluid domain, fluid velocity
and fluid domain velocity, respectively, and introduce the following spaces:

V ⋆ = {v ∈ H 1 (Ω⋆f ) : v|Γ⋆D,f = 0}, Q⋆ = L2 (Ω⋆f ), W = {ψ ∈ H 1 (Ωs ) : ψ ΓD,s = 0},

where Γ⋆D,f and ΓD,s are the portions of the boundary where a Dirichlet condition is
b Σ⋆ /∆t = 0} and
prescribed. We denote moreover Z ⋆ = {(v, ψ) ∈ V ⋆ × W : v|Σ⋆ − ψ|
introduce the following bilinear forms:
µ ¶
ρf ⋆ ⋆ ⋆ ⋆ ⋆ d ψ
a(v, d; w, ψ)⋆ = (v, w)f + (T f (v, p), ∇w)f + ρf (((v − v f ) · ∇)v, w)f + ρs , ,
∆t ∆t2 ∆t s
b(q; w)⋆ = −(q, ∇ · w)⋆f ,
R R
where (v, w)⋆f = Ω⋆ v · w dx and (ψ, χ)s = Ωs ψ · χ dx. Then, for any given E ∈ Ead , at
f
each time tn+1 , the time discrete forward FSI problem consists of the following steps:

1. compute extrapolations Ω⋆f , v ⋆ and v ⋆f for approximating Ωn+1

f , v n+1 and v n+1
f ;

2. given f n+1
f ∈ L2 (Ω⋆f ) and f n+1
s ∈ L2 (Ωs ), find (v n+1 , dn+1 ) ∈ Z ⋆ and pn+1 ∈ Q⋆
such that
¡ ¢
a(v n+1 , dn+1 ; w, ψ)⋆ + E S s (dn+1 ),
³ ´
1
∆t ∇ψ s
+ b(pn+1 ; v)⋆
= Ffn+1 (w) + Fsn+1 ψ
∆t − a(Rn+1
f , 0; w, ψ)⋆ ∀(w, ψ) ∈ Z ⋆
b(q; v n+1 )⋆ = 0 ∀q ∈ Q⋆ ;
(146)

3. update the fluid domain to obtain Ωn+1

f .

For the sake of notation, here we have set

1 ν
S s (d) = (∇d + (∇d)T ) + (∇ · d)I
2(1 + ν) (1 + ν)(1 − 2ν)

137
so that we can write T s = E S s ; moreover, given g ∈ H 1/2 (Σ⋆ ), we have denoted by
Rf (g) ∈ H div (Ω⋆f ) = {v ∈ H 1 (Ω⋆f ) : ∇ · v = 0} and Rs (g) ∈ H 1 (Ωs ) two lifting functions
for the fluid-structure interface continuity, defined as
Rf (g) − Rs (g) = −g/∆t on Σ⋆ .
Due to the arbitrariness of one of these functions, in what follows we set Rs (g) = 0;
moreover, when applied to function dm−1 , it is possible to set Rm f = Rf (d
m−1
). Finally,
n+1
the functionals Ff and Fsn+1 in (146) account for forcing terms, boundary data on
∂Ω⋆f \ Σ⋆ and ∂Ωs \ Σ and terms coming from the time discretization at previous time
steps.
Regarding the parameter estimation problem, once the problem has been discretized
in time, the minimization of the following cost functional
Z ³ ´2 Z ³ ´2
αE
Jek (d, E) = dmeas (x, τ k ) − d(x, τ k ) dσ + E(x, τ k ) − Eref dx
2 Ωs
Σ

for each k = 1, 2, . . . , K is considered in [442], under the constraint (146). Hence, a system
of KKT conditions can be derived relying on the standard Lagrange multiplier approach
(see Sect. 9.1.2), by introducing the Lagrangian functional at time τ k
µ ¶
e ⋆ 1
L(v, p, d; λv , λp , λd ; E) = Jk (d, E) + a(v, d; λv , λd ) + E S s (d), ∇λd
µ ∆t¶ s
⋆ ⋆ λd
+b(p; λv ) + b(λp ; v) − Ff (λv ) − Fs + a(Rf , 0; λv , λd )⋆ .
∆t
By imposing that the gradient of L vanishes, we obtain:
• the adjoint problem, by forcing to zero the (Gâteaux) derivatives of the Lagrangian
functional with respect to (v, p, d), which in fact is a discretized-in-time FSI problem
(see below for the interface condition): find (λu , λd ) ∈ Z ⋆ , λp ∈ Q⋆ such that
µ ¶
ψ
a(v, ψ; λu , λd ) + E S s ( ), ∇λd + b(λp ; v)⋆
⋆
∆t Z s
+ (d − dmeas ) · ψ dσ = 0
Σ
b(q; λu )⋆ = 0 ∀(v, ψ) ∈ Z ⋆ , q ∈ Q⋆ ;
(147)
• the optimality condition
(ϕ S s (d), ∇λη )s = 0 ∀ϕ ∈ L∞ (Ωs ),
by forcing to zero the derivative with respect to E.
These two problems, together with the state problem: find (u, d) ∈ Z ⋆ , p ∈ Q∗ such that
µ ¶ µ ¶
1 ⋆ ψ
⋆
a(u, d; v, ψ) + E S s (d), ∇ψ + b(p; v) = Ff (v) + Fs − a(Rf , 0; v, ψ)⋆
∆t s ∆t
b(q; u)⋆ = 0 ∀(v, ψ) ∈ Z ⋆ , q ∈ Q⋆ ,

138
formally obtained by deriving the Lagrangian with respect to (λv , λp , λd ), yield at each
τ k the system of KKT conditions, which couples two linearized FSI problems and a scalar
equation.
In particular, for the adjoint problem the interface velocity condition reads
λd
λu = on Σ⋆ ,
∆t
whereas the interface stress condition is

T s (λd ) n − T f (λu , λp ) n = − (d − dmeas ) on Σ⋆ , (148)

taking into account the mismatch between the data and the solution, and modifying the
homogeneous interface stress condition (22d) accordingly.
The same strategies described in Sect. 9.2 (e.g., gradient-based methods) can be ex-
ploited to solve numerically the KKT system. To take into account the constraint E > 0
in the case at hand, it is possible to transform the parameter as ψ = log(E), so that
E = exp(ψ) > 0 for every ψ ∈ L∞ (Ωs ) and then optimize with respect to ψ. Finally,
a finite element discretization in space is required to solve numerically the state and the
adjoint problem.
By construction, this approach provides an estimate for E at each time; a possible
option is then to average them in order to obtain a unique estimate, although this may
suffer of instabilities in presence of highly noisy data. A more robust approach consists of
solving the minimization problem over time. In this case, however, the complexity of the
problem would increase due to the inclusion of the shape derivatives into the minimization
problem. In Fig. 29, we report the number of iterations and the convergence history for
the algorithm described above for the estimation of the Young modulus.

Figure 29: Convergence history for the estimation of the Young modulus by means of
the algorithm proposed in [442]. Rectangular fluid and structure domains are used, with
synthetic measures generated by means of forward FSI simulations. These numerical
results are obtained using the Matlab Finite Element library MLife

139
10.3.2 A Kalman filter approach for estimating material parameters in car-
diac mechanics
In this section we illustrate a Kalman filter approach introduced in [597] for estimating
material parameters entering in the mechanical model describing myocardial contractility.
Only passive mechanics is modeled making a quasi-static assumption, and without taking
into account the electro-mechanical coupling. A transversely isotropic Guccione’s consti-
tutive law [225] is used and four parameters are estimated in-silico from noisy displacement
measurements of material points located on the myocardial surface. A similar approach
can be in principle applied, e.g., also to identify the material parameters affecting the
orthotropic model (94).
The transversely isotropic strain energy function in the Guccione law can be expressed
as
1
Θ(E) = θ1 (eQ − 1)
2
where
Q = θ2 Ef2f + θ3 (Ess
2 2
+ Enn 2
+ 2Esn ) + θ4 (2Ef2s + 2Ef2n );
here θ = (θ1 , . . . , θ4 ) ∈ Rp denotes a vector of p = 4 material parameters, whereas Eij is
the (i, j)-th component of the Green-Lagrange strain tensor E = 12 (F T F − I) = 21 (C −
I), with i, j ∈ {f, s, n} denoting fiber, sheet and sheet-normal directions, respectively.
After performing spatial discretization, at each time step the augmented state X fk =
(X k , θ k )T ∈ Rnx +p made by the deformed configuration X k and the material parameter
vector θ k has to be determined by solving a nonlinear problem of the form
· k−1 k−1
¸
fk = f (X
X fk−1 , uk−1 ) + εk , fk−1 , uk−1 ) = G(θ , u ) ,
f (X
θ k−1

being the input vector uk−1 at time k − 1 given by the external forces, with observations
k
f ) + wk ,
Z k = h(X k = 1, . . . , K.

In [597] h : Rnx +p → Rnz is taken as a linear (interpolation) matrix H ∈ R(nx +p)×nz

mapping the augmented state vector to the coordinates of the points where measured are
acquired; those measurements are assumed to be corrupted by a noise term wk with zero
mean and covariance Rk ; note that the material parameter has no time dependence itself.
Nevertheless, as already pointed out, the filter will provide recursively updated estimates
of θ k ; the goal is thus to compute the estimate θ K at the final time t = tK .
Moreover, since the augmented state vector X fk does not depend on the deformed
configuration Xfk−1 , the rank of the error covariance matrix Pek at each step will be equal
to p; this is a crucial point in order to devise a reduced-order UKF, making possible to
(i) store a covariance matrix of dimension (nx + p) × p, (ii) perform inversion of p × p
matrices and, even more importantly, (iii) decrease from O(nx + p) to O(p) the number
of model evaluations required by the filter at each time step.

140
 Slight differences arise with respect to the UKF formulation addressed in Sect. 10.2.3,
fk corresponding to the p
due to the reduction of the estimation process to the subset of X
parameter components. The filtering algorithm consists of three steps:

• inizialization: a singular value decomposition P0a = La0 Λa0 (La0 )T of the initial error
covariance P0a ∈ R(nx +p)×(nx +p) is performed; for instance, P0a can be given by a
diagonal matrix with p non-zero entries representing the variances of the initial (or
background) parameter estimates. Denoting by La0,p the matrix whose columns are
the first p singular vectors of La0 and Λa0,p the p × p diagonal matrix of the singular
values, the reduced-rank square-root approximation S0a ∈ R(nx +p)×p can be obtained
as q
S0a = La0,p Λa0,p ;

• prediction (or time update): at each step k = 1, . . ., starting from the assimilated
k−1
c
f
state X a at step k − 1, a set of sigma points is selected as
k−1 ¡ a ¢ k−1 ¡ a ¢
c
f c
f
ek−1
xa,(i) = X a + Sk−1 i
, ek−1
xa,(p+i) = X a − Sk−1 i
, i = 1, . . . , p.

Then, the sigma points are transformed through the state dynamics to obtain
³ ´
ekf,(i) = f x
x ek−1
a,(i) , u k−1
, i = 1, . . . , 2p,

k
c
f and the error covariance P f are estimated as
the mean X f k

k X 2p 2p
X k k
c c c
f =
X f ekf,(i) ,
x Pkf = f −x
(X f
f −x
ekf,(i) )(X f ekf,(i) )T ;
i=1 i=1

to preserve the low-rank structure of the correlation matrix Pkf , a singular value
decomposition Pkf = Lfk Λf0 (Lfk )T is performed, thus yielding
q
Pkf = Skf (Skf )T , Skf = Lfk,p Λfk,p ;

• correction (or measurement update): at each step k = 1, . . ., the observation Z k is

k
ê f yielding the corrected state
assimilated into the forecast X
k k k
X f̂ ).
f̂ + Kk (Z k − H X
f̂ = X
a f f

In this case the Kalman gain can be obtained as

Kk = Skf (I + (HSkf )T Rk−1 HSkf )−1 (HSkf )T Rk−1

141
 along with the square-root of the error covariance, given by

Ska = Skf (I + (HSkf )T Rk−1 HSkf )−1/2 .

Note that here, compared to the UKF formulation derived in Sect. 10.2.3, update
formulas for the Kalman gain and the error covariance are directly obtained for the
square-root of the covariance matrix; see, e.g., [520, 597] for further details.

In [597] such a reduced-order UKF is shown to be able to estimate p = 4 (and, subse-

quently, p = 12) material parameters starting from a set of synthetic measurements gen-
erated by adding a Gaussian noise to a set of noise-free measurements obtained through
model simulation for a chosen, ground-truth, parameter vector. Myocardial dilation is
considered by loading the structure with a given ventricular pressure on the endocardium,
and nz = 600 material points located on the epicardium and endocardium.
We have chosen to report this example to illustrate how to take advantage of a KF
strategy in a relatively simple cardiovascular application; further details and more involved
cases can be found, e.g., in [377, 51, 87, 376, 123]. Numerical results related to parameter
identification in electrophysiology by means of an Ensemble Kalman Filter – a technique
which shares several similarities with the UKF, indeed – will be shown in Sect. 11.2.2.

11 Including uncertainty
Moving towards model personalization – that is, the adaptation of model inputs to subject-
specific conditions – the question on the sensitivity of model predictions to errors and
uncertainties in the model inputs dramatically arises. The inputs to be personalized may
include the computational domain (e.g., vascular networks), physical parameters (such
as vascular material properties), and boundary conditions. Because of noise in input
measurements, as well as of their large biological variability, model inputs are unavoidably
hampered by uncertainty. Furthermore, as already seen in Sect. 10, not all model inputs
are directly measurable, as in the case of the local mechanical properties of the arterial wall,
thus calling into play the need of solving parameter estimation problems. The uncertainties
carried by patient-specific features should then be incorporated into the computational
model, to quantify their impact on the computed results and to obtain more reliable
predictions or best/worst-case scenarios [151]. These are the main reasons behind the
very rapid growth of applications of sensitivity analysis and uncertainty quantification
(UQ) to cardiovascular problems in the last decade. Being UQ a very active (and less
mature than PDE-constrained optimization or parameter estimation) field from both a
mathematical and a computational viewpoint, in this section we only touch some basic,
yet relevant, issues related to UQ techniques; see, e.g., [531, 283, 323] for a more in-depth
overview of numerical aspects of UQ.
Typically, UQ problems involve a mathematical model for a process of interest, subject
to some uncertainty about the correct form of this model or, more frequently, about some

142
of its parameters. Although featuring substantial overlap with the field of parameter
estimation and data assimilation, UQ problems additionally involve the propagation of
uncertainty on output of interest, reliability or certification problems, prediction problems
and, very often, consist of several of these aspects coupled together. For instance, after
estimating some model parameters, we may use them to forward propagate some other
uncertainties. Often, though not always, uncertainties are treated probabilistically, thus
calling into play probabilistic and statistical methods. Here we focus on basic aspects
related with forward propagation and backward propagation of uncertainty, focusing for
the latter case on statistical inversion methods within a Bayesian framework.

11.1 Forward Uncertainty Quantification

The goal of forward UQ is to derive information about the uncertainties in system outputs
of interest, given information about the uncertainties in the system inputs. The goal is to
obtain the probability density function (PDF) of some output of interest Z = f (θ) ∈ Rnz
given the probability distribution of the input θ ∈ Rp , or evaluate moments (such as
the expected value and the variance), correlation functions, confidence regions, quantiles,
etc4 . Here f : Rp → Rnz denotes an input-output map; for instance, in [506] the ra-
dius of an abdominal aortic aneurysm, the radii and the inflow velocity of the carotid
artery bifurcation, and the flow split of the left and right pulmonary arteries are taken as
random variables to account for the uncertainty impact on blood flows modeled by three-
dimensional Navier–Stokes equations with rigid arterial walls, by considering as outputs
blood velocity and wall shear stresses.
Once a probabilistic description of the random inputs has been provided, a suitable
strategy to propagate uncertainties through the model is needed. Regarding the former
aspect, the simplest case is the one where the PDE system depends on a set of random
inputs that are constant with respect to space and time. In this case, they can be described
by means of a finite-dimensional random vector, with a given probability distribution. A
more involved case is the one where input data may vary randomly from one point of
the physical domain to another (and, possibly, from one time instant to another); in this
case, they are described in terms of random fields. Two popular strategies to describe
correlated random fields include Karhunen-Loève expansions and expansions in terms of
global orthogonal polynomials [210, 602]. Made by infinitely many terms, these expansions
are usually truncated to approximate random fields: the milder the variations in space
and time of the random field realizations, the fewer the terms retained in the expansion.
Several approaches to solve PDE problems with random input data are available.
Monte Carlo (MC) methods are the most popular; they are based on independent real-
izations θ 1 , . . . , θ M of the random variable θ (whose probability distribution is denoted
by µ) yielding approximations of the expectation by averaging over the corresponding
realizations of that quantity, that is,
4
Sensitivity analysis (see, e.g., [502] for a detailed review) may be seen as a precursor of forward UQ,
sharing the need of providing a quantitative description of the dependence of the solution of a model on
input parameters. See, e.g., [147] for the application of SA to a pulse wave propagation model of arterial
flows.

143
 Z M
1 X
E[f (θ)] = f (θ)µ(dθ) ≈ f (θ i ) (149)
Rp M
i=1

where the sample θ 1 , . . . , θ M is generated randomly from the probability distribution of

θ. Such a method requires a deterministic PDE query for each realization, and a √ very
large number of queries to achieve a small error, this latter being proportional to 1/ M .
This entails an excessive computational burden especially for systems which are already
computationally expensive in their deterministic setting. See, e.g., [181, 491, 578] for
more on MC methods, [141] and [212] for quasi Monte Carlo and multilevel Monte Carlo
methods, respectively.
When the solution depends analytically on the random input parameters, other ap-
proaches feature much faster convergence rates. These include spectral (global) stochastic
Galerkin (SG) methods [210, 601, 22] and stochastic collocation (SC) methods [21, 600,
412]; see, e.g., the detailed review [236] and the recent books [323, 531] for a mathematical
and numerical discussion. These methods are based on the discretization of a PDE sys-
tem not only with respect to spatial variables, but also with respect to the random inputs.
SG methods are intrusive approaches since the physical and the probabilistic degrees of
freedom are coupled; stochastic sampling (SS) and interpolatory-type SC methods are
instead non-intrusive. SG methods require the solution of discrete systems that couple all
spatial and probabilistic degrees of freedom, whereas SC methods can combine standard
approximations in physical space with globally defined polynomial approximation in the
probability domain, either by full polynomial spaces, tensor product polynomial spaces,
or sparse tensor product polynomials. The stochastic space can then be queried at any
point in order to construct the PDF of the output, the interpolation in the stochastic
space being built on top, and independently, of the PDE solver.

For the sake of illustration, we report in Fig. 30 the activation times obtained using
the monodomain model coupled with the Aliev-Panfilov cell model. This computation is
carried out on a patient-specific left ventricle geometry, obtained by considering different
levels of tissue damage (or ischemia), caused by an insufficient blood perfusion of the my-
ocardium. In this case, each component of the conductivity Σ is multiplied by a spatial
field σ(x; θ) ranging from 0 (lack of conductivity, damaged tissue) to 1 (regular conductiv-
ity, healthy tissue), which we refer to as relative conductivity; as soon as the position and
the size of the ischemia are unknown, σ is a random field, depending on (a finite number
p of) random inputs θ1 , . . . , θp . We highlight that in this case input uncertainty yields
significant variability in the output, thus making uncertainty propagation a problem of
interest in this context.

11.2 Inverse Uncertainty Quantification

Inverse UQ (or backward uncertainty propagation) problems refer to those situations in
which input quantities are inferred starting from observed system outputs. Variational or
sequential methods described in Sect. 10.1–10.2 provide point estimates of the quantities
of interest, given a set of observations. The statistical inversion approach provides a well-

144
Figure 30: Different degrees of tissue damage in terms of relative conductivity (top)
and activation times in milliseconds (bottom) for an healthy case (left) and different
ischemic regions on the myocardium. The patient-specific geometry of the left ventri-
cle has been reconstructed using the semi-automatic segmentation method proposed in
[170]. These numerical results are obtained using the Finite Element library redbKIT
v2.1 ([Link]/redbKIT/redbKIT/releases)

established framework to better characterize the uncertainty in the data, and the resulting
uncertainty in the computed estimates, adopting a Bayesian formulation. In this setting,
all the model inputs included in the model are described by random variables, where
randomness accounts for the available degree of information concerning their realizations,
and is expressed in terms of probability distributions; as a result, the solution of the
inverse UQ problem is the posterior PDF of the unknown inputs. Compared to variational
methods of Sect. 10.1, where classical regularization methods yield point estimates by
curing the ill-posedness of the problem, statistical inversion aims at removing ill-posedness
by recasting the inverse problem in a larger space of probability distributions [283]. This
strategy also allows to better characterize the prior information brought, e.g., by the
regularization terms in (128), under the form of a prior PDF of the unknown inputs. The
task of Bayesian inversion is to improve the knowledge on the unknown system features
starting from the prior belief and exploiting information from a set of model realizations.
For more on this topic, see also, e.g., [530, 531].

11.2.1 Static problems

Assume that we are measuring an output z = f (θ, ε) ∈ Rnz in order to get information
on the unknown input θ ∈ Rp , and that the measured quantity is polluted by noise ε . By
treating input and output as random quantities, let us denote by Z ∈ Rnz and Θ ∈ Rp two
random variables, of which z an θ represent the corresponding realizations. Very often,

145
the noise is modeled as additive and mutually independent of Θ, so that the input-output
map takes the form
Z = f (Θ) + ε
where Z, ε ∈ Rnz , Θ ∈ Rp are random variables. Here Z is called measurement, and its
realization Z = z in the actual measurement process the data; moreover, let us denote by
πnoise (ε) the PDF of the noise ε, usually encoding experimental errors. Before performing
output measurements, all the information (structure, regularity, etc.) about the distribu-
tion of the input Θ are encapsulated in the prior PDF πprior (θ), to be selected according
to problem-specific considerations.
The conditional probability π(z | θ) of Z | Θ = θ (that is, of Z conditioned on Θ = θ)
is the so-called (conditional) likelihood function, and expresses the likelihood of different
measurement outcomes z being Θ = θ given. In our case the input/output map f : Rp →
Rnz the solution of a PDE problem, and the evaluation of the output f (θ). Thanks to
the assumption of mutual independence of Θ and ε, Z | Θ = θ is distributed like ε, that
is, the likelihood function is

π(z | θ) = πnoise (z − f (θ)).

Assuming that the measurement data Z = z is given, in the Bayesian framework the
inverse problem is to find the conditional PDF π(θ | z) of Θ; this latter is the posterior
PDF of Θ given the data Z = z and can be expressed through the Bayes theorem as
πprior (θ)π(z | θ)
π(θ | z) = ;
π(z)
R
π(z) = Rnz π(z | θ)πprior (θ)dθ plays the role of a normalization constant, and has often
little importance from a computational standpoint.
Solving an inverse UQ problem in the static case – alternatively, in literature such a
problem is referred to as stationary inverse problem – thus consists in finding a prior PDF
πprior (θ), expressing the likelihood function π(z | θ) using the interplay between the obser-
vation and the unknown and finally developing suitable numerical techniques to explore
the posterior PDF. Each of these tasks is a challenging problem from a computational
standpoint; here we provide some hints on how to cast in this framework a wide range of
applications related with backward uncertainty quantification in cardiovascular modeling.
In the case where the unknown is a random variable with few components, the posterior
PDF can also be visualized under the form of a nonnegative function of these variables;
most applications however yield larger scale inverse UQ problems, and resulting PDFs
in high-dimensional spaces, for which it is much more effective to evaluate suitable point
estimators, like the maximum a posteriori estimator

θ M AP = arg maxp π(θ | z)

θ∈R

or the conditional mean

Z
θ CM = E[θ | z] = θ π(θ | z) dθ.
Rp

146
Evaluating the former requires the solution of an optimization problem, using e.g. it-
erative, gradient-based methods; computing the latter involves a numerical quadrature
problem in high-dimensional spaces. The evaluation of variability estimators like the con-
ditional covariance
Z
Cov(θ | z) = (θ − θ CM )(θ − θ CM )T π(θ | z) dθ ∈ Rp×p
Rp

or confidence regions, also provides further indicators for the sake of uncertainty quantifi-
cation.

The simplest probabilistic model that can be used to describe experimental uncertain-
ties is the Gaussian model, for which the noise ε ∼ N (0, Σε ) is normally distributed, with
mean 0 and covariance matrix Σε . In this case, the likelihood function is
µ ¶
1 2
π(z | θ) ∝ exp − kz − f (θ)kΣ−1 .
2 ε

If we can assume a Gaussian model also on the prior knowledge of the parameters distri-
butions, i.e. πprior ∼ N (θ p , Σp ), then the posterior PDF will be normally distributed as
well, µ ¶
1 2 1 2
π(θ | z) ∝ exp − kz − f (θ)kΣ−1 − kθ − θ p kΣ−1 . (150)
2 ε 2 p

In this case, the maximum a posteriori estimator is

µ ¶
1 1
θ M AP = arg minp kz − f (θ)k2Σ−1 + kθ − θ p k2Σ−1 ,
θ∈R 2 ε 2 p

that is, it coincides with the estimator obtained by solving the (static version, with K = 1
of the) regularized least-squares problem (130), as long as we choose θ 0 = θ p , Pθ = Σp
and M = Σ−1 ε . If we assume instead that no information is available about the parameter
distribution except that it resides in a subset D of the parameter space Rp , πprior (θ) ∼
U(D) is a uniform distribution over D.

Remark 9. Note that parameter estimation techniques of Sect. 10.1 can be seen as strate-
gies yielding point estimates without any reference to underlying statistical models; how-
ever, when in the Bayesian framework a Gaussian assumption is made, the regularization
term and the norms k · kΣ−1
ε
, k · kΣ−1
p
and the value θ p have a clear interpretation in terms
of measurement noise and prior probability distribution; hence, a prior that carries suf-
ficient information about the true underlying structure of the parameters often provides
more meaningful estimates and regularizes the inverse problem in a more natural way
than relying on abstract regularization terms, as in (130), that might not have any inter-
pretation. The benefit of casting parameter estimation problems under the form of inverse
UQ problems is that one is also able to characterize the variance of the prediction due to
measurement and model errors more precisely than from the single-point estimates.

147
 Exploring the posterior distribution in the case where the input space D has a larger
dimension than p = 2 or 3 calls into play suitable sampling methods, among which the
Markov chain Monte Carlo (MCMC) techinques represent the most relevant example.
Instead than evaluating the posterior PDF at a single point, an MCMC technique is a
systematic way to generate a sample which can be used to explore the distribution, as well
as to perform MC integration like in (149) in order, e.g., to compute the conditional mean
or the conditional covariance. In the former case, the posterior π(θ | z) plays the role of
target probability distribution that we want to explore, and is obtained as a realization of
a Markov chain by relying, e.g., on the following Metropolis-Hastings algorithm:

1. Pick an initial θ 1 .

2. For m = 1, . . . , M :

(a) compute π(θ m | z);

b =
(b) draw w ∼ N (0, Σε ) and take a random step to find the next candidate θ
θ k + w;
(c) compute π(θ b | z);
( )
b | z)
π(θ
(d) define the acceptance ratio αm = min 1, ;
π(θ m | z)
b and add to the set of samples
(e) let u ∼ U([0, 1]); if u ≤ αm , accept, set θ m+1 = θ,
Ξ; otherwise reject, and keep θ m+1 = θ m .

After M steps of the algorithm, the set Ξ of samples contains realizations of the probability
distribution π(θ | z). The random step in point (b) from the current sample to the next
candidate is distributed as white noise; its covariance should be chosen as large as possible
while still maintaining a reasonable acceptance rate; moreover, the initial points of the
generated set usually poorly represent the distribution to be explored and are then removed
from the sample; to learn more on MCMC algorithms we refer, e.g., to [283, 491, 578] and
references therein.

11.2.2 Dynamical problems

Backward UQ problems whose forward system is stationary, or with observations that
are not acquired sequentially in time, can be cast under the form of a Bayesian inverse
problem as shown in the previous section. In these cases, backward UQ can be formu-
lated in the Bayesian framework by means of Bayesian filtering methods, among which
the Kalman filter can be seen as a particular instance. These problems are also referred
to as nonstationary inverse problems. As in Sect. 10.2, depending on the quantities that
have to be estimated, the problem can be formulated as a state estimation or a joint
state-parameter estimation problem. We will treat both these cases, but we will limit our
discussion to finite-dimensional models (that usually arise from space and time discretiza-
tion of unsteady PDEs) and using discrete time evolution models. Further details can be
found, e.g., in [283, 508, 531].

148
 Let us denote by {X k }K k K
k=0 and {Z }k=1 two stochastic processes; the former is related
with the quantity we are interested in, whereas the latter represents the measurement. In
particular, the random vector X k ∈ Rnx is referred to as the state vector, whereas the
random vector Z k ∈ Rnz is referred to as the observation, both considered at the k-th
time instant τ k . From a Bayesian standpoint, the goal is to use the observations until
time k to get information about the state X k and quantify the uncertainty related to this
estimate. To frame this problem in the Bayesian setting, we assume that {X k }K k=0 and
k K
{Z }k=1 are an evolution-observation model, that is:
1. {X k }K k K
k=0 and {Z }k=1 are Markov processes, that is

π(xk+1 | x0 , x1 , . . . , xk ) = π(xk+1 | xk ), k = 0, 1, . . .

π(z k | x0 , x1 , . . . , xk ) = π(z k | xk ), k = 1, 2, . . . ;

2. {X k }K
k=0 depends on the past observations only through its own history, that is

π(xk+1 | xk , z 1 , . . . , z k ) = π(xk+1 | z k ), k = 0, 1, . . . .

Here xk , z k denote the realizations of the processes {X k }K k K

k=0 and {Z }k=1 , respectively.
In order to characterize such a model, we need to specify the PDF of the initial state X 0 ,
πprior (x0 ), the so-called transition kernel π(xk+1 | xk ), k = 0, 1, . . . and the conditional
probability π(z k | xk ), k = 1, 2, . . ., the so-called (conditional) likelihood function. We
assume to deal with a state evolution equation under the form

X k+1 = f k+1 (X k , V k+1 ), k = 0, 1, . . . (151)

and an observation equation under the form

Z k = hk (X k , W k ), k = 1, 2, . . . ; (152)

f k+1 and hk are known functions, whereas V k+1 ∈ Rnx and W k ∈ Rnz represent the state
noise and the observation noise, respectively. We want to determine the conditional PDF
π(xk | D k ) of the state at the k-th time instant given the observations D k = (z 1 , . . . , z k )
up to the same time instant; this procedure is usually referred to as filtering problem. By
recursive application of the Bayes theorem, we have that
• the time evolution updating, that is, the problem of determining π(xk+1 | D k ) given
π(xk | D k ) and the transition kernel π(xk+1 | xk ), provides
Z
k+1 k
π(x |D ) = π(xk+1 | xk )π(xk | D k )dxk ; (153)
Rnx

• the observation updating, that is, the problem of determining the posterior distribu-
tion π(xk+1 | D k+1 ) of X k | D k based on the new observation Z k+1 given π(xk+1 | D k )
and the likelihood function π(z k+1 | xk+1 ), provides

π(z k+1 | xk+1 )π(xk+1 | D k )

π(xk+1 | D k+1 ) = (154)
π(z k+1 | D k )

149
 where Z
k+1 k
π(z |D ) = π(z k+1 | xk+1 )π(xk+1 | D k )dxk+1 .
Rnx

Formula (154) is the Bayes formula where π(xk+1 | D k ) is considered as prior distri-
bution for xk+1 .

The Kalman filter introduced in Sect. 10.2.1 is a remarkable instance of Bayesian filter
method. Indeed, let us assume that the state and the observation equations are linear
with additive noise processes, that is

f k+1 (xk , v k ) = Ak|k+1 xk + v k , h k = H k xk + w k ,

for given matrices Ak|k+1 , Hk , that the noise vectors V k+1 and W k are mutually inde-
pendent, Gaussian, with zero mean and known covariances Qk+1 and Rk , respectively,
and that the prior PDF of X 0 is Gaussian with mean m0 and covariance P 0 . Under
these assumptions, the time evolution and the observation updating formulas (151)–(152)
involve Gaussian distributions, whose means and covariances can be updated at each step
according to prediction (141a)–(141b) and correction (141d)–(141e) formulas, respectively.
In particular, we have that

π(xk+1 | xk ) ∼ N (Ak|k+1 xk , Qk )
π(z k | xk ) ∼ N (Hk xk , Rk ).

The Bayesian filtering equations can be evaluated in closed form, yielding the following
Gaussian distributions:

π(xk | D k−1 ) ∼ N (mkf , Pkf )

π(xk | D k ) ∼ N (mka , Pka )
π(z k | D k−1 ) ∼ N (Hk mkf , Hk Pkf HkT + Rk )

where the means and the variances can be computed with the following KF prediction
step:
mkf = Ak−1|k mk−1
(155)
Pkf = Ak−1|k Pk−1
a AT
k−1|k + Qk−1

and the consequent correction step:

Kk = Pkf HkT (Hk Pkf HkT + Rk )−1

mka = mkf + Fk (z k − Hk mkf ) (156)
Pka = (I − Kk Hk )Pkf .

In other words, under the Gaussian assumption, the density is updated only through the
mean and the covariance. A similar interpretation also holds for the EKF, as soon as a
Gaussian approximation of the densities is considered, and the evolution of these densities
is taken into account. In this respect, Bayesian filtering can be seen as a generalization
of deterministic filters, like the KF, the EKF and the UKF introduced in Sect. 10.2. See,

150
e.g., [283, 508, 531].

As already remarked in Sect. 10.2.2, when the evolution model is fully nonlinear, the
EKF, which can be seen as a particular instance of approximate Gaussian filter, may per-
form badly: this can be explained by considering that the push-forward of the previous
state estimate (which has a Gaussian distribution) by a nonlinear map is poorly approx-
imated by a Gaussian distribution. To avoid the linearization of the evolution and the
observation models, one can rely on Monte Carlo methods to simulate the distributions
by random samples, similarly to what has been done in the static case. This strategy
yields the so-called particle filters (also referred to as sequential Monte Carlo methods),
nowadays very popular for complex backward UQ problems.
k|k k|k
The goal of a particle filter is to sequentially produce an ensemble {x1 , . . . , xNe } of
Ne particles, that is, a random sample distributed according to the conditional probability
distribution π(xk | D k ). The Ensemble Kalman Filter (EnKF), introduced by Evensen
[163, 164], is a particle filter exploiting the idea of approximating the means and the
covariances of the current estimate involved in the Kalman filter prediction-correction
strategy by a set of particles sampled from the distribution. Unlike the KF, we evaluate the
error covariance predictions and corrections by the ensemble covariance matrices around
the corresponding ensemble mean, instead of classical covariance equations (141b)–(141e)
given in the KF algorithm. The covariance matrices of the state vector X need not be
evolved, thus eliminating the costs associated with storing, multiplying and inverting the
matrices appearing in the equations (141b)–(141e).
The ensemble is initialized by drawing Ne independent particles from, say, a Gaussian
distribution with mean m0 and covariance P0 . Then:

• at each prediction step, each particle is evolved using the KF prediction step,

xk|k−1
e = Ak−1|k xk−1|k−1
e + v k−1

if the system is linear, or

xk|k−1
e = f k (xk−1|k−1
e , v k−1 )

if the system is nonlinear;

• at each correction step, the observation z k is replicated Ne times, obtaining

dke = z k + η ke , η ke ∼ N (0, Rk ).

Then, the empirical mean

Ne
1 X
x̄k|k−1
e = xk|k−1
e
Ne
e=1

151
 and the empirical covariance
e ³
N ´³ ´T
E 1 X
Ck|k−1 = xk|k−1
e − x̄ k|k−1
e x k|k−1
e − x̄ k|k−1
e
Ne − 1
e=1

k|k−1
of the particles set {xe }N
e=1 are computed. The exact Kalman gain is then
e

approximated by
KkE = Ck|k−1
E
HkT (Hk Ck|k−1
E
HkT + Rk )−1
and, finally, the state correction is obtained by applying the formula (141d) to each
particle, that is,
xk|k
e = xe
k|k−1
+ KkE (dke − Hk xk|k−1
e ).

Several alternative implementations can be found , e.g., in [163, 165]; see, e.g., also
[275, 291, 158] for more on the Ensemble Kalman Filter.

If the backward UQ problem also involves random inputs (as in the case of model
parameters θ, which can also be time-varying), the problem of state estimation and pa-
rameter estimation simultaneously arises. Generally speaking, there is no unique opti-
mal solution for this problem. Similarly to the state augmentation technique presented
in Sect. 10.2, a possible way to face this problem is to treat the unknown parameters
θ as part of the state, and use conventional filtering technique to infer the parameter
and state simultaneously; this strategy goes under the name of joint estimation; see, e.g.
[382, 101, 165] for more details on this aspect. An example related to cardiovascular mod-
eling where an EnKF can be exploited to deal with state and parameter estimation is
addressed in Sect. 11.3.2.

11.3 Applications to Cardiovascular Modeling

Until recently, UQ has not represented a priority for cardiovascular modeling. Today, a
growing number of works focus on both forward and inverse UQ problems, taking into ac-
count uncertainties related to (i) measurement errors in experimental data (also referred
to as observational uncertainty); (ii) model parameters, which may result from observa-
tional uncertainty as well as from variability, or lack of information; (iii) boundary and/or
initial conditions and, possibly, (iv) the computational model itself, because of model lim-
itations (model uncertainty) or systematic approximation errors (which can be seen as a
form of epistemic uncertainty) introduced, e.g., when the original high-fidelity model is
replaced by a cheaper surrogate model or a reduced-order model. The list of contributions
we mention is unavoidably incomplete.

The effect of uncertain parameters in one dimensional models of the arterial network
has first been considered in [603]; a high-order stochastic collocation method based on the
generalized polynomial chaos expansion, combined with a discontinuous Galerkin spec-
tral/hp element discretization in physical space, has been exploited to analyze the effects

152
of parametric uncertainties related with material properties and the initial cross-sectional
area of the arterial vessel in pulse wave propagation in a network up to 37 vessels con-
nected via 16 bifurcations. A more general setting taking into account a wider network and
many more sources of parametric uncertainties, including parameter-dependent boundary
conditions in each distal boundary site and geometrical parameters describing the cross
section area) in each arterial segment, has been considered in [94]. Similar problems,
involving both sensitivity analysis and uncertainty propagation for networks built over
subject-specific datasets, can also be found in [328, 264, 263].

The use of an adaptive stochastic collocation framework to deal with UQ in haemody-

namical simulations has been considered in [506], where relevant hemodynamic features
are extracted – and their uncertainty is quantified – on two idealized problems, namely an
abdominal aortic aneurysm and a carotid artery bifurcation, and on a first case of subject-
specific problem, a Fontan procedure for congenital heart defects. In the former case, for
instance, the radius of the abdominal aortic aneurysm, the radius and inflow velocity of
the carotid artery bifurcation, and the flow split of the left and right pulmonary arteries
are treated as random variables, following either Gaussian or uniform distributions, to
account for the uncertainty impact on blood flows described by three-dimensional Navier-
Stokes equations with rigid walls in small arterial portions. More detailed versions of this
methodology have been recently exploited in [503, 504] to analyze the impact of geometric
uncertainties (that is, anatomic uncertainties resulting in error of the reconstructed geom-
etry) and physiological uncertainties (yielding to errors in boundary conditions or blood
viscosity) on the blood flow and pressures in the coronary artery.
The problem of calibrating outflow boundary conditions of blood flow simulations in
truncated arterial domains is cast in a Bayesian framework in [134, 441], with the goal
of quantifying the uncertainty affecting velocity and flow related variables of interest, all
treated as random variables.
A complete uncertainty propagation pipeline from clinical data to computational re-
sults has been considered in [511] focusing on single ventricle palliation surgery example.
After determining the probability density functions of right pulmonary artery flow split
ratio and average pulmonary pressures from clinical measurements, Bayesian parameter
estimation is carried out in order to characterize the distributions of boundary conditions
yielding the observed flow splits and average pressure distributions. Then, uncertain-
ties in the boundary conditions are propagated to simulation predictions by employing
sparse grid stochastic collocation to statistically characterize model predictions of post-
operative hemodynamics in models with and without pulmonary artery stenosis, in order
to quantify the statistical variability in virtual surgery predictions. Finally, the impact of
uncertainty on the optimal design of idealized bypass graft models has been considered,
e.g., in [505, 318, 319].

Regarding cardiac electrophysiology, a systematic application of UQ techniques to the

forward problem of electrocardiography (namely, the characterization of the torso poten-
tial given the electrical conductivity inside the heart and the torso) has been considered
in [205], focusing on a simplified two-dimensional configuration representing a geometric

153
model of a slice through the human thorax. The forward UQ propagation problem yielding
the standard deviation of the resulting stochastic torso potentials depending on several un-
certainties affecting, e.g., the electrical conductivities of the organs and the representation
of cardiac sources, has been faced by a SG method based on the generalized polynomial
chaos expansion. Model personalization in cardiac electrophysiology has been considered
in [301], where an efficient Bayesian method exploiting polynomial chaos and compressed
sensing has been applied to an eikonal-diffusion model involving a large (> 10) number of
parameters, by integrating uncertainty on data and parameters.
In [278] a UQ framework for cardiac action potential models has been addressed, focus-
ing on (i) the inverse UQ problem of inferring the maximal conductance of ionic channels
from noisy experimental recordings, and (ii) the forward UQ problem of propagating the
uncertainty in maximal ion channel conductances to suitable outputs of interest, such as
the action potential duration. To solve the inverse UQ problem a Bayesian framework,
such as the one described in Sect. 11.2, has been successfully employed, whereas the UQ
propagation problem has been tackled by means of a MC approach exploiting suitable
surrogate models, such as Gaussian Process emulators, to speed up the evaluation of the
system model under analysis. A detailed analysis of the mechanisms underlying physio-
logical variability in cardiac electrophysiology and pro-arrytmic risks under a variety of
conditions can be found, e.g., in [390].
Concerning heart electromechanics, the quantification of the effect of uncertainties in
the fiber orientation and the elasticity parameters included in the strain energy function
on global output quantities such as the increase in cavity volume, the elongation of the
ventricle, the increase in inner radius, the decrease in wall thickness, and the rotation
at apex during the passive filling phase, has been addressed in [422]. In [584], cardiac
tissue properties have been estimated by integrating structural information with electro-
physiological data from, e.g., electroanatomical mapping systems. By means of Bayesian
inference tools, a simultaneous description of clinically-relevant electrophysiological con-
duction properties and their associated uncertainty for various levels of noise have been
obtained, together with suitable design strategies to optimize the location and number of
measurements required to maximize information and reduce uncertainty. A relevant ap-
plication of UQ techniques to nonlinear biomechanics has been considered in [54], where
parametric uncertainties related with the constitutive law for the artery wall of an AAA
are modeled as random fields. In particular, a lognormal three-dimensional random field is
used to describe the inter- and intra-patient variations of one constitutive parameter of a
hyperelastic constitutive model, and sampling-based approaches like MC are used to solve
the resulting stochastic mechanical problem. A multi-fidelity Bayesian framework incor-
porating information from different low-fidelity models has been developed to speed up the
intensive approximation of the resulting problem, following some general ideas reported in
[292, 304]. Employing surrogate models or more reliable reduced-order models (for more
on this subject, see Sect. 12) to speed up the numerical solution of direct and inverse UQ
problems represents indeed a powerful alternative to SC and SG methods, which has only
been partially touched in the case of simple problems, like in [126, 95, 142, 356, 426].

154
11.3.1 Backward uncertainty propagation in a simplified blood flow model
We want to compare the solution of a parameter estimation problem in a variational
framework and in a Bayesian setting, on a simplified blood flow problem. We consider a
two-dimensional section of a carotid bifurcation where the diameters dc , db of the common
carotid artery (CCA) at the bifurcation and of the mid-sinus level of the internal carotid
artery (ICA) are the input parameters, assumed to be uncertain. In this idealized setting,
our goal is to identify θ = (dc , db )T from the observation of the mean pressure drop
Z Z
z= p dΓ − p dΓ
Γin ΓICA

between the internal carotid outflow ΓICA and the inflow Γin , see Fig. 31 (a). For the sake
of simplicity a steady incompressible Navier-Stokes model is employed to model the blood
flow, although the same approach concerning the solution of the inverse UQ problem can
be applied to the case of an unsteady fluid model, as soon as a peak (or time average)
pressure drop is evaluated. Although presented here in a very simplified way for the sake
of illustration, the problem of recovering information about vessel features by evaluating
physical indices related to flow variables is of general interest also for diagnostic purposes;
for instance, the fractional flow reserve is a procedure exploited in coronary catheterization
to measure pressure differences across a coronary artery stenosis in order to determine the
stenosis degree.
The parameter estimation problem is first solved by minimizing a least-squares func-
tional for different observations of the pressure drop, z = −1400 and z = −2200, by
assuming 5% relative additive noise in the measurements. The results of the inverse iden-
tification problem are represented in Fig. 31(b) for 100 realizations of random noise in
both cases. The recovered diameter values are shown to be quite sensitive to small noise
variations; this is due to the fact that several geometrical configurations – in terms of
diameters (dc , db ) – may correspond to the same output observation. The backward UQ
propagation problem is then solved in a Bayesian setting, characterizing the posterior PDF
of θ = (dc , db ) starting from a Gaussian prior and assuming a Gaussian model to describe
experimental noise; in particular, we can assume that the two diameters are either a pri-
ori independent (case (c), left), or correlated (case (c), right)). The two corresponding
posterior distributions, obtained for the observations z = −1400 and z = −2200 of the
pressure drop, are reported in Fig. 31 (d-e).

11.3.2 Backward uncertainty propagation in cardiac electrophysiology

We finally consider the problem of identifying the size and the position of an ischemic
region in the myocardial tissue, as a remarkable instance of backward UQ problem dealing
with cardiovascular applications. Such a problem case can be cast under the form of a joint
state/parameter estimation problem, and can be efficiently faced by an EnKF technique.
The state evolution equation is given by the (space and time discretization of the) mon-
odomain equation for the transmembrane potential Vm , coupled with an Alliev-Panfilov
cellular model involving a single gating variable w; the state variable is thus u = (Vm , w)T .

155
Figure 31: (a) Velocity profiles [cm/s] for different carotid bifurcations parametrized with
respect to the diameters dc , db ; (b) variational parameter estimation and isolines of the
pressure drop; (c) two different choices of the prior distribution on diameters θ = (dc , db )T ;
(d)-(e): results of the backward UQ problem obtained with the priors in (c) with observed
pressure drop z obs = −1400 and z obs = −2200 [dyn/cm2 ], respectively. These numerical
results are obtained using the Matlab Finite Element library MLife

Similarly to the example in Sect. 11.1, to represent lacks of conductivity in the tissue we
introduce the relative conductivity σ = σ(x; θ) ∈ [0, 1]; σ = 1 and σ = 0 correspond to
proper electrical conduction or total lack of conductivity, respectively.
Since the position and the extension of the ischemic region is unknown, σ(x; θ) is a
random field. To make its representation low-dimensional, we parametrize the field σ(x; θ)
in terms of p = 20 random inputs θ = (θ1 , . . . , θ20 )T entering into a linear combination of
p = 20 radial basis functions (RBFs), around p given centers {xi }20 i=1 ,

p µ ¶ p µ ¶
1 X kx − xi k2 X kx − xi k2
σ(x; θ) = θi exp − , ν(x) = exp − ;
ν(x) λ2 λ2
i=1 i=1

ν(x) is a normalization factor so that σ(x; θ) ∈ [0, 1]. We consider a subject-specific left-
ventricle geometry, and assume that a set of data z 1 , . . . , z K is acquired sequentially over
a time interval; in particular, at each time instant τ k , k = 1, . . . , K data are given by the

156
transmembrane potential values computed at a set of 9 points located on the endocardium,
polluted by Gaussian noise with zero mean and known covariance, to simulate the effect
of experimental noise while evaluating the so-called simultaneous endocardial mapping in
the human left ventricle; these invasive measurements are usually obtained using a non-
contact catheter (see, e.g., [512, 6]). For the sake of the computational experiment, we
assume that the data z k , k = 1, . . . , K, are generated by the monodomain model with
relative conductivity σ ∗ = σ ∗ (x; θ ∗ ) for a particular choice θ = θ ∗ of the input vector,
that is,
z k = h(u(θ ∗ )) + η k , η k ∼ N (0, Rk ).
The goal is thus to recover the (posterior) PDF of θ K | D k given the observations D k =
(z 1 , . . . , z k ), evaluate the conditional mean θ K K k
CM = E[θ | D ], and finally estimate the
conductivity field as σ(x; θ K CM ).

Starting from a prior distribution for the input vector θ 0 ∼ U([0, 1]20 ), reflecting the
lack of information about the possible presence and position of the ischemic region, we
rely on the EnKF to produce sequentially an ensemble {θ ke }N e=1 of Ne particles distributed
e

according to the conditional distribution π(θ k | D k ), and the associated ensemble of Ne

states {uk|k (θ ke )}N
e=1 . Hence, starting from the initial ensemble {P
e (0) , U (0) } sampled from

the prior distribution, the prediction-analysis procedure of the EnKF is given by the
following two stages recursion:
1. at each prediction step, compute the solution uk|k−1 (θ k−1
e ) of the state system over
k−1 k k|k−1 k−1
[τ , τ ) with initial datum u (θ e );

2. at each correction step, the observation z k is replicated Ne times, obtaining

dke = z k + η ke , η ke ∼ N (0, Rk ).

Then, the sample means

Ne Ne N
1 X 1 X k−1 1 X e

uk|k−1 = uk|k−1 (θ k−1

e ), sk|k−1
e = h(uk|k−1 (θ k−1
e )), θe = θ k−1
e ,
Ne Ne Ne
e=1 e=1 e=1
(157)
and the sample covariances
N
1 X e

Ckss = (h(uk|k−1 (θ k−1

e )) − sek|k−1 )(h(uk|k−1 (θ k−1
e )) − sk|k−1
e )T ∈ Rnz ×nz ,
Ne − 1
e=1
(158)
Ne
X
1 k−1
Ckθs = (θ k−1
e − θ e )(h(uk|k−1 (θ k−1e )) − sek|k−1 )T ∈ Rp×nz , (159)
Ne − 1
e=1
Ne
X
1
Ckus = (uk|k−1 (θ k−1
e ) − uk|k−1 )(h(uk|k−1 (θ k−1
e )) − sk|k−1
e )T ∈ Rnx ×nz ;
Ne − 1
e=1
(160)

157
 are computed. Finally, the state/parameters ensembles are updated through the
following Kalman formula
· ¸ · ¸ · k ¸
θ ke θ k−1 Cθz
k
e
= k k−1 + (Rk + Ckzz )−1 (dke − h(uk|k−1 (θ k−1
e ))), (161)
k|k
u (θ e ) u (θ e ) Ckuz

for each e = 1, . . . , Ne .

The numerical results obtained by this procedure are reported in Fig. 32. The data
have been generated by the monodomain model with the relative conductivity field σ ∗ =
σ ∗ (x; θ ∗ ) reported on Fig. 32, left; a uniform prior distribution θ 0 ∼ U([0, 1]20 ) is assumed.
After executing the EnKF algorithm, we obtain the posterior distribution whose mean
and 5%, 95% quantiles are reported on Fig. 32, bottom. The uncertainty on the input
parameters is greatly reduced, while the conditional mean of the posterior distribution
correctly approximates the realization of the random field reported on the left, by which
data exploited for the sake of parameter estimation have been generated.

12 Reduced order modeling

As seen in Sects. 9–11, several numerical strategies used to face PDE-constrained opti-
mization, parameter estimation, and uncertainty quantification problems arising in cardio-
vascular modeling involve the approximation of PDE systems for several input parameter
values. On the other hand, the repeated evaluation of input/output maps to characterize
different scenarios requires many queries to numerical models, too. Not only, if quantita-
tive outputs are meant to support clinicians and medical doctors in their decisions, each
new numerical simulation should be carried out very rapidly (say, order of minutes) on
deployed platforms rather than on huge parallel hardware architectures, possibly requiring
limited data storage and memory capacity. Meeting all these requirements is a challenging
task, that makes traditional high-fidelity, or full-order, techniques (such as the FE method)
ill-suited, despite the constant growth of computer resources available.

Reduced-order models (ROMs) represent emerging methodologies aimed at reducing the

computational complexity and costs entailed by the repeated solution of PDE problems
[12, 45, 474]. In the case of parametrized PDEs (that is, PDEs depending on a vector of
parameters µ ∈ P ⊂ Rp ) PDEs, the reduced basis (RB) method is a remarkable example
of ROM that allows to dramatically reduce the dimension of the discrete problems arising
from numerical approximation – from millions to hundreds, or thousands at most, degrees
of freedom.
Here, y(t; µ) represent the solution of a time-dependent nonlinear problem (here µ ∈
P ⊂ Rp denotes a set of input parameters) under the form:

dy(t; µ)
M (µ) + A(µ)y(t; µ) + F (y(t; µ)) = f (t; µ), t ∈ (0, T ]
dt (162)
y(0; µ) = y0 (µ)

158
Figure 32: Top: 5% quantile, mean and 95% quantile of the uniform prior distribution of
θ 0 . Middle: identification of p = 20 parameters through the EnKF algorithm. References
values θi∗ , i = 1, . . . , 20, estimates (θ̄ek )i and confidence intervals are reported in dot red,
blue and dot blue lines, respectively. Bottom: 5% quantile, conditional mean and 95%
quantile of the posterior distribution of θ K . These numerical results are obtained using
the Finite Element library redbKIT v2.1 ([Link]/redbKIT/redbKIT/releases)

stemming from the (e.g. finite element) discretization of a parametrized PDE, being
A(µ) ∈ RNh ×Nh , M (µ) ∈ RNh ×Nh and F (µ) : RNh → RNh three operators corresponding
to the linear, the mass and the nonlinear terms of the PDE.
The dimensional reduction is made possible by exploiting the parametric dependence
of the solution manifold, that is, the set Mh = {y(t; µ) : t ∈ (0, T ], µ ∈ P}, thanks to
the evaluation of a database of solutions, or snapshots, for selected parameter values, and
to a (Petrov-)Galerkin projection onto the RB space spanned by a set of RB functions.
In the case of a stationary problem, these latter are the snapshots themselves if a greedy
algorithm is used, or the first singular vectors of the snapshot matrix if proper orthogonal

159
decomposition (POD) is used; see, e.g. [471, 247, 112]. For time-dependent problems,
the parameter space can still be sampled relying on one of the two mentioned techniques,
whereas POD is usually exploited for reducing trajectories of the system over the time
interval.
Hence, the RB method seeks an approximation VN yN (t; µ) ≈ y(t; µ) to the solution of
(162), being yN (t; µ) the reduced state vector and VN = [v1 , . . . , vN ] ∈ RNh ×N a matrix
stacking by columns the RB functions. A possible RB approximation of (162) would read:

dyN (t; µ)
MN (µ) + AN (µ)yN (t; µ) + VNT F(VN yN (t; µ)) = fN (t; µ), t ∈ (0, T ]
dt
yN (0; µ) = yN,0 (µ)
(163)
where

AN (µ) = VNT A(µ)VN , MN (µ) = VNT M (µ)VN , fN (t; µ) = VNT f (t; µ)

The arrays appearing in (163) can be efficiently assembled in a rapid online phase by com-
bining parameter-independent quantities stored during a more expensive offline phase.
Suitable hyper-reduction techniques are instead required to manage nonlinear terms in
order to make their assembling independent of the dimension Nh of the high-fidelity prob-
lem.
We emphasize that a RB method requires the solution of some full-order, and therefore
very expensive, discrete equations. The key-idea is that these demanding calculations can
be done offline, before the optimization with respect to the control parameters, or the
parameter estimation, is attempted. As a matter of fact, the cost of each optimization
step performed online is much smaller than that involving the full-order state approxima-
tion; similarly, computing the evolution of a particles set in the Ensemble Kalman Filter
algorithm by querying the ROM can thus be performed in a substantially inexpensive way.

Describing the mathematical principles and the numerical algorithms on which the RB
method is rooted would bring us quite far; the interested reader can refer, e.g., to [471]
for a detailed presentation. We limit ourselves to cite some remarkable applications of
the RB method to problems of interest in cardiovascular modeling (this list is of course
incomplete). Parametrized blood flows in idealized cardiovascular geometries have been
considered in [320, 357] and in [30, 113, 397] by taking into account more complex (and
computationally challenging) subject-specific configurations; in all these cases, solutions
of Navier-Stokes equations are computed with respect to inflow and/or geometrical pa-
rameters. Applications to PDE-constrained optimization problems arising in the context
of optimal design of prosthetic devices can be found, e.g., in [358, 318]. A reduced-order
model based on POD is proposed for the bidomain equations of cardiac electrophysiology
in [75], yielding the efficient approximation of a restitution curve and the estimation of
ionic parameters and infarction locations from synthetic electrocardiograms with an evo-
lutionary algorithm. Alternative options in this respect have been more recently proposed
in [208]. Applications of POD to parameter estimation problems with sequential filtering
techniques can be found in [90], whereas an application of the RB method for the compu-

160
tational speedup of Bayesian inverse problems related to blood flows modeling is reported
in [319, 355].
We highlight that while physical coefficients, boundary and/or initial conditions, as well
as source terms can be usually described in terms of input parameters in a straightforward
way, dealing with geometries of varying shape is much more involved. In this latter case,
additional techniques providing flexible descriptions of complex shapes, possibly involving
few parameters, are required. Notable examples are given by volume-based representations,
which operate on a control volume and define parametric maps by introducing a set of
control points over the control volume; control points displacements, actually inducing a
shape deformation, can thus be treated as input parameters. Within this class, free-form
deformation techniques and interpolants built over a family of radial basis functions have
been successfully employed, see, e.g., [358, 357, 30] for further details. We also point out
that the need of deriving flexible and low-dimensional parametrizations is not confined to
the realm of RB methods for parametrized PDEs. Indeed, it also arises when dealing with
random inputs or fields; in this latter case, input uncertainties are usually parametrized
with respect to a finite number of random variables, corresponding to the retained terms
after truncating Karhunen-Loève or Polynomial Chaos expansions.
The analysis, development and application of reduced-order modeling techniques is a
very active field in the context of numerical approximation for PDEs; without any doubt,
cardiovascular applications represent one of the most relevant testing environment.

Disclaimers
Despite being 200 pages long5 , several topics related to the cardiovascular system have
not been addressed in this review paper. Among others, we mention the venous system
(mandatory when one wants to consider a closed loop model of the CS and playing a
crucial role in some specific pathologies, see e.g. [547]), the metabolic system [128], the
respiratory system [368, 583, 550], the cerebro-spinal fluid circulation [179], the nervous
system [337], and the limphatyc system [362]. For some of them (for example the venous
and the respiratory systems) research has progressed remarkably in the last years. Yet,
the mathematical investigation of these systems is still in its infancy. In particular, their
coupling with CS is almost lacking. Many research avenues are open to the contribu-
tion of both pure and applied mathematicians, with the dream of bringing mathematical
achievements to play a decisive role in clinical everyday’s practice.

Acknowledgments
The authors would like to thank P. Biondetti, M. Domanin, L. Forzenigo (Fondazione
IRCSS Ca’ Granda, Ospedale Maggiore Policlinico, Milan, Italy), S. Ippolito, R. Scro-
fani (Ospedale Sacco, Milan, Italy), D. Catanzariti, M. Centonze (Ospedale S. Maria del
5
By slightly rephrasing Blase Pascal’s quotation, we can state that “we were not good enough to make
it shorter.”

161
Carmine, Rovereto (TN), Italy), for providing the radiological images; G. Aloe (Politecnico
di Milano) for his help in preparing the figures; D. Bonomi, B. Guerciotti, R.M. Lancel-
lotti, S. Pagani (Politecnico di Milano), D. Forti, F. Negri (EPFL, Lausanne, Switzerland),
L. Barbarotta (Technische Universiteit Eindhoven), E. Faggiano (University of Pavia) for
their help in preparing the plots of some numerical results; E. Faggiano for her fruitful sug-
gestions. AQ acknowledges the project “Integrative HPC Framework for Coupled Cardiac
Simulations” (IFCCS) within the PASC (Platform for Advanced Scientific Computing)
network “Life Sciences Across Scales” and the Swiss National Supercomputing Centre
(CSCS), project ID s635. CV has been partially supported by the Italian MIUR PRIN09
project no. 2009Y4RC3B 001.

References
[1] S. Abboud, O. Berenfeld, and D. Sadeh. Simulation of high-resolution QRS complex
using a ventricular model with a fractal conduction system. effects of ischemia on
high-frequency QRS potentials. Circ Res, 68(6):1751–1760, 1991.

[2] S.A. Ahmed and D.P. Giddens. Pulsatile poststenotic flow studies with laser Doppler
anemometry. J Biomech, 17(9):695–705, 1984.

[3] I. Akkerman, Y. Bazilevs, V.M. Calo, T.J.R. Hughes, and S. Hulshoff. The role of
continuity in residual-based variational multiscale modeling of turbulence. Compu-
tational Mechanics, 41:371–378, 2008.

[4] F. Alauzet, B. Fabréges, M.A. Fernández, and M. Landajuela. Nitsche-xfem for the
coupling of an incompressible fluid with immersed thin-walled structures. Comput.
Methods Appl. Mech. Engrg., 301:300–335, 2016.

[5] R.R. Aliev and A.V. Panfilov. A simple two-variable model of cardiac excitation.
Chaos, Solitons & Fractals, 7(3):293–301, 1996.

[6] A. Álvarez, F. Alonso-Atienza, J.L. Rojo-Álvarez, A. Garcia-Alberola, and

M. Moscoso. Shape reconstruction of cardiac ischemia from non-contact intracardiac
recordings: a model study. Math. Computer Modeling, 55:1770–1781, 2012.

[7] D. Ambrosi, G. Arioli, F. Nobile, and A. Quarteroni. Electromechanical coupling in

cardiac dynamics: The active strain approach. SIAM J. Appl. Math., 2(71):605–621,
2011.

[8] D. Ambrosi and S. Pezzuto. Active stress vs. active strain in mechanobiology: Con-
stitutive issues. Journal of Elasticity, 2(107):199–212, 2012.

[9] B. Andreianov, M. Bendahmane, A. Quarteroni, and R. Ruiz-Baier. Solvability

analysis and numerical approximation of linearized cardiac electromechanics. Math-
ematical Models and Methods in Applied Sciences, 25(5):959–993, 2015.

162
[10] L. Antiga, J. Peiró, and D.A. Steinman. From image data to computational do-
mains. In Cardiovascular mathematics, edited by L. Formaggia, A. Quarteroni, A.
Veneziani, Chapter 4, pages 123–175. Springer, 2009.

[11] L. Antiga, M. Piccinelli, L. Botti, B. Ene-Iordache, A. Remuzzi, and D.A. Stein-

man. An image-based modeling framework for patient-specific computational hemo-
dynamics. Med Biol Eng Comput, 46:10971112, 2008.

[12] A. C. Antoulas. Approximation of Large-Scale Dynamical Systems. Society for

Industrial and Applied Mathematics, Philadelphia, 2005.

[13] M. Astorino, F. Chouly, and M. Fernández. Robin based semi-implicit coupling in

fluid-structure interaction: stability analysis and numerics. SIAM J. Sci. Comput.,
31(6):4041–4065, 2009.

[14] M. Astorino, J.F Gerbeau, O. Pantz, and K.F. Traoré. Fluid-structure interaction
and multi-body contact. application to the aortic valves. Comput. Methods Appl.
Mech. Engrg., 116(4):721–767, 2010.

[15] M. Astorino and C. Grandmont. Convergence analysis of a projection semi-

implicit coupling scheme for fluid-structure interaction problems. Numer. Math.,
45-46(198):3603–3612, 2009.

[16] M. Astorino, J. Hamers, C.S. Shadden, and J.F. Gerbeau. A robust and efficient
valve model based on resistive immersed surfaces. International Journal for Numer-
ical Methods in Biomedical Engineering, 28(9):937959, 2012.

[17] C.M. Augustin, G.A. Holzapfel, and O. Steinbach. Classical and all-floating feti
methods for the simulation of arterial tissues. International Journal for Numerical
Methods in Engineering, 99(4):290–312, 2014.

[18] F. Auricchio, A. Ferrara, and S. Morganti. Comparison and critical analysis of

invariant-based models with respect to their ability in fitting human aortic valve
data. Annals of Solid and Structural Mechanics, 4(1):1–14, 2012.

[19] F. Auricchio, F. Lefieux, A. Reali, and A. Veneziani. A locally anisotropic fluid-

structure interaction remeshing strategy for thin structures with application to a
hinged rigid leaflet. International Journal for Numerical Methods in Engineering,
107(2):155–180, 2016.

[20] A. P. Avolio. Multi-branched model of the human arterial system. Medical and
Biological Engineering and Computing, 18:709–718, 1980.

[21] I. M. Babuška, F. Nobile, and R. Tempone. A stochastic collocation method for

elliptic partial differential equations with random input data. SIAM J. Numer.
Anal., 45:1005–1034, 2007.

163
[22] I. M. Babuška, R. Tempone, and G. E. Zouraris. Galerkin finite element approxi-
mations of stochastic elliptic partial differential equations. SIAM J. Numer. Anal.,
42:800–825, 2004.

[23] S. Badia, F. Nobile, and C. Vergara. Fluid-structure partitioned procedures based

on Robin transmission conditions. J. Comput. Physics, 227:7027–7051, 2008.

[24] S. Badia, F. Nobile, and C. Vergara. Robin-Robin preconditioned Krylov methods for
fluid-structure interaction problems. Comput. Methods Appl. Mech. Engrg., 198(33-
36):2768–2784, 2009.

[25] S. Badia, A. Quaini, and A. Quarteroni. Modular vs. non-modular preconditioners

for fluid-structure systems with large added-mass effect. Comput. Methods Appl.
Mech. Engrg., 197:4216–4232, 2008.

[26] S. Badia, A. Quaini, and A. Quarteroni. Splitting methods based on algebraic

factorization for fluid-structure interaction. SIAM J Sc Comp, 30(4):1778–1805,
2008.

[27] E.Z. Bagci, Y. Vodovotz, T.R. Billiar, B. Ermentrout, and I. Bahar. Computational
insights on the competing effects of nitric oxide in regulating apoptosis. PLOS One,
3(5):e2249, 2008.

[28] J. M. Ball. Constitutive inequalities and existence theorems in nonlinear elastostat-

ics. In R.J. Knowps, editor, Nonlinear Analysis and Mechanics, 1997.

[29] J.M. Ball. Convexity conditions and existence theorems in nonlinear elasticity.
Archive for Rational Mechanics and Analysis, 63(4):337–403, 1976.

[30] F. Ballarin, E. Faggiano, S. Ippolito, A. Manzoni, A. Quarteroni, G. Rozza, and

R. Scrofani. Fast simulations of patient-specific haemodynamics of coronary artery
bypass grafts based on a pod?galerkin method and a vascular shape parametrization.
J. Comput. Phys., 315:609–628, 2016.

[31] D. Balzani, D. Brands, A. Klawonn, O. Rheinbach, and J. Schroder. On the me-

chanical modeling of anisotropic biological soft tissue and iterative parallel solution
strategies. Archive of Applied Mechanics, 80(5):479–488, 2010.

[32] D. Balzani, P. Neff, J. Schroder, and G.A. Holzapfel. A polyconvex framewok for
soft biological tissues. adjustment to experimental data. International Journal of
Solids and Structures, 43:6052–6070, 2006.

[33] H.T. Banks and K. Kunisch. Estimation Techniques for Distributed Parameter Sys-
tems. Systems & Control: Foundations & Applications. Birkhäuser Basel, 1989.

[34] J.W. Banks, W.D. Henshaw, and D.W. Schwendeman. An analysis of a new stable
partitioned algorithm for fsi problems. part i: Incompressible flow and elastic solids.
J. Comput. Physics, 269:108–137, 2014.

164
[35] L. Barbarotta. A Mathematical and Numerical Study of the Left Ventricular Con-
traction based on the Reconstruction of a Patient Specific Geometry. PhD thesis,
MSc thesis, Mathematical Engineering, Politecnico di Milano, 2014.

[36] A. Barker and X.C. Cai. Scalable parallel methods for monolithic coupling in fluid-
structure interaction with application to blood flow modeling. J. Comput. Phys.,
229:642–659, 2010.

[37] A. Barker and X.C. Cai. Two-level newton and hybrid schwarz preconditioners for
fluid-structure interaction. SIAM J. Sc. Comp., 32(4):23952417, 2010.

[38] A.C.L. Barnard, W.A. Hunt, W.P. Timlake, and E. Varley. A theory of fluid flow
in compliant tubes. Biophysical journal, 6(6):717–724, 1966.

[39] J.D. Bayer, R.C. Blake, G. Plank, and N.A. Trayanova. A novel rule-based algorithm
for assigning myocardial fiber orientation to computational heart models. Annals of
Biomedical Engineering, 40(10):54, 2012.

[40] Y. Bazilevs, V.M Calo, Y. Zhang, and T.J.R Hughes. Isogeometric fluid-structure
interaction analysis with applications to arterial blood flow. Computational Mechan-
ics, 38(4–5):310–322, 2006.

[41] Y. Bazilevs, J.R. Gohean, T.J.R. Hughes, R.D. Moser, and Y. Zhang. Patient-
specific isogeometric fluid–structure interaction analysis of thoracic aortic blood flow
due to implantation of the jarvik 2000 left ventricular assist device. Computer
Methods in Applied Mechanics and Engineering, 198(45):3534–3550, 2009.

[42] Y. Bazilevs, K. Takizawa, and T.E. Tezduyar. Computational fluid-structure inter-

action: methods and applications. John Wiley & Sons, 2012.

[43] G.W. Beeler and H. Reuter. Reconstruction of the action potential of ventricular
myocardial fibres. J Physiol, 268(1):177–210, 1977.

[44] H. Beirao da Veiga. On the existence of strong solutions to a coupled fluid-structure

evolution problem. Journal of Mathematical Fluid Mechanics, 6(1):21–52, 2004.

[45] P. Benner, S. Gugercin, and K. Willcox. A survey of model reduction methods for
parametric dynamical systems. SIAM Review, 57(4):483–531, 2015.

[46] J.L. Bentley and J.H. Friedman. Data structures for range searching. ACM Com-
puting Surveys, 11(4):397–409, 1979.

[47] M. Benzi, G.H. Golub, and J. Liesen. Numerical solution of saddle point problems.
Acta Numerica, 14:1–137, 2005.

[48] M. Benzi, H. G.H. Golub, and J. Liesen. Numerical solution of saddle point problems.
Acta Numerica, pages 1–137, 2005.

165
[49] L. Bertagna, M. D’Elia, M. Perego, and A. Veneziani. Data assimilation in car-
diovascular fluid–structure interaction problems: An introduction. In T. Bodnár,
P. G. Galdi, and Š. Nečasová, editors, Fluid-Structure Interaction and Biomedical
Applications, pages 395–481. Springer Basel, Basel, 2014.
[50] L. Bertagna and A. Veneziani. A model reduction approach for the variational
estimation of vascular compliance by solving an inverse fluid–structure interaction
problem. Inverse Problems, 30(5), 2014.
[51] C. Bertoglio, P. Moireau, and J.F. Gerbeau. Sequential parameter estimation for
fluid-structure problems: Application to hemodynamics. Int. J. Numer. Meth.
Biomed. Engng., 28(4):434–455, 2012.
[52] F. Bertrand, P.A. Tanguy, and F. Thibault. A three-dimensional fictitious domain
method for incompressible fluid flow problems. International Journal for Numerical
Methods in Fluids, 25(6):719–736, 1997.
[53] R.L.T. Bevan, P. Nithiarasu, R. Van Loon, I. Sazonov, H. Luckraz, and A. Garnham.
Application of a locally conservative galerkin (lcg) method for modelling blood flow
through a patient-specific carotid bifurcation. International Journal for Numerical
Methods in Fluids, 64:1274–1295, 2010.
[54] J. Biehler, M.W. Gee, and W.A. Wall. Towards efficient uncertainty quantification in
complex and large-scale biomechanical problems based on a Bayesian multi-fidelity
scheme. Biomech. Model. Mechanobiol., 14(3):489–513, 2015.
[55] P.J. Blanco, S. Deparis, and A.C.I. Malossi. On the continuity of mean total normal
stress in geometrical multiscale cardiovascular problems. J Comp Phys, 51:136–155,
2013.
[56] P.J. Blanco and R.A. Feijóo. A dimensionally-heterogeneous closed-loop model for
the cardiovascular system and its applications. Medical Engineering & Physics,
35(5):652–667, 2013.
[57] P.J. Blanco, R.A. Feijòo, and S.A. Urquiza. A unified variational approach for
coupling 3d-1d models and its blood flow applications. Comput. Methods Appl.
Mech. Engrg., 196:4391–4410, 2007.
[58] P.J. Blanco, M.R. Pivello, S.A. Urquiza, and R.A. Feijòo. On the potentialities of
3d-1d coupled models in hemodynamics simulations. J. Biomech., 42:919–930, 2009.
[59] P.J. Blanco, S.M. Watanabe, and R.A. Feijóo. Identification of vascular territory re-
sistances in one-dimensional hemodynamics simulations. J. Biomech., 45(12):2066–
2073, 2012.
[60] P.J. Blanco, S.M. Watanabe, M.A.R.F. Passos, Lemos P., Feijoo R.A., et al. An
anatomically detailed arterial network model for one-dimensional computational
hemodynamics. Biomedical Engineering, IEEE Transactions on, 62(2):736–753,
2015.

166
[61] J. Blum, F.-X. Le Dimet, and I.M. Navon. Data assimilation for geophysical fluids.
In P.G. Ciarlet, editor, Handbook of Numerical Analysis, volume 14 of Handbook of
Numerical Analysis, pages 385–441. Elsevier, 2009.

[62] T. Bodnár, G.P. Galdi, and Š. Nečasová. Fluid-Structure Interaction and Biomedical
Applications. Springer, 2014.

[63] J.M. Boese, M. Bock, S.O. Schoenberg, and L.R. Schad. Estimation of aortic com-
pliance using magnetic resonance pulse wave velocity measurement. Physics in
Medicine and Biology, 45(6):1703–1713, 2000.

[64] D. Boffi, F. Brezzi, and M. Fortin. Mixed Finite Element Methods and Applications.
Springer, 2013.

[65] D. Boffi and L. Gastaldi. A finite element approach for the immersed boundary
method. Comp. and Struct, 81(8-11):491–501, 2003.

[66] D. Boffi, L. Gastaldi, and L. Heltai. Numerical stability of the finite element im-
mersed boundary method. Mathematical Models and Methods in Applied Sciences,
17(10):1479–1505, 2007.

[67] D. Boffi, L. Gastaldi, L. Heltai, and C. Peskin. On the hyper-elastic formulation

of the immersed boundary method. Computer Methods in Applied Mechanics and
Engineering, 197(25–28):2210–2231, 2008.

[68] E. Boileau, P. Nithiarasu, P.J. Blanco, L.O. Muller, F.E. Fossan, L.R. Hellevik,
W.P. Donders, W. Huberts, M. Willemet, and J. Alastruey. A benchmark study of
numerical schemes for one-dimensional arterial blood flow modelling. International
journal for numerical methods in biomedical engineering, DOI: 10.1002/cnm.2732,
2015.

[69] C. Boldak, Y. Rolland, and C. Toumoulin. An improved model-based vessel track-

ing algorithm with application to computed tomography angiography. Biocybern.
Biomed., 23:41–63, 2003.

[70] D. Bonomi, C. Vergara, E. Faggiano, M. Stevanella, C. Conti, A. Redaelli, G. Pup-

pini, G. Faggian, L. Formaggia, and G.B. Luciani. Influence of the aortic valve
leaflets on the fluid-dynamics in aorta in presence of a normally functioning bicus-
pid valve. Biomechanics and Modeling in Mechanobiology, 6(14):1349–1361, 2015.

[71] I. Borazjani, L. Ge, and F. Sotiropoulos. Curvilinear immersed boundary method

for simulating fluid structure interaction with complex 3d rigid bodies. Journal of
Computational Physics, 227(16):7587–7620, 2008.

[72] R.M. Bordas, K. Gillow, D. Gavaghan, B. Rodrı́guez, and D. Kay. A Bidomain

model of the ventricular specialized conduction system of the heart. SIAM J. Appl.
Math., 72(5):1618–1643, 2012.

167
[73] A. Borzı̀ and V. Schulz. Computational optimization of systems governed by partial
differential equations. Society for Industrial and Applied Mathematics, Philadephia,
PA, 2011.
[74] M. Boulakia, M. A. Fernandez, J. F. Gerbeau, and N. Zemzemi. Direct and in-
verse problems in electrocardiography. AIP Conference Proceedings, 1048(1):113–
117, 2008.
[75] M. Boulakia, E. Schenone, and J-F. Gerbeau. Reduced-order modeling for cardiac
electrophysiology. application to parameter identification. Int. J. Numer. Meth.
Biomed. Engng., 28(6–7):727–744, 2012.
[76] Y. Bourgault, Y. Coudière, and C. Pierre. Existence and uniqueness of the solution
for the bidomain model used in cardiac electrophysiology. Nonlinear Analysis: Real
World Applications, 10(1):458–482, 2006.
[77] Y. Bourgault, M. Ethier, and V.G. LeBlanc. Simulation of electrophysiological waves
with an unstructured finite element method. ESAIM: Math. Modelling Numer.
Anal., 37(4):649–661, 2003.
[78] A. Bueno-Orovio, E.M. Cherry, and F.H. Fenton. Minimal model for human ven-
tricular action potentials in tissue. J Theor Biol, 3(253):544–560, 2008.
[79] E. Burman, S. Claus, P. Hansbo, M.G. Larson, and A. Massing. Cutfem: Discretiz-
ing geometry and partial differential equations. International Journal of Numerical
Methods for Engineering, 104(7):472501, 2015.
[80] E. Burman and M.A. Fernández. Stabilization of explicit coupling in fluidstructure
interaction involving fluid incompressibility. Comput. Methods Appl. Mech. Engrg.,
198:766–784, 2009.
[81] E. Burman and M.A. Fernández. An unfitted nitsche method for incompressible
fluidstructure interaction using overlapping meshes. Comput. Methods Appl. Mech.
Engrg., 279:497–514, 2014.
[82] I.C. Campbell, J. Ries, S.S. Dhawan, A.A. Quyyumi, W.R. Taylor, and J.N. Oshin-
ski. Effect of inlet velocity profiles on patient-specific computational fluid dynamics
simulations of the carotid bifurcation. J Biomech Eng, 134(5):051001, 2012.
[83] T.E. Carew, R.N. Vaishnav, and D.J. Patel. Compressibility of the arterial wall.
Circ Res, 23:61–68, 1968.
[84] J.C. Carr, W. Fright, and R.K. Beatson. Surface interpolation with radial basis
functions for medical imaging. IEEE Transactions on Medical Imaging, 16(1):96–
107, 1997.
[85] P. Causin, J.F. Gerbeau, and F. Nobile. Added-mass effect in the design of parti-
tioned algorithms for fluid-structure problems. Comput. Methods Appl. Mech. En-
grg., 194(42-44):4506–4527, 2005.

168
[86] I.B. Celik, U. Ghia, P.J. Roache, C.J. Freitas, and H. Coleman P.E. Raad. Procedure
for estimation and reporting of uncertainty due to discretization in cfd applications.
J Fluids Eng Trans ASME, 130(7):078001, 2008.

[87] R. Chabiniok, Moireau P., Lesault P.-F., A. Rahmouni, J.-F. Deux, and D. Chapelle.
Estimation of tissue contractility from cardiac cine-mri using a biomechanical heart
model. Biomech. Model. Mechanobiol., 11:609–630, 2012.

[88] R. Chabiniok, V.Y. Wang, M. Hadjicharalambous, L. Asner, J. Lee, M. Serme-

sant, E. Kuhl, A.A. Young, P. Moireau, M.P. Nash, D. Chapelle, and D. A. Nord-
sletten. Multiphysics and multiscale modelling, datamodel fusion and integration
of organ physiology in the clinic: ventricular cardiac mechanics. Interface Focus,
6(2):20150083, 2016.

[89] D. Chapelle, M. Fragu, V. Mallet, and P. Moireau. Fundamental principles of data

assimilation underlying the Verdandi library: applications to biophysical model per-
sonalization within euheart. Med. Biol. Eng. Comput., 51(11):1221–1233, 2013.

[90] D. Chapelle, A. Gariah, P. Moireau, and J. Sainte-Marie. A galerkin strategy with

proper orthogonal decomposition for parameter-dependent problems: Analysis, as-
sessments and applications to parameter estimation. ESAIM: Math. Modelling Nu-
mer. Anal., 47(6):1821–1843, 2013.

[91] J.J. Charonko, R. Kumar, K. Stewart, W.C. Little, and P.P. Vlachos. Vortices
formed on the mitral valve tips aid normal left ventricular filling. Annals of Biomed-
ical Engineering, 41(5):10491061, 2013.

[92] G. Chavent. Nonlinear Least Squares for Inverse Problems. Theoretical Foundations
and Step-by-Step Guide for Applications. Scientific Computation. Springer Nether-
lands, 2010.

[93] J Chen, X-Y Lu, and W Wang. Non-newtonian effects of blood flow on hemody-
namics in distal vascular graft anastomoses. J. Biomech., 39:1983–1995, 2006.

[94] P. Chen, A. Quarteroni, and G. Rozza. Simulation-based uncertainty quantification

of human arterial network hemodynamics. Int. J. Numer. Methods Biomed. Engng.,
29(6):698–721, 2013.

[95] P. Chen and C. Schwab. Sparse-grid, reduced-basis Bayesian inversion. Comput.

Methods Appl. Mech. Engrg., 297:84–115, 2015.

[96] L.K. Cheng, J.M. Bodley, and A. Pullan. Comparison of potential- and activation-
based formulations for the inverse problem of electrocardiology. IEEE Trans.
Biomed. Eng., 50(1):11–22, 2003.

[97] Y. Cheng, H. Oertel, and T. Schenkel. Fluid-structure coupled cfd simulation of

the left ventricular flow during filling phase. Annals of Biomedical Engineering,
5(33):567–576, 2005.

169
 [98] C. Cherubini, S. Filippi, P. Nardinocchi, and L. Teresi. An electromechanical model
of cardiac tissue: Constitutive issues and electrophysiological effects. Progress in
Biophysics and Molecular Biology, 2–3(97):562–573, 2008.

[99] S.C.P. Cheung, Kelvin K. L. Wong, Guan Heng Yeoh, William Yang, Jiyuan Tu,
Richard Beare, and Thanh Phan. Experimental and numerical study on the hemody-
namics of stenosed carotid bifurcation. Australasian Physical & Engineering Sciences
in Medicine, 33(4):319–328, 2010.

[100] P. Chinchapatnam, K.S. Rhode, M. Ginks, C.A. Rinaldi, P. Lambiase, R. Razavi,

S. Arridge, and M. Sermesant. Model-based imaging of cardiac apparent conductivity
and local conduction velocity for diagnosis and planning of therapy. IEEE Trans.
Med. Imaging, 27(11):1631–1642, 2008.

[101] J. Ching, J.L. Beck, and K.A. Porter. Bayesian state and parameter estimation of
uncertain dynamical systems. Probabilist. Eng. Mech., 21(1):81–96, 2006.

[102] C. Chnafa, S. Mendez, and F. Nicoud. Image-based large-eddy simulation in a

realistic left heart. Computers & Fluids, 94:173–187, 2014.

[103] Y.J. Choi, J. Constantino, V. Vedula, N. Trayanova, and R. Mittal. A new mri-based
model of heart function with coupled hemodynamics and application to normal and
diseased canine left ventricles. Front Bioeng Biotechnol, 3:140, 2015.

[104] A.J. Chorin. Numerical solution of the Navier–Stokes equations. Mathematics of

Computation, 22:745–762, 1968.

[105] J. Chung and G. M. Hulbert. A time integration algorithm for structural dynamics
with improved numerical dissipation: the generalized-α method. Trans. ASME J.
Appl. Mech., 60(2):371–375, 1993.

[106] P.G. Ciarlet. Mathematical Elasticity, Volume 1: Three Dimensional Elasticity.

Elsevier Science Publishers, North-Holland, 1988.

[107] P.G. Ciarlet and J. Necas. Unilateral problems in nonlinear, three-dimensional elas-
ticity. Archive for Rational Mechanics and Analysis, 87(4):319–338, 1985.

[108] R.H. Clayton, O. Bernus, E.M. Cherry, H. Dierckx, F.H. Fenton, L. Mirabella,
A.V. Panfilov, F.B. Sachse, G. Seemann, and H. Zhang. Models of cardiac tissue
electrophysiology: Progress, challenges and open questions. Progress in Biophysics
and Molecular Biology, 104:22–48, 2011.

[109] R.H. Clayton and A.V. Panfilov. A guide to modelling cardiac electrical activity
in anatomically detailed ventricles. Progress in Biophysics and Molecular Biology,
96(1-3):19–43, 2008.

170
[110] C.A. Cocosco, T. Netsch, J. Sénǵas, D. Bystrov, W.J. Niessen, and M.A. Viergever.
Automatic cardiac region-of-interest computation in cine 3d structural mri. Com-
puter Assisted Radiology and Surgery. Proceedings of the 18th International Congress
and Exhibition, 1268:1126–1131, 2004.
[111] R. Codina and S. Badia. On some pressure segregation methods of fractional-step
type for the finite element approximation of incompressible flow problems. Comput.
Methods Appl. Mech. Engrg., 195:2900–2918, 2006.
[112] A. Cohen and R. DeVore. Approximation of high-dimensional parametric pdes. Acta
Numerica, 24:1–159, 2015.
[113] C.M. Colciago, S. Deparis, and A. Quarteroni. Comparisons between reduced order
models and full 3d models for fluid-structure interaction problems in haemodynam-
ics. J. Comput. Appl. Math., 2754:120–138, 2014.
[114] P. Colli Franzone and L. Guerri. Spreading excitation in 3-d models of the anisotropic
cardiac tissue, I. validation of the Eikonal model. Math Biosci, 113:145–209, 1993.
[115] P. Colli Franzone, L. Guerri, and S. Rovida. Wavefront propagation in an acti-
vation model of the anisotropic cardiac tissue: asymptotic analysis and numerical
simulations. Journal of Mathematical Biology, 28(2):121–176, 1990.
[116] P. Colli Franzone, L. Guerri, C. Viganotti, and B. Taccardi. Finite element approxi-
mation of regularized solution of the inverse potential problem of electrocardiography
and applications to experimental data. Calcolo, 12(1):91–186, 1985.
[117] P. Colli Franzone and L.F. Pavarino. A parallel solver for reaction-diffusion systems
in computational electrocardiology. Mathematical Models and Methods in Applied
Sciences, 14(6):883–911, 2004.
[118] P. Colli Franzone, L.F. Pavarino, and S. Scacchi. Mathematical Cardiac Electrophys-
iology. Springer, 2014.
[119] P. Colli Franzone, L.F. Pavarino, and S. Scacchi. Bioelectrical effects of mechanical
feedbacks in a strongly coupled cardiac electro-mechanical model. Mathematical
Models and Methods in Applied Sciences, 26(1):27–57, 2016.
[120] P. Colli Franzone, L.F. Pavarino, and B. Taccardi. Simulating patterns of excitation,
repolarization and action potential duration with cardiac bidomain and monodomain
models. Math. Biosci., 197(1):35–66, 2005.
[121] P. Colli Franzone and G. Savaré. Degenerate evolution systems modeling the cardiac
electric field at micro- and macroscopic level. In A. Lorenzi and B. Ruf, editors,
Evolution Equations, Semigroups and Functional Analysis: In Memory of Brunello
Terreni, pages 49–78. Birkhäuser Basel, 2002.
[122] P. Colli Franzone, B. Taccardi, and C. Viganotti. An approach to inverse calculation
of epicardial potentials from body surface maps. Adv. Cardiol., 21:50–54, 1978.

171
[123] C. Corrado, J.-F. Gerbeau, and P. Moireau. Identification of weakly coupled mul-
tiphysics problems. application to the inverse problem of electrocardiography. J.
Comput. Phys., 283:271 – 298, 2015.

[124] K.D. Costa, J.W. Holmes, and A.D. Mcculloch. Modelling cardiac mechanical
properties in three dimensions. American Journal of PhysiologyPHILOSOPHICAL
TRANSACTIONS of the Royal society A, 359(6):1233–1250, 2001.

[125] P. Crosetto, S. Deparis, G. Fourestey, and A. Quarteroni. Parallel algorithms for

fluid-structure interaction problems in haemodynamics. SIAM J. Sci. Comput.,
33:1598–1622, 2011.

[126] T. Cui, Y. Marzouk, and K. Willcox. Data-driven model reduction for the Bayesian
solution of inverse problems. Int. J. Numer. Methods Engng, 102(5):966–990, 2015.

[127] B. Dacorogna. Direct Methods in the Calculus of Variations - Applied Mathematical

Science (vol 78). Springer, New York, 2000.

[128] C. D’Angelo. Multiscale modelling of metabolism and transport phenomena in living

tissues. PhD thesis, Ecole Polytechnique Fédérale de Lausanne, 2007.

[129] J De Hart, FPT Baaijens, GWM Peters, and PJG Schreurs. A computational fluid-
structure interaction analysis of a fiber-reinforced stentless aortic valve. Journal of
biomechanics, 36(5):699–712, 2003.

[130] J. Degroote. On the similarity between dirichletneumann with interface artificial

compressibility and robinneumann schemes for the solution of fluid-structure inter-
action problems. J. Comput. Physics, 230:6399640363996403, 2011.

[131] J. Degroote, A. Swillens, P. Bruggeman, Haelterman, P. Segers, and J. Vierendeels.

Simulation of fluidstructure interaction with the interface artificial compressibility
method. International Journal for Numerical Methods in Biomedical Engineering,
26(3-4):276–289, 2010.

[132] J. Degroote and J. Vierendeels. Multi-solver algorithms for the partitioned sim-
ulation of fluidstructure interaction. Comput. Methods Appl. Mech. Engrg., 25–
28(200):21952210, 2011.

[133] M. D’Elia, M. Perego, and A. Veneziani. A variational data assimilation procedure

for the incompressible navier-stokes equations in hemodynamics. J. Sc. Comp.,
52(2):340–359, 2012.

[134] M. D’Elia and A. Veneziani. Uncertainty quantification for data assimilation in

a steady incompressible navier-stokes problem. ESAIM: Math. Modelling Numer.
Anal.., 47(04):1037–1057, 2013.

172
[135] H. Delingette, F. Billet, K.C. Wong, M. Sermesant, K. Rhode, M. Ginks, C.A Ri-
naldi, R. Razavi, and N. Ayache. Personalization of cardiac motion and contractil-
ity from images using variational data assimilation. IEEE Trans. Biomed. Engng.,
59(1):20–24, 2012.

[136] S. Deparis. Numerical analysis of axisymmetric flows and methods for fluid-structure
interaction arising in blood flow simulation. PhD thesis, École Polytechnique
Fédérale de Lausanne, 2004.

[137] S. Deparis, M. Discacciati, G. Fourestey, and A. Quarteroni. Fluid-structure algo-

rithms based on Steklov-Poincaré operators. Comput. Methods Appl. Mech. Engrg.,
195(41-43):5797–5812, 2006.

[138] S. Deparis, D. Forti, G. Grandperrin, and A. Quarteroni. Facsi: A block parallel

preconditioner for fluid-structure interaction in hemodynamics. J. Comput. Physics,
(327):700–718, 2016.

[139] S. Deparis, G. Grandperrin, and A. Quarteroni. Parallel preconditioners for the

unsteady Navier–Stokes equations and applications to hemodynamics simulations.
Computer & Fluids, 92:253–273, 2014.

[140] D. Di Francesco and D. Noble. A model of cardiac electrical activity incorporating

ionic pumps and concentration changes. Phil. Trans. R. Soc. B, 307(1133):353–398,
1985.

[141] J. Dick, F. Kuo, and I. Sloan. High-dimensional integration: the quasi-Monte Carlo
way. Acta Numerica, 22:133–288, 2013.

[142] M. Dihlmann and B. Haasdonk. A reduced basis Kalman filter for parametrized
partial differential equations. SAIM: Control Optim. Calc. Var., 22(3):625–669,
2016.

[143] F.-X. Dimet and O. Talagrand. Variational algorithms for analysis and assimilation
of meteorological observations: theoretical aspects. Tellus A, 38A(2):97–110, 1986.

[144] H. Do, A. A. Owida, W. Yang, and Y.S. Morsi. Numerical simulation of the haemo-
dynamics in end-to-side anastomoses. Int. J. Numer. Meth. Fluids, 67(5):638–650,
2011.

[145] C.R. Dohrmann and O.B. Widlund. An overlapping schwarz algorithm for almost
incompressible elasticity. SIAM J. Num. Anal., 4(47):8811–8823, 2009.

[146] S. Dokos, B.H. Smaill, A.A. Young, and I.J. LeGrice. Shear properties of passive ven-
tricular myocardium. American Journal of Physiology, 283(6):H2650–H2659, 2002.

[147] W. P. Donders, W. Huberts, F. N. van de Vosse, and T. Delhaas. Personalization

of models with many model parameters: an efficient sensitivity analysis approach.
Int. J. Numer. Methods Biomed. Engng., 31(10):e02727, 2015.

173
[148] J. Donea. An arbitrary Lagrangian-Eulerian finite element method for transient
dynamic fluid-structure interaction. Comput. Methods Appl. Mech. Engrg., 33:689–
723, 1982.
[149] O. Dur, S.T. Coskun, K.O. Coskun, D. Frakes, L.B. Kara, and K. Pekkan. Computer-
aided patient-specific coronary artery graft design improvements using CFD coupled
shape optimizer. Cardiovasc. Engr. Tech., pages 1–13, 2011.
[150] D. Durrer, R.R. van Dam, G.E. Freud, M.J. Janse, F.L. Meijler, and R.C.
Arzbaecher. Total excitation of the isolated human heart. Circulation, 41(6):899–
912, 1970.
[151] V.G. Eck, W.P. Donders, J. Sturdy, J. Feinberg, T. Delhaas, L.R. Hellevik, and
W. Huberts. A guide to uncertainty quantification and sensitivity analysis for car-
diovascular applications. Int. J. Numer. Methods Biomed. Engng., page in press,
2015.
[152] D.R. Einstein, K.S. Kunzelman, P.G. Reinhall, M.A. Nicosia, and R.P. Cochran.
The effects of cellular contraction on aortic valve leaflet flexural stiffness. J Heart
Valve Disease, 14:376–385, 2005.
[153] C. Eitel, G. Hindricks, N. Dagres, P. Sommer, and C. Piorkowski. Ensite velocityT M
cardiac mapping system: a new platform for 3d mapping of cardiac arrhythmias.
Expert Rev Med Devices, 7(2):185–192, 2010.
[154] H. Elman, D. Silvester, and A. Wathen. Finite Elements and Fast Iterative Solvers.
Oxford Science Publications, 2005.
[155] H.C. Elman and D.J. Silvester. Fast nonsymmetric iterations and preconditioning
for navierstokes equations. SIAM J. Sci. Comp., 17(1):33–46, 1996.
[156] G. Enden and A. Popel. A numerical study of the shape of the surface separating
flow into branches in microvascular bifurcations. J Biomech Eng, 114:398405., 1992.
[157] T.S.E. Eriksson, A.J. Prassl, G. Plank, and G.A. Holzapfel. Modeling the dispersion
in electromechanically coupled myocardium. International Journal for Numerical
Methods in Biomedical Engineering, 29(11):1267–1284, 2013.
[158] O.G. Ernst, B. Sprungk, and H.-J. Starkloff. Bayesian inverse problems and Kalman
filters. In Extraction of Quantifiable Information from Complex Systems, volume 102
of Lecture Notes in Computational Science and Engineering, pages 133–159. Springer
International Publishing, 2014.
[159] V.J. Ervin and H. Lee. Numerical approximation of a quasi-newtonian stokes
flow problem with defective boundary conditions. SIAM J. Numer. Anal.,
45(5):21202140, 2007.
[160] C.R. Ethier, D.A. Steinman, X. Zhang, S.R. Karpik, and M. Ojha. Flow waveform
effects on end-to-side anastomotic flow patterns. J. Biomech., 31(7):609–617, 1998.

174
[161] M. Ethier and Y. Bourgault. Semi-implicit time-discretization schemes for the bido-
main model. SIAM J. Numer. Anal., 5(46):2443–2468, 2008.

[162] L. Euler. Principia pro motu sanguinis per arterias determinando. Opera posthuma
mathematica et physica anno 1844 detecta. Ediderunt P.H. Fuss et N. Fuss Petropoli;
Apund Eggers et Socios., 1:814–823, 1775.

[163] G. Evensen. Sequential data assimilation with a nonlinear quasi-geostrophic model

using Monte Carlo methods to forecast error statistics. J. Geophys. Res., 9:10143–
10162, 1994.

[164] G. Evensen. The Ensemble Kalman Filter: theoretical formulation and practical
implementation. Ocean Dynamics, 53:343–367, 2003.

[165] G. Evensen. The ensemble Kalman filter for combined state and parameter estima-
tion. IEEE Control Syst. Mag., 29(3):83–104, 2009.

[166] E. Faggiano, A. Antiga, G. Puppini, A. Quarteroni, G.B. Luciani, and C Ver-

gara. Helical flows and asymmetry of blood jet in dilated ascending aorta with
normally functioning bicuspid valve. Biomechanics and Modeling in Mechanobiol-
ogy, 4(12):801–813, 2013.

[167] E. Faggiano, T. Lorenzi, and A. Quarteroni. Metal artefact reduction in computed

tomography images by a fourth-order total variation flow. Computer Methods in
Biomechanics and Biomedical Engineering: Imaging & Visualization, 3-4(4):202–
213, 2014.

[168] C. Farhat and F.X. Roux. A method of finite element tearing and interconnecting
and its parallel solution algorithm. International Journal for Numerical Methods in
Engineering, 32(6):1205–1227, 1991.

[169] A. Fasano, R.F. Santos, and A. Sequeira. Blood coagulation: A puzzle for biolo-
gists, a maze for mathematicians. In Modeling of Physiological Flows, edited by D.
Ambrosi, A. Quarteroni, G. Rozza, Chapter 3, pages 41–75. Springer, 2012.

[170] M. Fedele, E. Faggiano, L. Barbarotta, F. Cremonesi, L. Formaggia, and S. Perotto.

Semi-automatic three-dimensional vessel segmentation using a connected component
localization of the region-scalable fitting energy. In 2015 9th International Sympo-
sium on Image and Signal Processing and Analysis (ISPA), pages 72–77. IEEE,
2015.

[171] M. Fedele, E. Faggiano, L Dedè, and A. Quarteroni. A patient-specific aortic valve

model based on moving resistive immersed implicit surfaces. MOX-Report n. 23-
2016, Department of Mathematics, Politecnico di Milano, Italy, 2016.

[172] F. Fenton and A. Karma. Vortex dynamics in three-dimensional continuous my-

ocardium with fiber rotation: Filament instability and fibrillation. Chaos, 8(1),
1998.

175
[173] M.A. Fernández. Incremental displacement-correction schemes for incompressible
fluid-structure interaction - stability and convergence analysis. Numerische Mathe-
matik, 123(1):21–65, 2013.

[174] M.A. Fernández, J.F. Gerbeau, and C. Grandmont. A projection semi-implicit

scheme for the coupling of an elastic structure with an incompressible fluid. Int.
J. Num. Methods Engrg., 69(4):794–821, 2007.

[175] M.A. Fernàndez, V. Milisic, and A. Quarteroni. Analysis of a geometrical multiscale

blood flow model based on the coupling of odes and hyperbolic pdes. Multiscale
Model Simul, 4(1):215–236, 2005.

[176] M.A. Fernández and M. Moubachir. A Newton method using exact Jacobians for
solving fluid-structure coupling. Computers & Structures, 83(2-3):127–142, 2005.

[177] M.A. Fernandez and N. Zemzemi. Decoupled time-marching schemes in computa-

tional cardiac electrophysiology and ecg numerical simulation. Mathematical Bio-
sciences, 226(1):58–75, 2010.

[178] C.A. Figueroa, I.E. Vignon-Clementel, K.E. Jansen, T.J.R Hughes, and C.A. Tay-
lor. A coupled momentum method for modeling blood flow in three-dimensional
deformable arteries. Computer Methods in Applied Mechanics and Engineering,
195(41–43):5685–5706, 2006.

[179] L. Fin and R. Grebe. Three dimensional modeling of the cerebrospinal fluid dy-
namics and brain interactions in the aqueduct of sylvius. Computer Methods in
Biomechanics and Biomedical Engineering, 3(6):163–170, 2003.

[180] P.F. Fischer, F. Loth, S.E. Lee, S.W. Lee, D.S. Smith, and H.S. Bassiouny. Simula-
tion of high-Reynolds number vascular flows. Comput. Methods Appl. Mech. Engrg.,
196:3049–3060, 2007.

[181] G. Fishman. Monte Carlo. Concepts, Algorithms, and Applications. Springer Series
in Operations Research and Financial Engineering. Springer-Verlag New York, 1996.

[182] R. FitzHugh. Impulses and physiological states in theoretical models of nerve mem-
brane. Biophys J, 1(6):445–466, 1961.

[183] L. Formaggia, J.F. Gerbeau, F. Nobile, and A. Quarteroni. On the coupling of

3D an 1D Navier-Stokes equations for flow problems in compliant vessels. Comput.
Methods Appl. Mech. Engrg., 191(6-7):561–582, 2001.

[184] L. Formaggia, J.F. Gerbeau, F. Nobile, and A. Quarteroni. Numerical treatment

of defective boundary conditions for the Navier-Stokes equation. SIAM J. Numer.
Anal., 40(1):376–401, 2002.

[185] L. Formaggia, D. Lamponi, and A. Quarteroni. One-dimensional models for blood

flow in arteries. Journal of Engineering Mathematics, 47(3-4):251–276, 2003.

176
[186] L. Formaggia, A. Moura, and F. Nobile. On the stability of the coupling of 3d and 1d
fluid-structure interaction models for blood flow simulations. M2AN Math. Model.
Numer. Anal., 41(4):743–769, 2007.
[187] L. Formaggia, F. Nobile, A. Quarteroni, and A. Veneziani. Multiscale modelling
of the circulatory system: a preliminary analysis. Comput and Visual in Science,
2:75–83, 1999.
[188] L. Formaggia, A. Quarteroni, and A. Veneziani (Eds.). Cardiovascular Mathematics
- Modeling and simulation of the circulatory system. Springer-Verlag Milan, 2009.
[189] L. Formaggia, A. Quarteroni, and C. Vergara. On the physical consistency be-
tween three-dimensional and one-dimensional models in haemodynamics. J. Com-
put. Physics, 244:97–112, 2013.
[190] L. Formaggia, A. Veneziani, and C. Vergara. A new approach to numerical solution
of defective boundary value problems in incompressible fluid dynamics. SIAM J.
Numer. Anal., 46(6):2769–2794, 2008.
[191] L. Formaggia, A. Veneziani, and C. Vergara. Flow rate boundary problems for
an incompressible fluid in deformable domains: formulations and solution methods.
Comput. Methods Appl. Mech. Engrg., 199 (9-12):677–688, 2009.
[192] L. Formaggia and C. Vergara. Prescription of general defective boundary conditions
in fluid-dynamics. Milan Journal of Mathematics, 80(2):333–350, 2012.
[193] M. Fornefett, K. Rohr, and H.S. Stiehl. Radial basis functions with compact support
for elastic registration of medical images. Image and Vision Computing, 19(1–2):87–
96, 2001.
[194] C. Forster, W. Wall, and E. Ramm. Artificial added mass instabilities in sequential
staggered coupling of nonlinear structures and incompressible viscous flow. Comput.
Methods Appl. Mech. Engrg., 196(7):1278–1293, 2007.
[195] D. Forti and L. Dedé. Semi-implicit bdf time discretization of the navier-stokes equa-
tions with vms-les modeling in a high performance computing framework. Computer
& Fluids, 117:168–182, 2015.
[196] A.F. Frangi, W.J. Niessen, R.M. Hoogeveen, T. Van Walsum, and M.A. Viergever.
Model-based quantitation of 3-d magnetic resonance angiographic images. IEEE
Transactions on Medical Imaging, 18(10):946–956, 1999.
[197] D.W. Frazier, W. Krassowska, P.S. Chen, P.D. Wolf, N.D. Danieley, W.M. Smith,
and R.E. Ideker. Transmural activations and stimulus potentials in three-dimensional
anisotropic canine myocardium. Circ Res, 63(1):135–146, 1988.
[198] T. Fritz, C. Wieners, G. Seemann, H. Steen, and O. Dossel. Simulation of the
contraction of the ventricles in a human heart model including atria and pericardium.
Biomechanics and Modeling in Mechanobiology, 13(3):627–641, 2014.

177
[199] Y. Fung. Biomechanics: mechanical properties of living tissues. Springer, 1993.

[200] Y. C. Fung, K. Fronek, and P. Patitucci. Pseudoelasticity of arteries and the choice
of its mathematical expression. American Journal of Physiology, 237(5):H620–H631,
1979.

[201] K. Galvin and H. Lee. Analysis and approximation of the cross model for
quasi-newtonian flows with defective boundary conditions. Appl. Math. Comp.,
222:244254, 2013.

[202] K. Galvin, H. Lee, and L.G. Rebholz. Approximation of viscoelastic flows with
defective boundary conditions. J. Non Newt. Fl. Mech., 169-170:104113, 2012.

[203] M.W. Gee, U. Kuttler, and W.A. Wall. Truly monolithic algebraic multigrid for
fluid-structure interaction. Int. J. Num. Methods Engrg., 85(8):987–1016, 2011.

[204] B. Geiger. Three-dimensional modeling of human organs and its application to

diagnosis and surgical planning. Technical Report RR-2105, INRIA, 1993.

[205] S.E. Geneser, R.M. Kirby, and R.S. MacLeod. Application of stochastic finite ele-
ment methods to study the sensitivity of ecg forward modeling to organ conductivity.
IEEE Trans. Biomed. Engng., 55(1):31–40, 2008.

[206] L. Gerardo Giorda, L. Mirabella, F. Nobile, M. Perego, and A. Veneziani. A model-

based block-triangular preconditioner for the bidomain system in electrocardiology.
Journal of Computational Physics, 228(10):36253639, 2009.

[207] L. Gerardo Giorda, F. Nobile, and C. Vergara. Analysis and optimization of robin-
robin partitioned procedures in fluid-structure interaction problems. SIAM J. Nu-
mer. Anal., 48(6):2091–2116, 2010.

[208] J.-F. Gerbeau, D. Lombardi, and E. Schenone. Reduced order model in cardiac
electrophysiology with approximated lax pairs. Adv. Comput. Math., 41(5):1103–
1130, 2015.

[209] P. Gervasio, F. Saleri, and A. Veneziani. Algebraic fractional-step schemes with spec-
tral methods for the incompressible Navier-Stokes equations. J. Comput. Physics,
214(1):347–365, 2006.

[210] R. Ghanem and P. D. Spanos. Stochastic Finite Elements: A Spectral Ap- proach,
revised edition. Dover, 2003. Reprint of the Springer-Verlag, New York, 1991 edition.

[211] G. Gigante and C. Vergara. Analysis and optimization of the generalized schwarz
method for elliptic problems with application to fluid-structure interaction. Numer.
Math., 131(2):369–404, 2015.

[212] M.B. Giles. Multilevel Monte Carlo methods. Acta Numerica, 24:259–328, 2015.

178
[213] S. Giordana, S.J. Sherwin, J. Peiró, D.J. Doorly, J.S. Crane, K.E. Lee, N.J.W.
Cheshire, and C.G. Caro. Local and global geometric influence on steady flow in
distal anastomoses of peripheral bypass grafts. J. Biomech. Engr., 127:1087, 2005.

[214] V. Girault and P.A. Raviart. Finite element methods for Navier-Stokes equations.
Springer–Verlag, 1986.

[215] S. Glagov, C. Zarins, D.P. Giddens, and D.N. Ku. Hemodynamics and atheroscle-
rosis. insights and perspectives gained from studies of human arteries. Archives of
Pathology & Laboratory Medicine, 112(10):1018–1031, 1988.

[216] R. Glowinski, T.W. Pan, and J. Periaux. A lagrange multiplier/fictitious domain

method for the numerical simulation of incompressible viscous flow around moving
rigid bodies: (i) case where the rigid body motions are known a priori. Comptes
Rendus de l’Acadmie des Sciences - Series I - Mathematics, 324(3):361–369, 1997.

[217] R. Gnyaneshwar, R.K. Kumar, and K.R. Balakrishnan. Dynamic analysis of the
aortic valve using a finite element model. Ann Thorac Surg, 73(4):1122–1129, 2002.

[218] S. Goktepe and E. Kuhl. Electromechanics of the heart: a unified approach to

the strongly coupled excitationcontraction problem. Computational Mechanics,
45(2):227–243, 2010.

[219] A. Goshtasby and D. Turner. Segmentation of cardiac cine mr images for extraction
of right and left ventricular chambers. IEEE Transactions on Medical Imaging,
14(1):56–64, 1995.

[220] C. Grandmont. Analyse Mathematique et Numerique de Quelques Problemes d In-

teraction Fluide-Structure. PhD thesis, Laboratoire d Analyse Numrique de Paris
VI, 1998.

[221] B.E. Griffith, R.D. Hornung, D.M. McQueen, and C.S. Peskin. An adaptive, for-
mally second order accurate version of the immersed boundary method. Journal of
Computational Physics, 223(1):10–419, 2007.

[222] B.E. Griffith, X. Luo, D.M. McQueen, and C.S. Peskin. Simulating the fluid dy-
namics of natural and prosthetic heart valves using the immersed boundary method.
International Journal of Applied Mechanics, 1:137–176, 2009.

[223] L. Grinberg, E. Cheever, T. Anor, J.R. Madsen, and G.E. Karniadakis. Model-
ing blood flow circulation in intracranial arterial networks: a comparative 3d/1d
simulation study. Annals of Biomed Eng, 39(1):297–309, 2010.

[224] L. Grinberg, A. Yakhot, and G.E. Karniadakis. Analyzing Transient Turbulence

in a Stenosed Carotid Artery by Proper Orthogonal Decomposition. Annals of
Biomedical Engineering, 37(11):2200–2217, 2009.

179
[225] J. M. Guccione, A. D. McCulloch, and L. K. Waldman. Passive material properties
of intact ventricular myocardium determined from a cylindrical model. Journal of
Biomechanical Engineering, 113(1):42–55, 1991.

[226] B. Guerciotti, C. Vergara, L. Azzimonti, L. Forzenigo, A. Buora, P. Biondetti, and

M. Domanin. Computational study of the fluid-dynamics in carotids before and
after endarterectomy. J. Biomech., 195:2088–2099, 2015.

[227] B. Guerciotti, C. Vergara, S. Ippolito, A. Quarteroni, C. Antona, and R. Scrofani.

Computational study of the risk of restenosis in coronary bypasses. Biomechanics
and Modeling in Mechanobiology, DOI:10.1007/s10237-016-0818-x, 2016.

[228] J.L. Guermond, P. Minev, and J. Shen. Error analysis of pressure-correction schemes
for the time-dependent stokes equations with open boundary conditions. SIAM J.
Numer. Anal., 43(1):239–258, 2005.

[229] J.L. Guermond, P. Minev, and J. Shen. An overview of projection methods for
incompressible flows. Comput. Methods Appl. Mech. Engrg., 195:6011–6045, 2006.

[230] J.L. Guermond and L. Quartapelle. On the approximation of the unsteady Navier-
Stokes equations by finite element projection methods. Numer. Math., 80:207–238,
1998.

[231] J.L. Guermond and J. Shen. Velocity-correction projection methods for incompress-
ible flows. SIAM J. Numer. Anal., 41:112–134, 2003.

[232] J.L. Guermond and J. Shen. An overview of projection methods for incompressible
flows. Comput. Methods Appl. Mech. Engrg., 195:6011–6045, 2006.

[233] G. Guidoboni, R. Glowinski, N. Cavallini, and S. Canic. Stable loosely-coupled-

type algorithm for fluidstructure interaction in blood flow. J. Comput. Physics,
228:69166937, 2009.

[234] O. Gultekin, G. Sommer, and G.A. Holzapfel. An orthotropic viscoelastic model

for the passive myocardium: continuum basis and numerical treatment. Computer
Methods in Biomechanics and Biomedical Engineering, 19(15):1647–1664, 2016.

[235] M.D. Gunzburger. Perspectives in Flow Control and Optimization. Advances in

Design and Control. Society for Industrial and Applied Mathematics, Philadephia,
PA, 2003.

[236] M.D. Gunzburger, C.G. Webster, and G. Zhang. Stochastic finite element methods
for partial differential equations with random input data. Acta Numerica, 23:521–
650, 2014.

[237] A. Gupta, L. von Kurowski, A. Singh, D. Geiger, C.C. Liang, M.Y. Chiu, L. Adler,
M. Haacke, and D. Wilson. Cardiac mr image segmentation using deformable models.
Computers in Cardiology, Proceedings, pages 747–750, 1993.

180
[238] V. Gurev, P. Pathmanathan, J.L. Fattebert, H.F. Wen, J. Magerlein, R.A. Gray,
D.F. Richards, and J.J. Rice. A high-resolution computational model of the deform-
ing human heart. Biomechanics and Modeling in Mechanobiology, 14(4):132–140,
2015.
[239] C.M. Haggerty, L. Mirabella, M. Restrepo, D. A de Zélicourt, J. Rossignac,
F. Sotiropoulos, T.L. Spray, K.R. Kanter, M.A. Fogel, and A.P. Yoganathan.
Patient-specific surgery planning for the fontan procedure. In Computer Models
in Biomechanics, pages 217–228. Springer, 2013.
[240] A. Hansbo and P. Hansbo. An unfitted finite element method, based on nitsche’s
method, for elliptic interface problems. Comput. Methods Appl. Mech. Engrg.,
191(47–48):5537–5552, 2002.
[241] A. Hansbo, P. Hansbo, and M.G. Larson. An unfitted finite element method, based
on nitsche’s method, for elliptic interface problems. ESAIM: Math. Modelling Nu-
mer. Anal., 37(3):495 – 514, 2003.
[242] I. Hariton, G. de Botton, T.C. Gasser, and G.A. Holzapfel. Stress-driven collagen
fiber remodeling in arterial walls. Biomechanics and Modeling in Mechanobiology,
3(6):163–175, 2006.
[243] J. De Hart, G.W.M. Peters, P.J.G. Schreurs, and F.P.T. Baaijens. A two-dimensional
fluidstructure interaction model of the aortic value. Journal of biomechanics,
33(9):10791088, 2000.
[244] H. Haruguchi and S. Teraoka. Intimal hyperplasia and hemodynamic factors in
arterial bypass and arteriovenous grafts: a review. J. Artif. Organs, 6(4):227–235,
2003.
[245] X. He, D.N. Ku, and J.E. Moore Jr. Simple calculation of the velocity profiles for
pulsatile flow in a blood vessel using mathematica. Ann Biomed Eng, 21:45–49,
1993.
[246] M. Heil. An efficient solver for the fully coupled solution of large-displacement fluid-
structure interaction problems. Comput. Methods Appl. Mech. Engrg., 193:1–23,
2004.
[247] J.S. Hesthaven, G. Rozza, and B. Stamm. Certified Reduced Basis Methods
for Parametrized Partial Differential Equations. SpringerBriefs in Mathematics.
Springer, 2016.
[248] J.G. Heywood, R. Rannacher, and S. Turek. Artificial boundaries and flux and
pressure conditions for the incompressible Navier-Stokes equations. Int. J. Num.
Methods Fluids, 22:325–352, 1996.
[249] D.W. Hilgemann and D. Noble. Excitation-contraction coupling and extracellular
calcium transients in rabbit atrium: reconstruction of basic cellular mechanisms.
Proc R soc Lond B, 230:163–205, 1987.

181
[250] B. Hillen, H.W. Hoogstraten, and L. Post. A wave propagation model of blood flow in
large vessels using an approximate velocity profile function. J Biomech, 19:187–194,
1986.

[251] M. Hintermüller, K. Kunisch, Y. Spasov, and S. Volkwein. Dynamical systems-based

optimal control of incompressible fluids. Int. J. Numer. Meth. Fluids, 46(4):345–359,
2004.

[252] M. Hinze, R. Pinnau, M. Ulbrich, and S. Ulbrich. Optimization with PDE Con-
straints, volume 23 of Mathematical Modelling: Theory and Applications. Springer,
2009.

[253] C.W. Hirt, A.A. Amsden, and J.L. Cook. An arbitrary lagrangian eulerian comput-
ing method for all flow speeds. J. Comput. Physics, 69:277–324, 1974.

[254] A.L. Hodgkin and A.F. Hukley. A quantitative description of membrane current
and its application to conduction and excitation in nerve. J Physiol, 117(4):500–
544, 1952.

[255] G.A. Holzapfel. Nonlinear Solid Mechanics: a ContinuumApproach for Engineering.

John Wiley & Sons Ltd, England, 2000.

[256] G.A. Holzapfel and T.C. Gasser. A viscoelastic model for fiber-reinforced composites
at finite strains: continuumbasis, computational aspects, and applications. Comput.
Methods Appl. Mech. Engrg., 190:4379–4403, 2001.

[257] G.A. Holzapfel, T.C. Gasser, and R.W. Ogden. A new constitutive framework for
arterial wall mechanics and a comparative study of material models. J Elast, 61:1–48,
2000.

[258] G.A. Holzapfel and R.W. Ogden. Constitutive modelling of passive myocardium:
a structurally based framework for material characterization. PHILOSOPHICAL
TRANSACTIONS of the Royal society A, 367:3445–3475, 2009.

[259] G.A. Holzapfel and R.W. Ogden. Constitutive modelling of arteries. Proc. R. Soc.
Lond. Ser. A Math. Phys. Eng. Sci., 466(2118):1551–1596, 2010.

[260] E. Hopf. Uber die aufangswertaufgabe für die hydrodynamischen grundgliechungen.

Mathematische Nachrichten, 4:213–231, 1951.

[261] I. Hoteit, D.T. Pham, and J. Blum. A simplified reduced order kalman filtering
and application to altimetric data assimilation in tropical pacific. J. Mar. Syst.,
36(1–2):101–127, 2002.

[262] M.C. Hsu, D. Kamensky, Y. Bazilevs, M.S. Sacks, and T.J.R. Hughes. Fluidstruc-
ture interaction analysis of bioprosthetic heart valves: significance of arterial wall
deformation. Comp Mech, 54(4):10551071, 2014.

182
[263] W. Huberts, C. de Jonge, W.P.M. van der Linden, M. Inda, K. Passera, J.H.M. Tor-
doir, F.N. van de Vosse, and E.M.H Bosboom. A sensitivity analysis of a personalized
pulse wave propagation model for arteriovenous fistula surgery. part b: identification
of possible generic model parameters. Med. Eng. Phys., 35(6):827–837, 2013.

[264] W. Huberts, C. de Jonge, W.P.M. van der Linden, M. Inda, J.H.M. Tordoir, F.N.
van de Vosse, and E.M.H Bosboom. A sensitivity analysis of a personalized pulse
wave propagation model for arteriovenous fistula surgery. part a: identification of
most influential model parameters. Med. Eng. Phys., 35(6):810–826, 2013.

[265] T.J.R. Hughes. A Study of the One-Dimensional Theory of Arterial Pulse Propaga-
tion. PhD thesis, University of California, Berkeley, 1974.

[266] T.J.R. Hughes. Multiscale phenomena: Green’s function, the Dirichlet-to-Neumann

formulation, subgrid scale models, bubbles and the origins of stabilized formulations.
Comput. Methods Appl. Mech. Engrg., 127:387–401, 1995.

[267] T.J.R. Hughes and J. Lubliner. On the one-dimensional theory of blood flow in the
larger vessels. Mathematical Biosciences, 18(1-2):161–170, 1973.

[268] T.J.R. Hughes, L. Mazzei, and K.E. Jansen. Large eddy simulation and the vari-
ational multiscale method. Computing and Visualization in Science, 3(1):47–59,
2000.

[269] G. Huiskamp. Simulation of depolarization in a membrane- equations-based model of

the anisotropic ventricle. IEEE Transaction on Biomedical Engineering, 45(7):847–
855, 1998.

[270] J. Humpherys, P. Redd, and J. West. A fresh look at the kalman filter. SIAM
Review, 54(4):801–823, 2012.

[271] J.D. Humphrey and F.C.P. Yin. On constitutive relations and finite deformations of
passive cardiac tissue: I. a pseudostrain-energy function. Journal of Biomechanical
Engineering, 109(4):298–304, 1987.

[272] P.J. Hunter, M.P. Nash, and G.B. Sands. Computational electromechanics of the
heart. Computational biology of the heart (eds. Panfilov, Holden), pages 345–407,
1997.

[273] Paul A Iaizzo. Handbook of cardiac anatomy, physiology, and devices. Springer,
2009.

[274] K. Ide, P. Courtier, M. Ghil, and A. C. Lorenc. Unified notation for data assimilation:
Operational, sequential and variational. J. Met. Soc. Japan, 75(1B):181–189, 1997.

[275] M.A Iglesias, K.J.H. Law, and A.M Stuart. Ensemble Kalman methods for inverse
problems. Inverse Problems, 29(4):045001, 2013.

183
[276] T. Ijiri, T. Ashihara, T. Yamaguchi, K. Takayama, T. Igarashi, T. Shimada,
T. Namba, R. Haraguchi, and K. Nakazawa. A procedural method for modeling
the purkinje fibers of the heart. J Physiol Sci, 58(7):90–100, 2008.

[277] J.G. Isaksen, Y. Bazilevs, T. Kvamsdal, Y. Zhang, J.H. Kaspersen, K. Waterloo,

B. Romner, and T. Ingebrigtsen. Determination of wall tension in cerebral artery
aneurysms by numerical simulation. Stroke, 39:3172–3178, 2008.

[278] R.H. Johnstone, E.T. Chang, R. Bardenet, T.P. de Boer, D.J. Gavaghan, P. Path-
manathan, R.H. Clayton, and G.R. Mirams. Uncertainty and variability in models
of the cardiac action potential: Can we build trustworthy models? J. Mol. Cell.
Cardiol., 96:49–62, 2016.

[279] S. Julier and J. Uhlmann. Unscented filtering and nonlinear estimation. Proceedings
of the IEEE, 92(3):401?422, 2004.

[280] S. Julier, J. Uhlmann, and H. Durrant-Whyte. A new method for the nonlinear
transformation of means and covariance in filters and estimators. IEEE Trans. on
Automatic Control, 45(3):477–482, 2000.

[281] S.J. Julier, J.K. Uhlmann, and H.F. Durrant-Whyte. A new approach for filter-
ing nonlinear systems. In American Control Conference, Proceedings of the 1995,
volume 3, pages 1628–1632, 1995.

[282] M. Juntunen and R. Stenberg. Nitsche’s method for general boundary conditions.
Math. Comp., 78:1353–1374, 2009.

[283] J. Kaipio and E. Somersalo. Statistical and computational inverse problems, volume
160 of Applied Mathematical Sciences. Springer Science+Business Media, Inc., 2005.

[284] R.E. Kalman. A new approach to linear filtering and prediction problems. ASME.
J. Basic Eng., 82(1):35–45, 1960.

[285] G.E. Karniadakis, M. Israeli, and S.E. Orszag. High order splitting methods for the
incompressible Navier-Stokes equations. J. Comput. Physics, 59:414–443, 1991.

[286] J.P. Keener. An eikonal-curvature equation for action potential propagation in my-
ocardium. J Math Biol, 29(7):629–651, 1991.

[287] J.P. Keener and K. Bogar. A numerical method for the solution of the bidomain
equations in cardiac tissue. Chaos, 8:234–241, 1998.

[288] S. Kefayati, D.W. Holdsworth, and T.L. Poepping. Turbulence intensity measure-
ments using particle image velocimetry in diseased carotid artery models: Effect
of stenosis severity, plaque eccentricity, and ulceration. Journal of Biomechanics,
47:253–263, 2014.

184
[289] R.H. Keldermann, M.P. Nash, and A.V. Panfilov. Modeling cardiac mechano-
electrical feedback using reaction-diffusion-mechanics systems. Physica D: Nonlinear
Phenomena, 238(11-12):1000–1007, 2009.

[290] C. Kelley. Iterative Methods for Optimization. Society for Industrial and Applied
Mathematics, 1999.

[291] D.T.B. Kelly, K.J.H. Law, and A.M. Stuart. Well-posedness and accuracy of the
ensemble Kalman filter in discrete and continuous time. Nonlinearity, 27(10):2579,
2014.

[292] M. Kennedy and A. O’Hagan. Predicting the output from a complex computer code
when fast approximations are available. Biometrika, 87:1–13, 2000.

[293] B. Kerckoffs, O.P. Faris, P.H. Boveenderd, F.W. Prinzen, K. Smits, and T. Arts.
Timing of depolarization and contraction in the paced canine left ventricle:. J.
Cardiovasc. Electrophysiol., 14:S188–S195, 2003.

[294] R.S. Keynton, M.M. Evancho, R.L. Sims, N.V. Rodway, A. Gobin, and S.E. Rittgers.
Intimal hyperplasia and wall shear in arterial bypass graft distal anastomoses: an in
vivo model study. J. Biomech. Engr., 123:464, 2001.

[295] S.S. Khalafvand, L. Zhong, and E.Y.K. Ng. Three-dimensional cfd/mri modeling
reveals that ventricular surgical restoration improves ventricular function by mod-
ifying intraventricular blood flow. International Journal for Numerical Methods in
Biomedical Engineering, 30(10):1044–1056, 2014.

[296] H. Kim, J. Lu, M.S. Sacks, and K.B. Chandran. Dynamic simulation of bioprosthetic
heart valves using a stress resultant shell model. Journal of Biomechanics, 36(2):262–
275, 2008.

[297] H. Kim, I.E. Vignon-Clementel, C. Figueroa, J. Ladisa, K. Jansen, J. Feinstein, and

C. Taylor. On coupling a lumped parameter heart model and a three-dimensional
finite element aorta model. Ann Biomed Eng, 37:2153–2169, 2009.

[298] C. Kleinstreuer. Biofluid Dynamics: Principles and Selected Applications. CRC

Press, 2006.

[299] P. Kohl, P. Hunter, and D. Noble. Stretch-induced changes in heart rate and rhythm:
clinical observations, experiments and mathematical models. Progress in Biophysics
and Molecular Biology, 71(1):91–138, 1999.

[300] P. Kohl and F. Sachs. Stretch-induced changes in heart rate and rhythm: clinical
observations, experiments and mathematical models. PHILOSOPHICAL TRANS-
ACTIONS of the Royal society A, 359:1173–1185, 2001.

[301] E. Konukoglu, J. Relan, U. Cilingir, B.H. Menze, P. Chinchapatnam, A. Jadidi,

H. Cochet, M. Hocini, H. Delingette, P. Jaı̈s, M. Haı̈ssaguerre, N. Ayache, and

185
 M. Sermesant. Efficient probabilistic model personalization integrating uncertainty
on data and parameters: Application to eikonal-diffusion models in cardiac electro-
physiology. Progress Biophys. Molec. Biology, 107(1):134–146, 2011.

[302] T. Korakianitis and Y. Shi. A concentrated parameter model for the human car-
diovascular system including heart valve dynamics and atrioventricular interaction.
Medical Engineering & Physics, 28(7):613–628, 2006.

[303] T. Korakianitis and Y. Shi. Numerical simulation of cardiovascular dynamics with

healthy and diseased heart valves. J Biomech, 39(11):1964–1982, 2006.

[304] P. Koutsourelakis. Accurate uncertainty quantification using inaccurate models.

SIAM J. Sci. Comput., 31(5):3274–3300, 2009.

[305] A. Krishnamurthy, C.T. Villongco, J. Chuang, L.R. Frank, V. Nigam, E. Belezzuoli,

P. Stark, D.E. Krummen, S. Narayan, J.H. Omens, A.D. McCulloch, and R.C.P.
Kerckhoffs. Patient-specific models of cardiac biomechanics. J. Comput. Phys.,
244:4–21, 2013.

[306] R.H. Kufahl and M.E. Clark. A circle of willis simulation using distensible vessels
and pulsatile flow. J Biomech Eng, 107(2):112–122, 1985.

[307] E. Kuhl and G.A. Holzapfel. A continuum model for remodeling in living structures.
Journal of Materials Science, 21(42):8811–8823, 2006.

[308] K. Kunisch and A. Rund. Time optimal control of the monodomain model in cardiac
electrophysiology. IMA J. Appl. Math., 80(6):1664–1683, 2015.

[309] K. Kunisch and B. Vexler. Optimal vortex reduction for instationary flows based
on translation invariant cost functionals. SIAM J. Control Optim, 46(4):1368–1397,
2007.

[310] K.S. Kunzelman and R.P. Cochran. Mechanical properties of basal and marginal
mitral valve chordae tendineae. ASAIO Trans., 36:M405, 1990.

[311] K.S. Kunzelman, R.P. Cochran, C. Chuong, W.S. Ring, Verrier, and R.D. Eberhart.
Finite element analysis of the mitral valve. J. Heart Valve Dis., 2:326–340, 1993.

[312] U. Kuttler, M. Gee, C. Forster, A. Comerford, and W.A. Wall. Coupling strategies
for biomedical fluid-structure interaction problems. Int. J. Numer. Meth. Biomed.
Engng., 26:305–321, 2010.

[313] U. Kuttler and W.A. Wall. Fixed-point fluidstructure interaction solvers with dy-
namic relaxation. Comp. Mech, 43(1):61–72, 2008.

[314] A. Laadhari and A. Quarteroni. Aortic valve dynamics using a fluid structure in-
teraction model - the physiology of opening and closing. International Journal for
Numerical Methods in Biomedical Engineering, 32(5):, 2016.

186
[315] R. Lal, B. Mohammadi, and F. Nicoud. Data assimilation for identification of car-
diovascular network characteristics. Int. J. Numer. Methods Biomed. Engrg., pages
n/a–n/a, 2016.
[316] R.M. Lancellotti, C. Vergara, L. Valdettaro, S. Bose, and A. Quarteroni. Large
eddy simulations for blood fluid-dynamics in real stenotic carotids. MOX-Report n.
63-2015, Department of Mathematics, Politecnico di Milano, Italy, 2015.
[317] S. Land, S.A. Niederer, J.M. Aronsen, E.S. Espe, L. Zhang, W.E. Louch, [Link],
O.M. Sejersted, and N.P. Smith. An analysis of deformation-dependent electrome-
chanical coupling in the mouse heart. The Journal of Physiology, 590(18):4553–4569,
2012.
[318] T. Lassila, A. Manzoni, A. Quarteroni, and G. Rozza. Boundary control and shape
optimization for the robust design of bypass anastomoses under uncertainty. ESAIM:
Math. Modelling Numer. Anal., 47(4):1107–1131, 2013.
[319] T. Lassila, A. Manzoni, A. Quarteroni, and G. Rozza. A reduced computational
and geometrical framework for inverse problems in haemodynamics. Int. J. Numer.
Methods Biomed. Engng., 29(7):741–776, 2013.
[320] T. Lassila, A. Quarteroni, and G. Rozza. A reduced basis model with parametric cou-
pling for fluid-structure interaction problem. SIAM J. Sci. Comput., 34(2):A1187–
A1213, 2012.
[321] K. Law, A. Stuart, and K. Zygalakis. Data Assimilation. A Mathematical Introduc-
tion, volume 62 of Texts in Applied Mathematics. Springer International Publishing
Switzerland, 2015.
[322] T.B. Le and F. Sotiropoulos. Fluidstructure interaction of an aortic heart valve
prosthesis driven by an animated anatomic left ventricle. Journal of Computational
Physics, 224(1):4162, 2013.
[323] O.P. Le Maı̂tre and O. Knio. Spectral Methods for Uncertainty Quantification With
Applications to Computational Fluid Dynamics. Springer Netherlands, 2010.
[324] H. Lee. Optimal control for quasi-newtonian flows with defective boundary condi-
tions. Comput. Methods Appl. Mech. Engrg., 200:24982506, 2011.
[325] J. Lee, A. Cookson, I. Roy, E. Kerfoot, L. Asner, G. Vigueras, T. Sochi, S. De-
paris, C. Michler, N. P. Smith, and D. A. Nordsletten. Multiphysics computational
modeling in cheart. SIAM J. Sci. Comput., 38(3):C150C178, 2016.
[326] S.E. Lee, S.W. Lee, P.F. Fischer, H.S. Bassiouny, and F. Loth. Direct numeri-
cal simulation of transitional flow in a stenosed carotid bifurcation. J Biomech,
41(11):2551–2561, 2008.
[327] M.J. Legato. Ultrastructure of the atrial, ventricular, and purkinje cell, with special
reference to the genesis of arrhythmias. Circulation, 47:178–189, 1973.

187
[328] C.A.D. Leguy, A.S.Z. Bosboom, E.M.H. abd Belloum, A.P.G. Hoeks, and F.N.
van de Vosse. Global sensitivity analysis of a wave propagation model for arm
arteries. Med. Eng. Phys., 33(8):1008–1016, 2011.

[329] M. Lei, J. Archie, and C. Kleinstreuer. Computational design of a bypass graft that
minimizes wall shear stress gradients in the region of the distal anastomosis. J. Vasc.
Surg., 25(4):637–646, 1997.

[330] J.S. Leiva, P.J. Blanco, and G.C. Buscaglia. Partitioned analysis for dimensionally-
heterogeneous hydraulic networks. Mult Model Simul, 9:872–903, 2011.

[331] J. Leray. Sur les mouvements d’un liquide visqueux emplissant l’espace. Acta Math-
ematica, 63:193–248, 1934.

[332] D. Lesagea, E.D. Angelini, I. Blochb, and G. Funka-Lea. A review of 3d vessel lumen
segmentation techniques: Models, features and extraction schemes. Medical Image
Analysis, 13:819845, 2009.

[333] A. Leuprecht, K. Perktold, Martin Prosi, T. Berk, W. Trubel, and H. Schima. Nu-
merical study of hemodynamics and wall mechanics in distal end-to-side anastomoses
of bypass grafts. Journal of Biomechanics, 2(35):225–236, 2002.

[334] R.J. LeVeque. Numerical methods for conservation laws, volume 132. Birkhäuser
Basel, 1992.

[335] C. Li and C. Vuik. Eigenvalue analysis of the simple preconditioning for incompress-
ible flow. Numerical Linear Algebra with Applications, 11(5-6):511–523, 2004.

[336] D. Li and A.M. Robertson. A structural multi-mechanism damage model for cerebral
arterial tissue. Journal of Biomechanical Engineering, 131(10):101013, 2013.

[337] F. Liang and H. Liu. Simulation of hemodynamic responses to the valsalva maneuver:
An integrative computational model of the cardiovascular system and the autonomic
nervous system. The Journal of Physiological Sciences, 1(56):45–65, 2006.

[338] J.L. Lions and G. Prodi. Un theoreme d’existence et d’unicite dans les equations de
navier-stokes en dimension 2. C.R. Acad. Sci. Paris, 248:3519–3521, 1959.

[339] Y. Liu, C. Charles, M. Gracia, H. Gregersen, and G. S. Kassab. Surrounding tissues

affect the passive mechanics of the vessel wall: theory and experiment. Am J Physiol
Heart Circ Physiol, 293:H3290–H3300, 2007.

[340] D. Lombardi. Inverse problems in 1d hemodynamics on systemic networks: A se-

quential approach. Int. J. Numer. Meth. Biomed. Engng, 30(2):160–179, 2014.

[341] M. Lorenzo-Valdés, G.I. Sanchez-Ortiz, A.G. Elkington, R.H. Mohiaddin, and

D. Rueckert. Segmentation of 4d cardiac mr images using a probabilistic atlas and
the em algorithm. Medical Images Analysis, 8(3):255265, 2004.

188
[342] F. Loth, P.F. Fischer, and H.S. Bassiouny. Blood flow in end-to-side anastomoses.
Annu. Rev. Fluid. Mech., 40:367–393, 2008.

[343] C.H. Luo and Y. Rudy. A model of the ventricular cardiac action potential. depo-
larization, repolarization, and their interaction. Circ Res, 68:1501–1526, 1991.

[344] C.H. Luo and Y. Rudy. A dynamic model of the cardiac ventricular action potential.
i. simulations of ionic currents and concentration changes. Circ Res, 74:1071–1096,
1994.

[345] C.H. Luo and Y. Rudy. A dynamic model of the cardiac ventricular action potential.
ii. afterdepolarizations, triggered activity, and potentiation. Circ Res, 74:1097–1113,
1994.

[346] M. Lykaser and B.F. Nielsen. Towards a level set framework for infarction modeling:
an inverse problem. Int. J. Numer. Anal. Model., 3(4):377–394, 2006.

[347] M.C. MacLachlan, B.F. Nielsen, m. Lysaker, and A. Tveito. Computing the size
and location of myocardial ischemia using measurements of ST-segment shift. IEEE
Trans. Biomed. Engng., 53(6):1024–1031, 2006.

[348] Y. Maday. Analysis of coupled models for fluid-structure interaction of internal

flows. In Cardiovascular mathematics, edited by A. Quarteroni, L. Formaggia, A.
Veneziani, Chapter 8, pages 279–306. Springer, 2009.

[349] A. Mahmoud, A. El-Barkouky, H.E. Farag, J. Graham, and A. Farag. A non-invasive

method for measuring blood flow rate in superficial veins from a single thermal
image. In Computer Vision and Pattern Recognition Workshops (CVPRW), 2013
IEEE Conference on, pages 354–359. IEEE, 2013.

[350] A.C.I. Malossi. Partitioned Solution of Geometrical Multiscale Problems for the
Cardiovascular System: Models, Algorithms, and Applications. PhD thesis, École
Polytechnique Fédérale de Lausanne, 2012. Thesis n◦ 5453.

[351] A.C.I. Malossi, P.J. Blanco, P. Crosetto, S. Deparis, and A. Quarteroni. Implicit
coupling of one-dimensional and three-dimensional blood flow models with compliant
vessels. Multiscale Model Simul, 11(2):474–506, 2013.

[352] A.C.I. Malossi, P.J. Blanco, and S. Deparis. A two-level time step technique for the
partitioned solution of one-dimensional arterial networks. Comput. Methods Appl.
Mech. Engrg., 237–240:212–226, 2012.

[353] T. Mansi, I. Voigt, B. Georgescu, X. Zheng, E.A. Mengue, M. Hackl, R.I. Ionasec,
T. Noack, J. Seeburger, and D. Comaniciu. An integrated framework for finite-
element modeling of mitral valve biomechanics from medical images: Application to
mitralclip intervention planning. Medical Image Analysis, 16(7):13301346, 2012.

189
[354] A. Manzoni. Reduced Models for optimal control, shape optimization and inverse
problems in haemodynamics. PhD thesis, Ecole Polytechnique Fédérale de Lausanne,
2012.

[355] A. Manzoni, T. Lassila, A. Quarteroni, and G. Rozza. A reduced-order strategy for

solving inverse bayesian shape identification problems in physiological flows. In G.H.
Bock, P.X. Hoang, R. Rannacher, and P.J. Schlöder, editors, Modeling, Simulation
and Optimization of Complex Processes - HPSC 2012: Proceedings of the Fifth Inter-
national Conference on High Performance Scientific Computing, March 5-9, 2012,
Hanoi, Vietnam, pages 145–155. Springer International Publishing, Cham, 2014.

[356] A. Manzoni, S. Pagani, and T. Lassila. Accurate solution of bayesian inverse un-
certainty quantification problems combining reduced basis methods and reduction
error models. SIAM/ASA J. Uncert. Quant., 4:380–412, 2016.

[357] A. Manzoni, A. Quarteroni, and G. Rozza. Model reduction techniques for fast
blood flow simulation in parametrized geometries. Int. J. Numer. Methods Biomed.
Engng., 28(6–7):604–625, 2012.

[358] A. Manzoni, A. Quarteroni, and G. Rozza. Shape optimization of cardiovascular

geometries by reduced basis methods and free-form deformation techniques. Int. J.
Numer. Methods Fluids, 70(5):646–670, 2012.

[359] S. Marchesseau, H. Delingette, M. Sermesant, and N. Ayache. Fast parameter cal-

ibration of a cardiac electromechanical model from medical images based on the
unscented transform. Biomech. Model. Mechanobiol., 12(4):815–831, 2012.

[360] S. Marchesseau, H. Delingette, M. Sermesant, R. Cabrera-Lozoya, C. Tobon-Gomez,

P. Moireau, R.M. Figueras i Ventura, K. Lekadir, A. Hernandez, M. Garreau,
E. Donal, C. Leclercq, S.G. Duckett, K. Rhode, C.A. Rinaldi, A.F. Frangi, R. Razavi,
D. Chapelle, and N. Ayache. Personalization of a cardiac electromechanical model
using reduced order unscented kalman filtering from regional volumes. Medical Image
Analysis, 17(7):816–829, 2013.

[361] K.A. Mardal, B.F. Nielsen, X. Cai, and A. Tveito. Semi-implicit time-discretization
schemes for the bidomain modelan order optimal solver for the discretized bidomain
equations. Numerical Linear Algebra with Applications, 14(2):83–98, 2007.

[362] K.N. Margaris and R.A. Black. Modelling the lymphatic system: challenges and
opportunities. J. R. Soc. Interface, 69(9):601–612, 2012.

[363] G. Marom. Numerical Methods for Fluid–Structure Interaction Models of Aortic

Valves. Arch Computat Methods Eng, 22(4):595–620, 2015.

[364] A.L. Marsden. Optimization in cardiovascular modeling. Annual Review of Fluid

Mechanics, 46(1):519–546, 2014.

190
[365] A.L. Marsden, J.A. Feinstein, and C.A. Taylor. A computational framework for
derivative-free optimization of cardiovascular geometries. Comput. Methods Appl.
Mech. Engrg., 197(21–24):1890–1905, 2008.

[366] V. Martin, F. Clément, A. Decoene, and J.-F. Gerbeau. Parameter identification for
a one-dimensional blood flow model. ESAIM: Proc., 14:174–200, 2005.

[367] K.S. Matthys, J. Alastruey, J. Peiró, A.W. Khir, P. Segers, P.R. Verdonck, K.H.
Parker, and S.J. Sherwin. Pulse wave propagation in a model human arterial net-
work: Assessment of 1-d numerical simulations against in vitro measurements. J
Biomech, 40(15):3476–3486, 2007.

[368] B. Maury. The Respiratory System in Equations. Springer-Verlag Italia, 2013.

[369] K. May-Newman, C. Lam, and F.C.P. Yin. A hyperelastic constitutive law for aortic
valve tissue. J Biomech Eng, 131:081009, 2009.

[370] K. May-Newman and F.C.P. Yin. A constitutive law for mitral valve tissue. J
Biomech Eng, 120(1):38–47, 1998.

[371] W.D. Merryman, H.Y.S. Huang, F.J. Schoen, and M.S. Sacks. The effects of cel-
lular contraction on aortic valve leaflet flexural stiffness. Journal of Biomechanics,
39(1):88–96, 2006.

[372] F. Migliavacca, R. Balossino, G. Pennati, G. Dubini, T.Y. Hsia, M.R. de Leval, and
E.L. Bove. Multiscale modelling in biofluidynamics: application to reconstructive
paediatric cardiac surgery. J Biomech, 39:1010–1020, 2006.

[373] F. Migliavacca and G. Dubini. Computational modeling of vascular anastomoses.

Biomech. Model. Mechanobiol., 3(4):235–250, 2005.

[374] R. Mittal and G. Iaccarino. Immersed boundary methods. Annual Review of Fluid
Mechanics, 37:239–261, 2005.

[375] R. Mittal, J.H. Seo, V. Vedula, Y.J. Choi, H. Liu, H.H. Huang, S. Jain, L. Younes,
T. Abraham, and R.T. George. Computational modeling of cardiac hemodynamics:
Current status and future outlook. Journal of Computational Physics, 305:1065–
1082, 2016.

[376] P. Moireau, C. Bertoglio, N. Xiao, C. A. Figueroa, C. A. Taylor, D. Chapelle, and

J.-F. Gerbeau. Sequential identification of boundary support parameters in a fluid-
structure vascular model using patient image data. Biomech. Model. Mechanobiol.,
12(3):475–496, 2013.

[377] P. Moireau and D. Chapelle. Reduced-order unscented kalman filtering with appli-
cation to parameter identification in large-dimensional systems. ESAIM: Control
Optim. Calc. Var., 17(2):380–405, 2011.

191
[378] P. Moireau, D. Chapelle, and P. Le Tallec. Joint state and parameter estimation
for distributed mechanical systems. Comput. Methods Appl. Mech. Engrg., 197(6–
8):659–677, 2008.

[379] P. Moireau, D. Chapelle, and P.L. Tallec. Filtering for distributed mechanical sys-
tems using position measurements: perspectives in medical imaging. Inverse Prob-
lems, 25:035010, 2009.

[380] P. Moireau, N. Xiao, M. Astorino, C. A. Figueroa, D. Chapelle, C. A. Taylor, and

J.F. Gerbeau. External tissue support and fluid–structure simulation in blood flows.
Biomechanics and Modeling in Mechanobiology, 11(1–2):1–18, 2012.

[381] J.A. Moore, D.A. Steinman, and C.R. Ethier. Computational blood flow modelling:
Errors associated with reconstructing finite element models from magnetic resonance
images. Journal of Biomechanics, 31(2):179–184, 1997.

[382] H. Moradkhani, S. Sorooshian, H.V. Gupta, and P.R. Houser. Dual state-parameter
estimation of hydrological models using ensemble kalman filter. Adv. Water Resour.,
28(2):135–147, 2005.

[383] U. Morbiducci, R. Ponzini, G. Rizzo, M. Cadioli, A. Esposito, F. De Cobelli, A. Del

Maschio, F.M. Montevecchi, and A. Redaelli. In vivo quantification of helical blood
flow in human aorta by time-resolved three-dimensional cine phase contrast magnetic
resonance imaging. Ann Biomed Eng, 37(3):516–531, 2009.

[384] D. Mozaffarian, E.J. Benjamin, A.S. Go, D.K. Arnett, M.J. Blaha, M. Cushman,
S.R. Das, S. de Ferranti, J.-P. Després, H.J. Fullerton, V.J. Howard, M.D. Huffman,
C.R. Isasi, M.C. Jiménez, S.E. Judd, B.M. Kissela, J.H. Lichtman, L.D. Lisabeth,
S. Liu, R.H. Mackey, D.J. Magid, D.K. McGuire, E.R. Mohler, C.S. Moy, P. Muntner,
M.E. Mussolino, K. Nasir, R.W. Neumar, G. Nichol, L. Palaniappan, D.K. Pandey,
M.J. Reeves, C.J. Rodriguez, W. Rosamond, P.D. Sorlie, J. Stein, A. Towfighi, T.N.
Turan, S.S. Virani, D. Woo, R.W. Yeh, and M.B. Turner. Heart disease and stroke
statistics—2016 update. Circulation, 132, 2015.

[385] J. Muller, O. Sahni, X. Lia, K.E. Jansen, M. S. Shephard, and C.A. Taylor.
Anisotropic adaptive finite element method for modelling blood flow. Computer
Methods in Biomechanics and Biomedical Engineering, 8(5):295–305, 2005.

[386] L. Muller and E.F. Toro. Well-balanced high-order solver for blood flow in networks
of vessels with variable properties. Int J Num Meth Biomed Eng, 29(12):1388–1411,
2013.

[387] L. Muller and E.F. Toro. A global multiscale mathematical model for the human
circulation with emphasis on the venous system. Int J Num Meth Biomed Eng,
30(7):681–725, 2014.

192
[388] M. Munteanu, L.F. Pavarino, and S. Scacchi. A scalable newton–krylov–schwarz
method for the bidomain reaction-diffusion system. SIAM J. Sci. Comput.,
31(5):3861–3883, 2009.

[389] M. Murillo and X.C. Cai. A fully implicit parallel algorithm for simulating the non-
linear electrical activity of the heart. Numerical Linear Algebra with Applications,
2-3(11):261–277, 2004.

[390] A. Muszkiewicz, O.J. Britton, P. Gemmell, E. Passini, C. Sánchez, X. Zhou,

A. Carusi, T.A. Quinn, K. Burrage, A. Bueno-Orovio, and B. Rodriguez. Vari-
ability in cardiac electrophysiology: Using experimentally-calibrated populations of
models to move beyond the single virtual physiological human paradigm. Progress
Biophys. Molec. Biology, 120(1-3):115–127, 2016.

[391] Ch. Nagaiah, K. Kunisch, and G. Plank. Numerical solution for optimal control of
the reaction-diffusion equations in cardiac electrophysiology. Comput. Optim. Appl.,
49:149–178, 2011.

[392] Ch. Nagaiah, K. Kunisch, and G. Plank. On boundary stimulation and optimal
boundary control of the bidomain equations. Mathem. Biosciences, 245(2):206–215,
2013.

[393] Ch. Nagaiah, K. Kunisch, and G. Plank. Optimal control approach to termination
of re-entry waves in cardiac electrophysiology. J. Math. Biol., 67(2):1–30, 2013.

[394] Ch. Nagaiah, K. Kunisch, and G. Plank. Pde constrained optimization of electrical
defibrillation in a 3d ventricular slice geometry. Int. J. Numer. Meth. Biomed.
Engng., 32(4):e02742, 2016.

[395] A. Nagler, C. Bertoglio, M. Gee, and W. Wall. Personalization of cardiac fiber

orientations from image data using the unscented kalman filter. Functional Imaging
and Modeling of the Heart, 7945:132–140, 2013.

[396] M.P. Nash and A.V. Panfilov. Electromechanical model of excitable tissue to study
reentrant cardiac arrhythmias. Progress in Biophysics and Molecular Biology, 2-
3(85):501–522, 2004.

[397] F. Negri. Efficient Reduction Techniques for the Simulation and Optimization of
Parametrized Systems: Analysis and Applications. PhD thesis, Ecole Polytechnique
Fédérale de Lausanne, 2016.

[398] M.G.C. Nestola, E. Faggiano, C Vergara, R.M. Lancellotti, S. Ippolito, C. Antona,

A. Quarteroni, and R. Scrofani. Computational comparison of aortic root stresses in
presence of stentless and stented aortic valve bio-prostheses. Computer Methods in
Biomechanics and Biomedical Engineering, DOI:10.1080/10255842.2016.1207171,
2016.

193
[399] N.M Newmark. A method of computation for structural dynamics. Journal of
Engineering Mechanics, 85:67–94, 1959.
[400] N. K. Nichols. Mathematical concepts of data assimilation. In W. Lahoz, B. Khat-
tatov, and R. Menard, editors, Data Assimilation: Making Sense of Observations,
pages 13–39. Springer Berlin Heidelberg, Berlin, Heidelberg, 2010.
[401] W.W. Nichols and M.F. O’Rourke, editors. McDonald’s Blood Flow in Arteries.
Hodder Arnold, 2005.
[402] S.A. Niederer, P.J. Hunter, and N.P. Smith. A quantitative analysis of cardiac
myocyte relaxation: A simulation study. Biophysical Journal, 90(5):1697–1722, 2006.
[403] S.A. Niederer, E. Kerfoot, A.P. Benson, M.O. Bernabeu, O. Bernus, C. Bradley,
E..M Cherry, R. Clayton, F.H. Fenton, A. Garny, et al. Verification of cardiac tissue
electrophysiology simulators using an n-version benchmark. Phil. Trans. R. Soc. A,
369(1954):4331–4351, 2011.
[404] S.A. Niederer and N.P Smith. A mathematical model of the slow force response to
stretch in rat ventricular myocytes. Biophysical Journal, 92(11):4030–4044, 2007.
[405] S.A. Niederer and N.P Smith. An improved numerical method for strong coupling of
excitation and contraction models in the heart. Progress in Biophysics and Molecular
Biology, 96(1–3):90–111, 2008.
[406] B.F. Nielsen, X. Cai, and M. Lykaser. On the possibility for computing the trans-
membrane potential in the heart with a one shot method: an inverse problem. Math.
Biosciences, 210:523–553, 2007.
[407] B.F. Nielsen, M. Lykaser, and A. Tveito. On the use of the resting potential and
level set methods for identifying ischemic heart disease: An inverse problem. J.
Comput. Phys., 220:772–790, 2007.
[408] J.A. Nitsche. Uber ein variationsprinzip zur lozung von dirichlet-problemen bei
verwendung von teilraumen, die keinen randbedingungen unterworfen sind. Ab-
handlungen aus dem Mathematischen Seminar der Universitat Hamburg, 36:9–15,
1970/71.
[409] F. Nobile. Numerical approximation of fluid-structure interaction problems with ap-
plication to haemodynamics. PhD thesis, École Polytechnique Fédérale de Lausanne,
2001. Thesis n◦ 2458.
[410] F. Nobile, M. Pozzoli, and C. Vergara. Time accurate partitioned algorithms for
the solution of fluid-structure interaction problems in haemodynamics. Computer &
Fluids, 86:470–482, 2013.
[411] F. Nobile, M. Pozzoli, and C. Vergara. Inexact accurate partitioned algorithms for
fluid-structure interaction problems with finite elasticity in haemodynamics. Journal
of Computational Physics, 273:598–617, 2014.

194
[412] F. Nobile, R. Tempone, and C. G. Webster. A sparse grid stochastic collocation
method for partial differential equations with random input data. SIAM J. Numer.
Anal., 46:2309–2345, 2008.

[413] F. Nobile and C. Vergara. An effective fluid-structure interaction formulation for

vascular dynamics by generalized Robin conditions. SIAM J Sc Comp, 30(2):731–
763, 2008.

[414] F. Nobile and C. Vergara. Partitioned algorithms for fluid-structure interaction

problems in haemodynamics. Milan Journal of Mathematics, 80(2):443–467, 2012.

[415] J. Nocedal. Theory of algorithms for unconstrained optimization. Acta Numerica,

1:199–242, 1992.

[416] D.A. Nordsletten, M. McCormick, P.J. Kilner, P. Hunter, D. Kayand, and N.P.
Smith. Fluid–solid coupling for the investigation of diastolic and systolic human left
ventricular function. International Journal for Numerical Methods in Biomedical
Engineering, 27(7):1017–1039, 2011.

[417] D.A. Nordsletten, S.A. Niederer, M.P. Nash, P.J. Hunter, and N.P. Smith. Coupling
multi-physics models to cardiac mechanics. Progress in Biophysics and Molecular
Biology, 104(1-3):77–88, 2011.

[418] A. A. Oberai, N. H. Gokhale, and G. R. Feijoo. Solution of inverse problems in

elasticity imaging using the adjoint method. Inverse Problems, 19(2):297–313, 2003.

[419] T. O’Donnell, M.P.D. Jolly, and A. Gupta. A cooperative framework for segmenta-
tion using 2d active contours and 3d hybrid models as applied to branching cylin-
drical structures. Proc. IEEE Int. Conf. Computer Vision, pages 454 – 459, 1998.

[420] M.S. Olufsen, C.S. Peskin, W.Y. Kim, E.M. Pedersen, A. Nadim, and J. Larsen.
Numerical simulation and experimental validation of blood flow in arteries with
structured-tree outflow conditions. Ann Biomed Eng, 28(11):1281–1299, 2000.

[421] S.A. Orszag, M. Israeli, and M. Deville. Boundary conditions for incompressible
flows. J. Sci. Comput., 1:75–111, 1986.

[422] H. Osnes and J. Sundnes. Uncertainty analysis of ventricular mechanics using the
probabilistic collocation method. IEEE Trans. Biomed. Engng., 59(8):2171–2179,
2012.

[423] A.A. Owida, H. Do, and Y.S. Morsi. Numerical analysis of coronary artery bypass
grafts: an over view. Comput. Methods Programs Biomed., 108(2):689–705, 2012.

[424] M. Padala, M.S. Sacks, S.W. Liou, K. Balachandran, Z. He, and A.P. Yoganathan.
Mechanics of the mitral valve strut chordae insertion region. Journal of Biomechan-
ical Engineering, 132(8):081004, 2010.

195
[425] S. Pagani. Reduced order models for inverse problems and uncertainty quantification
in cardiac electrophysiology. PhD thesis, Mathematical Models and Methods in
Engineering, Dipartimento di Matematica, Politecnico di Milano, 2016.
[426] S. Pagani, A. Manzoni, and A. Quarteroni. A reduced basis ensemble Kalman filter
for state/parameter identification in large-scale nonlinear dynamical systems. MOX-
Report n. 24-2016, Department of Mathematics, Politecnico di Milano, Italy, 2016.
[427] S. Palamara, C. Vergara, D. Catanzariti, E. Faggiano, M. Centonze, C. Pangrazzi,
M. Maines, and A. Quarteroni. Computational generation of the Purkinje network
driven by clinical measurements: the case of pathological propagations. Int. J.
Numer. Meth. Biomed. Engng., 30(12):1558–1577, 2014.
[428] S. Palamara, C. Vergara, E. Faggiano, and F. Nobile. An effective algorithm for the
generation of patient-specific purkinje networks in computational electrocardiology.
J. Comp. Phys., 283:495–517, 2015.
[429] A.V. Panfilov. Three-dimensional organization of electrical turbulence in the heart.
Phys. Rev. E, 59:R6251–R6254, 1999.
[430] S. Pant, B. Fabréges, J-F. Gerbeau, and I. E. Vignon-Clementel. A methodological
paradigm for patient-specific multi-scale cfd simulations: from clinical measurements
to parameter estimates for individual analysis. Int. J. Numer. Meth. Biomed. Engng.,
30(12):1614–1648, 2014.
[431] G. Papadakis. Coupling 3d and 1d fluid–structure-interaction models for wave prop-
agation in flexible vessels using a finite volume pressure-correction scheme. Comm
Numer Meth Eng, 25(5):533–551, 2009.
[432] S.V. Patankar and D.B. Spalding. A calculation procedure for heat, mass and mo-
mentum transfer in three-dimensional parabolic flows. International Journal of Heat
and Mass Transfer, 15(10):1787–1806, 1972.
[433] P. Pathmanathan, S. J. Chapman, D. J. Gavaghan, and J.P. Whiteley. Cardiac
electromechanics: The effect of contraction model on the mathematical problem
and accuracy of the numerical scheme. Quarterly Jnl of Mechanics & Applied Math,
63(3):375–399, 2010.
[434] P. Pathmanathan, G.R. Mirams, J. Southern, and J.P. Whiteley. The significant ef-
fect of the choice of ionic current integration method in cardiac electro-physiological
simulations. International Journal for Numerical Methods in Biomedical Engineer-
ing, 27(11):1751–1770, 2011.
[435] L.F. Pavarino and S. Scacchi. Multilevel additive schwarz preconditioners for the
bidomain reaction-diffusion system. SIAM J. Sci. Comput, 31(1):420–443, 2008.
[436] L.F. Pavarino, S. Scacchi, and S. Zampini. Newton–krylov-bddc solvers for nonlin-
ear cardiac mechanics. Computer Methods in Applied Mechanics and Engineering,
295:562–580, 2015.

196
[437] T.J. Pedley. The fluid mechanics of large blood vessels. Cambridge monographs on
mechanics and applied mathematics: Cambridge University Press, 1980.

[438] J. Peiró, L. Formaggia, M. Gazzola, A. Radaelli, and V. Rigamonti. Shape recon-

struction from medical images and quality mesh generation via implicit surfaces.
International Journal for Numerical Methods in Fluids, 53(8):1339–1360, 2007.

[439] J. Peiró and A. Veneziani. Reduced models of the cardiovascular system. In Cardio-
vascular mathematics, edited by A. Quarteroni, L. Formaggia, A. Veneziani, Chapter
10, pages 347–394. Springer, 2009.

[440] M. Pennacchio and V. Simoncini. Efficient algebraic solution of reaction-diffusion

systems for the cardiac excitation process. Journal of Computational and Applied
Mathematics, 145(1):49–70, 2002.

[441] P. Perdikaris and G.E. Karniadakis. Model inversion via multi-fidelity bayesian
optimization: a new paradigm for parameter estimation in haemodynamics, and
beyond. J. R. Soc. Interface, 13:20151107, 2015.

[442] M. Perego, A. Veneziani, and C. Vergara. A variational approach for estimating

the compliance of the cardiovascular tissue: An inverse fluid-structure interaction
problem. SIAM J. Sci. Comput., 33(3):1181–1211, 2011.

[443] K. Perktold and D. Hilbert. Numerical simulation of pulsatile flow in a carotid

bifurcation model. Journal of biomedical engineering, 8(3):193–199, 1986.

[444] K. Perktold, E. Thurner, and T. Kenner. Flow and stress characteristics in rigid
walled and compliant carotid artery bifurcation models. Medical and Biological
Engineering and Computing, 32(1):19–26, 1994.

[445] P. Perona and J. Malik. Scale-space and edge detection using anisotropic diffusion.
IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(7):629–639,
1990.

[446] C. Peskin. Flow patterns around heart valves: A numerical method. J. Comput.
Physics, 10(2):252–271, 1972.

[447] C. Peskin. Fiber architecture of the left ventricular wall: An asymptotic analysis.
Communications on Pure and Applied Mathematics, 42(1):1126–1131, 1989.

[448] C. Peskin. The immersed boundary method. Acta Numerica, 11:479–517, 2002.

[449] C. Petitjean and J.-N. Dacher. A review of segmentation methods in short axis
cardiac mr images. Medical Image Analysis, 15(2):169184, 2011.

[450] S. Pezzuto. Mechanics of the Heart. Constitutive Issues and Numerical Experiments.
PhD thesis, Mathematical Models and Methods in Engineering, Dipartimento di
Matematica, Politecnico di Milano, 2013.

197
[451] D.L. Pham, C. Xu, and J.L. Prince. Current methods in medical image segmentation.
Annual Review of Biomedical Engineering, 2:315–337, 2000.

[452] D.T. Pham. Stochastic methods for sequential data assimilation in strongly nonlinear
systems. Mon. Weather Rev., 129(5):1194–1207, 2001.

[453] M. Piccinelli, C. Vergara, L. Antiga, L. Forzenigo, P. Biondetti, and M. Domanin.

Impact of hemodynamics on lumen boundary displacements in abdominal aortic
aneurysms by means of dynamic computed tomography and computational fluid
dynamics. Biomech Model Mechanobiol, 2013.

[454] C. Pierre. Preconditioning the bidomain model with almost linear complexity. Jour-
nal of Computational Physics, 231(1):8297, 2012.

[455] S. Piperno and C. Farhat. Partitioned prodecures for the transient solution of cou-
pled aeroelastic problems-Part II: energy transfer analysis and three-dimensional
applications. Comput. Methods Appl. Mech. Engrg., 190:3147–3170, 2001.

[456] G. Plank, M. Liebmann, R. Weber dos Santos, E.J. Vigmond, and G. Haase. Alge-
braic multigrid preconditioner for the cardiac bidomain model. IEEE Transaction
on Biomedical Engineering, 54(4):585–596, 2007.

[457] R. Ponzini, C. Vergara, A. Redaelli, and A. Veneziani. Reliable cfd-based estimation

of flow rate in haemodynamics measures. Ultrasound in Med. and Biol., 32(10):1545–
1555, 2006.

[458] R. Ponzini, C. Vergara, G. Rizzo, A. Veneziani, A. Roghi, A. Vanzulli, O. Parodi,

and A. Redaelli. Womersley number-based estimates of blood flow rate in doppler
analysis: In vivo validation by means of phase contrast magnetic resonance imaging.
IEEE Transaction on Biomedical Engineering, 57(7):1807–1815, 2010.

[459] A. Porpora, P. Zunino, C. Vergara, and M. Piccinelli. Numerical treatment of bound-

ary conditions to replace lateral branches in haemodynamics. Int. J. Numer. Meth.
Biomed. Eng., 28(12):1165–1183, 2012.

[460] M. Potse, B. Dubé, J. Richer, A. Vinet, and R.M. Gulrajani. A comparison of mon-
odomain and bidomain reaction-diffusion models for action potential propagation in
the human heart. IEEE Transaction on Biomedical Engineering, 53(12):2425–2435,
2006.

[461] S.F. Pravdin, V.I. Berdyshev, A.V. Panfilov, L.B. Katsnelson, O. Solovyova, and
V.S. Markhasin. Mathematical model of the anatomy and fibre orientation field of
the left ventricle of the heart. Biomedical Engineering OnLine, 12:54, 1989.

[462] G. Prodi. Teoremi di tipo locale per il sistema di navier-stokes e stabilitá delle
soluzioni stazionarie. Rendiconti del Seminario Matematico della Universitá di
Padova, 32:374–397, 1962.

198
[463] V. Prot, B. Skallerud, and G.A. Holzapfel. Transversely isotropic membrane shells
with application to mitral valve mechanics. constitutive modelling and finite ele-
ment implementation. International Journal for Numerical Methods in Engineering,
71:987–1008, 2007.

[464] A.J. Pullan, L.K. Cheng, M.P. Nash, C.P. Bradley, and D.J. Paterson. Noninvasive
electrical imaging of the heart: Theory and model development. Ann. Biomed. Eng,
29:817–836, 2001.

[465] A.J. Pullan, L.K. Cheng, M.P. Nash, A. Ghodrati, R. MacLeod, and D.H. Brooks.
The inverse problem of electrocardiography. In P.W. Macfarlane, A. van Oosterom,
O. Pahlm, P. Kligfield, M. Janse, and J. Camm, editors, Comprehensive Electrocar-
diology, pages 299–344. Springer London, 2010.

[466] S. Puwal and B.J. Roth. Forward euler stability of the bidomain model of cardiac
tissue. IEEE Transaction on Biomedical Engineering, 5(54):951–953, 2007.

[467] Z. Qu and A. Garfinkel. An advanced algorithm for solving partial differential

equation in cardiac conduction. IEEE Transaction on Biomedical Engineering,
46(9):1166–1168, 1998.

[468] A. Quarteroni, editor. Modeling the Heart and the Circulatory System. Springer,
2015.

[469] A. Quarteroni and L. Formaggia. Mathematical modelling and numerical simula-

tion of the cardiovascular system. In Computational Models for the Human Body,
volume 12 of Handbook of Numerical Analysis, pages 3–127. Elsevier, 2004.

[470] A. Quarteroni, T. Lassila, S. Rossi, and R. Ruiz-Baier. Integrated heart - cou-

pling multiscale and multiphysics models for the simulation of the cardiac function.
Comput. Methods Appl. Mech. Eng., page in press, 2016.

[471] A. Quarteroni, A. Manzoni, and F. Negri. Reduced Basis Methods for Partial Dif-
ferential Equations. An Introduction, volume 92 of Unitext. Springer, 2016.

[472] A. Quarteroni, S. Ragni, and A. Veneziani. Coupling between lumped and dis-
tributed models for blood flow problems. Comp Vis Sc, 4(2):111–124, 2001.

[473] A. Quarteroni and G. Rozza. Optimal control and shape optimization of aorto-
coronaric bypass anastomoses. Math. Models Meth. Appl. Sci., 13(12):1801–1823,
2003.

[474] A. Quarteroni and G. Rozza, editors. Reduced Order Methods for Modeling and Com-
putational Reduction, volume 9 of Modeling, Simulation and Applications (MS&A).
Springer International Publishing, Switzerland, 2014.

[475] A. Quarteroni, R. Sacco, and F. Saleri. Numerical mathematics. Springer Berlin,

2000.

199
[476] A. Quarteroni, F. Saleri, and A. Veneziani. Analysis of the Yosida method for
the incompressible Navier-Stokes equations. Journal de Mathématiques Pures et
Appliquées, 78:473–503, 1999.

[477] A. Quarteroni, F. Saleri, and A. Veneziani. Factorization methods for the numerical
approximation of Navier-Stokes equations. Comput. Methods Appl. Mech. Engrg.,
188:505–526, 2000.

[478] A. Quarteroni, M. Tuveri, and A. Veneziani. Computational vascular fluid dynamics:

Problems, models and methods. Computing and Visualisation in Science, 2:163–197,
2000.

[479] A. Quarteroni and A. Valli. Numerical approximation of partial differential equa-

tions. Springer, 1994.

[480] A. Quarteroni and A. Veneziani. Modeling and simulation of blood flow problems.
Technical report, Bristeau, M.-O.(ed.) et al. Chichester: John Wiley & Sons., 1997.

[481] A. Quarteroni and A. Veneziani. Analysis of a geometrical multiscale model based

on the coupling of ode and pde for blood flow simulations. Multiscale Model. Simul.,
1(2):173–195, 2003.

[482] A. Quarteroni, A. Veneziani, and C. Vergara. Geometric multiscale modeling of the

cardiovascular system, between theory and practice. Comput. Methods Appl. Mech.
Engrg., 302:193–252, 2016.

[483] G. Querzoli, S. Fortini, and A. Cenedese. Effect of the prosthetic mitral valve on
vortex dynamics and turbulence of the left ventricular flow. Phys. Fluids, 22:041901,
2010.

[484] M. Raghavan and D. Vorp. Towards a biomechanical tool to evaluate rupture poten-
tial of abdominal aortic aneurysm: identification of a finite strain constitutive model
and evaluation of its applicability. Journal of Biomechanics, 33:475–482, 2000.

[485] R. Rannacher. On Chorin’s projection method for incompressible Navier-Stokes equa-

tions, Lecture Notes in Mathematics, volume 1530, pages 167–183. Springer, Berlin,
1992.

[486] S.P. Raya and J.K. Udupa. Shape-based interpolation of multidimensional objects .
IEEE Transactions on Medical Imaging, 9(1):32–42, 1990.

[487] V.L. Rayz, S.A. Berger, and D. Saloner. Transitional flows in arterial fluid dynamics.
Computer Methods in Applied Mechanics and Engineering, 196(31–32):3043–3048,
2007.

[488] T. Rees, H. S. Dollar, and A. J. Wathen. Optimal Solvers for PDE-Constrained

Optimization. SIAM J. Sci. Comput., 32:271–298, 2010.

200
[489] J. Relan, P. Chinchapatnam, M. Sermesant, K. Rhode, M. Ginks, H. Delingette,
C.A. Rinaldi, R. Razavi, and N. Ayache. Coupled personalization of cardiac elec-
trophysiology models for prediction of ischaemic ventricular tachycardia. Interface
Focus, 1(3):396–407, 2011.

[490] P. Reymond, F. Merenda, [Link], D. Rufenacht, and N. Stergiopulos. Validation

of a one-dimensional model of the systemic arterial tree. Am J Physiol Heart Circ
Physiol, 297:H208–H222, 2009.

[491] C.P. Robert and G. Casella. Monte Carlo Statistical Methods. Springer, 2nd edition,
2004.

[492] A.M. Robertson, A. Sequeira, and R.G. Owens. Rheological models for blood. In
Cardiovascular mathematics, edited by L. Formaggia, A. Quarteroni, A. Veneziani,
Chapter 6, pages 211–241. Springer, 2009.

[493] D.D. Robertson, J. Yuan, G. Wang, and M.W. Vannier. Total hip prosthesis metal-
artifact suppression using iterative deblurring reconstruction. J Comput Assist To-
mogr, 21(2):293–298, 1997.

[494] J.M. Rogers and A.D. McCulloch. A collocation-galerkin finite element model of
cardiac action potential propagation. Biophys JIEEE Transaction on Biomedical
Engineering, 41(8):743–757, 1994.

[495] S. Rossi. Anisotropic Modeling of Cardiac Mechanical Activation. PhD thesis, Ecole
Polytechnique Fédérale de Lausanne, 2014.

[496] S. Rossi, T. Lassila, R. Ruiz-Baier, Adélia Sequeira, and A. Quarteroni. Thermody-

namically consistent orthotropic activation model capturing ventricular systolic wall
thickening in cardiac electromechanics. European Journal of Mechanics - A/Solids,
48(1):129–142, 2014.

[497] B.J. Roth. Action potential propagation in a thick strand of cardiac muscle. Circ
Res, 68:162–173, 1991.

[498] Bradley J. Roth. Electrical conductivity values used with the bidomain model of
cardiac tissue. IEEE transactions on bio-medical engineering, 44(4):326–328, 1997.

[499] O. Rousseau. Geometrical modeling of the heart. PhD thesis, Université de Ottawa,
2010.

[500] Y. Rudy and B.J. Messinger-Rapport. The inverse problem in electrocardiography:

solutions in terms of epicardial potentials. Crit. Rev. Biomed. Eng., 16(3):215–268,
1988.

[501] Y. Rudy and J.R. Silva. Computational biology in the study of cardiac ion channels
and cell electrophysiology. Quarterly Reviews of Biophysics, 39(1):57–116, 2006.

201
[502] A. Saltelli, M. Ratto, T. Andres, F. Campolongo, J. Cariboni, D. Gatelli, M. Salsana,
and S. Tarantola. Global sensitivity analysis - The primer. Wiley, 2008.

[503] S. Sankaran, L. Grady, and C.A. Taylor. Impact of geometric uncertainty on hemo-
dynamic simulations using machine learning. Comput. Methods Appl. Mech. Engrg.,
297:167–190, 2015.

[504] S. Sankaran, H.J. Kim, G. Choi, and C.A. Taylor. Uncertainty quantification in
coronary blood flow simulations: Impact of geometry, boundary conditions and blood
viscosity. J. Biomech., 49(12):2540–2547, 2016.

[505] S. Sankaran and A.L. Marsden. The impact of uncertainty on shape optimization
of idealized bypass graft models in unsteady flow. Phys. Fluids, 22:121902, 2010.

[506] S. Sankaran and A.L Marsden. A stochastic collocation method for uncertainty
quantification and propagation in cardiovascular simulations. J. Biomech. Eng.,
133(3):031001, 2011.

[507] R. Weber Dos Santos, G. Plank, S. Bauer, and E.J. Vigmond. Preconditioning
techniques for the bidomain equations. In T.J. Barth, M. Field, M. Griebel, D.E.
Keyes, R.M. Nieminen, D. Roose, and T. Schlick, editors, Proceedings of the XV
International Symposium on Domain Decomposition Methods for PDE’s - Lecture
Notes in Computational Science and Engineering, pages 571–580. Springer, 2004.

[508] S. Särkkä. Bayesian Filtering and Smoothing. Cambridge University Press, 2013.

[509] S.J. Savader, G.B. Lund, and F.A. Osterman. Volumetric evaluation of blood flow
in normal renal arteries with a doppler flow wire: A feasibility study. J Vasc Inter-
ventional Radiol, 8(2):209–214, 1997.

[510] I. Sazonov, S.Y. Yeo, R.L.T. Bevan, X. Xie, R. van Loon, and P. Nithiarasu. Mod-
elling pipeline for subject-specific arterial blood flow: A review. International Jour-
nal for Numerical Methods in Biomedical Engineering, 27:1868–1910, 2011.

[511] D.E. Schiavazzi, G. Arbia, C. Baker, A.M. Hlavacek, T.Y. Hsia, A.L. Marsden, I.E.
Vignon-Clementel, and The Modeling Of Congenital Hearts Alliance (MOCHA) In-
vestigators. Uncertainty quantification in virtual surgery hemodynamics predictions
for single ventricle palliation. Int. J. Numer. Methods Biomed. Engng., 32(3):02737,
2016.

[512] R. J. Schilling, N. S. Peters, and D. W. Davies. Simultaneous endocardial mapping

in the human left ventricle using a noncontact catheter comparison of contact and
reconstructed electrograms during sinus rhythm. Circulation, 98(9):887–?898, 1998.

[513] R. Sebastian, V. Zimmerman, D. Romero, and A.F. Frangi. Construction of a

computational anatomical model of the perpheral cardiac conduction system. IEEE
Transaction on Biomedical Engineering, 58(12):90–100, 2011.

202
[514] J.H. Seo, V. Vedula, T. Abraham, A.C. Lardo, F. Dawoud, H. Luo, and R. Mittal.
Effect of the mitral valve on diastolic flow patterns. Physics of fluids, 26(12):121901,
2014.

[515] M. Sermesant, R. Chabiniok, P. Chinchapatnam, T. Mansi, F. Billet, P. Moireau,

J.M. Peyrat, K. Wong, J. Relan, K. Rhode, M. Ginks, P. Lambiase, H. Delingette,
M. Sorine, C.A. Rinaldi, D. Chapelle, R. Razavi, and N. Ayache. Patient-specific
electromechanical models of the heart for the prediction of pacing acute effects in
crt: a preliminary clinical validation. Med Image Anal, 16(1):201–215, 2012.

[516] M. Sermesant, P. Moireau, O. Camara, J. Sainte-Marie, R. Andriantsimiavona,

R. Cimrman, D.L.G. Hill, D. Chapelle, and R. Razavi. Cardiac function estimation
from {MRI} using a heart model and data assimilation: Advances and difficulties.
Medical Image Analysis, 10(4):642–656, 2006.

[517] J.A. Sethian. Level set methods and fast marching methods. Cambridge University,
1999.

[518] S.J. Sherwin, L. Formaggia, J. Peiró, and V. Franke. Computational modelling of 1d

blood ow with variable mechanical properties and its application to the simulation of
wave propagation in the human arterial system. Int J Num Meth Fluids, 43:673–700,
2003.

[519] S.J. Sherwin, V. Franke, J. Peiró, and K. Parker. One-dimensional modelling of

a vascular network in space-time variables. Journal of Engineering Mathematics,
47(3-4):217–259, 2003.

[520] D. Simon. Optimal State Estimation: Kalman, H-infinity, and Nonlinear Ap-
proaches. John Wiley and Sons, 2006.

[521] N. P. Smith, D. P. Nickerson, E. J. Crampin, and P. J. Hunter. Multiscale compu-

tational modelling of the heart. Acta Numerica, 13:371–431, 2004.

[522] R.L. Spilker and C.A. Taylor. Tuning multidomain hemodynamic simulations to
match physiological measurements. Ann. Biomed. Engng., 38(8):2635–2648, 2010.

[523] T Stankovičová, V Bito, F Heinzel, K Mubagwa, and KR Sipido. Isolation and

morphology of single purkinje cells from the porcine heart. Gen. Physiol. Biophys,
22:329–340, 2003.

[524] B.N. Steele, J. Wan, J.P. Ku, T.J.R Hughes, and C.A. Taylor. In vivo validation of
a one-dimensional finite-element method for predicting blood flow in cardiovascular
bypass grafts. IEEE Trans Biomed Eng, 50:649–656, 2003.

[525] D.A. Steinman, J.B. Thomas, H.M. Ladak, J.S. Milner, B.K. Rutt, and J.D. Spence.
Reconstruction of carotid bifurcation hemodynamics and wall thickness using com-
putational fluid dynamics and mri. Magnetic Resonance in Medicine, 47(1):149–159,
2001.

203
[526] N. Stergiopulos, B.E. Westerhof, J.J. Meister, and N. Westerhof. The four-element
windkessel model. In Engineering in Medicine and Biology Society, 1996. Bridging
Disciplines for Biomedicine. Proceedings of the 18th Annual International Confer-
ence of the IEEE, volume 4, pages 1715–1716, Oct 1996.

[527] N. Stergiopulos, B.E. Westerhof, and N. Westerhof. Total arterial inertance as the
fourth element of the windkessel model. American Journal of Physiology-Heart and
Circulatory Physiology, 276(1):H81–H88, 1999.

[528] J.M.A. Stijnen, J. de Hart, P.H.M. Bovendeerd, and F.N. van de Vosse. Evaluation
of a fictitious domain method for predicting dynamic response of mechanical heart
valves. International Journal for Numerical Methods in Fluids, 19(6):835–850, 2004.

[529] J.S. Stroud, S.A. Berger, and D. Saloner. Numerical Analysis of Flow Through a
Severely Stenotic Carotid Artery Bifurcation. J Biomech Eng, 124(1):9–20, 2002.

[530] A.M. Stuart. Inverse problems: A Bayesian perspective. Acta Numerica, 19:451–559,
2010.

[531] T.J. Sullivan. Introduction to Uncertainty Quantification, volume 63 of Texts in

Applied Mathematics. Springer International Publishing Switzerland, 2015.

[532] W. Sun, B. Starly, J. Nam, and A. Darling. Bio-cad modeling and its applications
in computer-aided tissue engineering. Computer-Aided Design, 11:1097–1114, 2005.

[533] G.B.K. Sundaram, K.R. Balakrishnan, and R.K. Kumar. Aortic valve dynamics
using a fluid structure interaction model - the physiology of opening and closing.
Journal of Biomechanics, 48(10):1737–1744, 2015.

[534] J. Sundnes, G.T. Lines, and A. Tveito. An operator splitting method for solving the
bidomain equations coupled to a volume conductor model for the torso. Mathematical
Biosciences, 2(194):233–248, 2005.

[535] J. Sundnes, S. Wall, H. Osnes, T. Thorvaldsen, and A.D. McCulloch. Improved

discretisation and linearisation of active tension in strongly coupled cardiac electro-
mechanics simulations. Computer Methods in Biomechanics and Biomedical Engi-
neering, 6(17):604–615, 2014.

[536] E.W. Swim and P. Seshaiyer. A nonconforming finite element method for fluid-
structure interaction problems. Comput. Methods Appl. Mech. Engrg., 195(17–
18):2088–2099, 2006.

[537] A. Tagliabue, L Dedè, and A. Quarteroni. Fluid dynamics of an idealized left ventri-
cle: the extended nitsche’s method for the treatment of heart valves as mixed time
varying boundary conditions. MOX-Report n. 61-2015, Department of Mathematics,
Politecnico di Milano, Italy, 2015.

204
[538] H. Talbot, S. Cotin, R. Razavi, C. Rinaldi, and H. Delingette. Personalization of
Cardiac Electrophysiology Model using the Unscented Kalman Filtering. In Com-
puter Assisted Radiology and Surgery (CARS 2015), Barcelona, Spain, 2015.

[539] A. Tarantola. Inverse Problem Theory and Methods for Model Parameter Estimation.
Society for Industrial and Applied Mathematics, Philadelphia, USA, 2004.

[540] C.A. Taylor and C.A. Figueroa. Patient-specific modeling of cardiovascular mechan-
ics. Annual Review of Biomedical Engineering, 11:109–134, 2009.

[541] C.A. Taylor, T.J.R. Hughes, and C.K. Zarins. Finite element analysis of pulsatile
flow in the abdominal aorta under resting and exercise conditions. American Society
of Mechanical Engineers, Bioengineering Division, 33:81–82, 1996.

[542] C.A. Taylor, T.J.R. Hughes, and C.K. Zarins. Finite element modeling of blood
flow in arteries. Computer Methods in Applied Mechanics and Engineering, 158(1-
2):155–196, 1998.

[543] R. Temam. Sur l’approximation de la solution des équations de Navier–Stokes par

la méthode des pas fractionaires (I). Archives for Rational Mechanics and Analysis,
32:135–153, 1969.

[544] T.E. Tezduyar, S. Sathe, T. Cragin, B. Nanna, B.S. Conklin, J. Pausewang, and
M. Schwaab. Modelling of fluid-structure interactions with the space-time finite
elements: arterial fluid mechanics. Int. J. Num. Methods Fluids, 54:901–922, 2007.

[545] L.J.P. Timmermans, P.D. Minev, and F.N. Van de Vosse. An approximate projection
scheme for incompressible flow using espectral elements. Int. J. Num. Methods
Fluids, 22:673–688, 1996.

[546] K.A. Tomlinson, P.J. Hunter, and A.J. Pullan. A finite element method for an
eikonal equation model of myocardial excitation wavefront propagation. SIAM J.
Appl. Math, 1(63):324–350, 2002.

[547] E.F. Toro. Brain venous haemodynamics, neurological diseases and mathematical
modelling. a review. Applied Mathematics and Computation, 272(2):542579, 2016.

[548] N.A. Trayanova. Defibrillation of the heart: insights into mechanisms from modelling
studies Authors. Experimental physiology, 91(2):323–337, 2006.

[549] N.A. Trayanova, W. li, J. Eason, and P. Kohl. Effect of stretch-activated channels
on defibrillation efficacy. Heart Rhythm, 1(1):67–77, 2004.

[550] P.R. Trenhago, L.G Fernandes, L.O. Müller, P.J. Blanco, and R.A. Feijóo. An
integrated mathematical model of the cardiovascular and respiratory systems. Int.
J. Numer. Meth. Biomed. Engng., 32(1):e02736–n/a, 2016. e02736 cnm.2736.

205
[551] F. Tröltzsch. Optimal control of partial differential equations: theory, methods and
applications, volume 112 of Graduate Studies in Mathematics. American Mathemat-
ical Society, Providence, 2010.

[552] C. Tu and C. Peskin. Stability and instability in the computation of flows with
moving immersed boundaries: A comparison of three methods. SIAM J. Sci. and
Stat. Comput., 6(13):13611376, 1992.

[553] S. Turek. Efficient Solvers for Incompressible Flow Problems. Lecture Notes in
Computational Science and Engineering. Springer, 1999.

[554] K.H.W.J. Ten Tusscher and A.V. Panfilov. Cell model for efficient simulation of wave
propagation in human ventricular tissue under normal and pathological conditions.
Physics in Medicine and Biology, 51(23):6141–6156, 2006.

[555] M. Unser. Splines: a perfect fit for signal and image processing. IEEE Transactions
on Signal Processing Magazine, 16(6):22–38, 1999.

[556] T.P. Usyk, I.J. LeGrice, and A.D. McCulloch. Computational model of three-
dimensional cardiac electromechanics. Computing and Visualization in Science,
4(4):249–257, 2002.

[557] R.J. van der Geest, E. Jansen, V. Buller, and J. Reiber. Automated detection of
left ventricular epi- and endocardial contours in short-axis mr images. Computers
in Cardiology, Proceedings, pages 33–36, 1994.

[558] A. Veneziani. Boundary conditions for blood flow problems. Proceedings of ENU-
MATH, Rannacher et al. eds., World Sci. Publishing, River Edge, NJ, 1998.

[559] A. Veneziani. Mathematical and Numerical Modeling of Blood Flow Problems. PhD
thesis, University of Milan, 1998.

[560] A. Veneziani. Block factorized preconditioners for high-order accurate in time ap-
proximation of the navier-stokes equations. Numerical Methods for Partial Differ-
ential Equations, 19(4):487–510, 2003.

[561] A. Veneziani and C. Vergara. Flow rate defective boundary conditions in haemodi-
namics simulations. Int. J. Num. Methods Fluids, 47:803–816, 2005.

[562] A. Veneziani and C. Vergara. An approximate method for solving incompressible

Navier-Stokes problems with flow rate conditions. Comput. Methods Appl. Mech.
Engrg., 196(9-12):1685–1700, 2007.

[563] A. Veneziani and C. Vergara. Inverse problems in Cardiovascular Mathematics:

toward patient-specific data assimilation and optimization. Int. J. Numer. Meth.
Biomed. Engng., 29(7):723–725, 2013.

[564] A. Veneziani and U. Villa. Aladins: An algebraic splitting time adaptive solver for
the incompressible navierstokes equations. J Comp Phys, 238:359–375, 2013.

206
[565] C. Vergara. Nitsche’s method for defective boundary value problems in incompress-
ibile fluid-dynamics. J Sci Comp, 46(1):100–123, 2011.
[566] C. Vergara, M. Lange, S. Palamara, T. Lassila, A.F. Frangi, and A. Quarteroni. A
coupled 3D-1D numerical monodomain solver for cardiac electrical activation in the
myocardium with detailed Purkinje network. J. Comput. Phys, 308:218–238, 2016.
[567] C. Vergara, S. Palamara, D. Catanzariti, F. Nobile, E. Faggiano, C. Pangrazzi,
M. Centonze, M. Maines, A. Quarteroni, and G. Vergara. Patient-specific generation
of the Purkinje network driven by clinical measurements of a normal propagation.
Med. Biol. Eng. Comput., 52(10):813–826, 2014.
[568] C. Vergara, R. Ponzini, A. Veneziani, A. Redaelli, D. Neglia, and O. Parodi. Womer-
sley number-based estimation of flow rate with doppler ultrasound: Sensitivity anal-
ysis and first clinical application. Computer Methods and Programs in Biomedicine,
98(2):151–160, 2010.
[569] C. Vergara, F. Viscardi, L. Antiga, and G.B. Luciani. Influence of bicuspid valve
geometry on ascending aortic fluid-dynamics: a parametric study. Artificial Organs,
36(4):368–378, 2012.
[570] J. A. Vierendeels, K. Riemslagh, E. Dick, and P.R. Verdonck. Computer simulation
of intraventricular flow and pressure gradients during diastole. Journal of Biome-
chanical Engineering, 6(122):667–674, 2000.
[571] E.J. Vigmond, F. Aguel, and N.A. Trayanova. Computational techniques for solv-
ing the bidomain equations in three dimensions. IEEE Transaction on Biomedical
Engineering, 49(11):1260–1269, 2002.
[572] E.J. Vigmond and C. Clements. Construction of a computer model to investigate
sawtooth effects in the Purkinje system. IEEE Trans Biomed Eng, 54(3):389–399,
2007.
[573] E.J. Vigmond, R. Weber dos Santos, A.J. Prassl, M. Deo, and G. Plank. Solvers
for the cardiac bidomain equations. Progress in Biophysics and Molecular Biology,
96(1–3):3–18, 2008.
[574] I.E. Vignon-Clementel, C.A. Figueroa, K. Jansen, and C. Taylor. Outflow boundary
conditions for three-dimensional finite element modeling of blood flow and pressure
waves in arteries. Comput. Methods Appl. Mech. Engrg., 195:3776–3996, 2006.
[575] N. Virag, V. Jacquemet, C.S. Henriquez, S. Zozor, O. Blanc, J.-M. Vesin, E. Pruvot,
and L. Kappenberger. Study of atrial arrhythmias in a computer model based on
magnetic resonance images of human atria. Chaos, 12:754, 2002.
[576] F. Viscardi, C. Vergara, L. Antiga, S. Merelli, A. Veneziani, G. Puppini, G. Fag-
gian, A. Mazzucco, and G.B. Luciani. Comparative finite element model analysis
of ascending aortic flow in bicuspid and tricuspid aortic valve. Artificial organs,
34(12):1114–20, December 2010.

207
[577] H. U. Voss, J. Timmer, and J. Kurths. Nonlinear dynamical system identification
from uncertain and indirect measurements. Int. J. Bifurc. Chaos, 14(6):1905–1933,
2004.

[578] J. Voss. An introduction to Statistical Computing: A Simulation-Based Approach.

Wiley, 2013.

[579] J. Vossoughi, R.N. Vaishnav, and D.J. Patel. Compressibility of the myocardial
tissue. Adv Bioeng (ed V.C. Mow), pages 45–48, 1980.

[580] E. Votta, T.B. Le, M. Stevanella, L. Fusinic, E.G. Caiani, A. Redaelli, and
F. Sotiropoulos. Toward patient-specific simulations of cardiac valves: State-of-
the-art and future directions. Journal of Biomechanics, 46(2):217–228, 2013.

[581] E. Votta, F. Maisano, S.F. Bolling, O. Alfieri, F.M. Montevecchi, and A. Redaelli.
The geoform disease-specific annuloplasty system: a finite element study. The Annals
of Thoracic Surgery, 84(1):92–101, 2007.

[582] G.D. Waiter, F.I. McKiddie, T.W. Redpath, S.I.K. Semple, and R.J. Trent. Deter-
mination of normal regional left ventricular function from cine-mr images using a
semi-automated edge detection method. Magnetic Resonance Imaging, 17(1):99–107,
1999.

[583] W. Wall, L. Wiechert, A. Comerford, and S. Rausch. Towards a comprehensive

computational model for the respiratory system. International Journal for Numerical
Methods in Biomedical Engineering, 26(7):807827, 2010.

[584] M. Wallman, N.P. Smith, and B. Rodriguez. Computational methods to reduce un-
certainty in the estimation of cardiac conduction properties from electroanatomical
recordings. Med. Image Anal., 18(1):228–240, 2014.

[585] D. Wang, R.M. Kirby, R. MacLeod, and C.R. Johnson. Inverse electrocardiographic
source localization of ischemia: An optimization framework and finite element solu-
tion. J. Comput. Phys., 250:403–424, 2013.

[586] K.C. Wang, R.W. Dutton, and C.A. Taylor. Improving geometric model construction
for blood flow modeling. Engineering in Medicine and Biology Magazine, IEEE,
18(6):33–39, 1999.

[587] L. Wang, K.C. L. Wong, H. Zhang, H. Liu, and P. Shi. Noninvasive computational
imaging of cardiac electrophysiology for 3-d infarct. IEEE Trans. Biomed. Engng.,
58(4):1033–1043, 2011.

[588] L. Wang, H. Zhang, K.C.L. Wong, and P. Shi. A reduced-rank square root filter-
ing framework for noninvasive functional imaging of volumetric cardiac electrical
activity. In 2009 IEEE International Conference on Acoustics, Speech and Signal
Processing, pages 533–536, 2009.

208
[589] H. Watanabe, T. Hisada, S. Sugiura, J. Okada, and H. Fukunari. Computer sim-
ulation of blood flow, left ventricular wall motion and their interrelationship by
fluid-structure interaction finite element method. JSME International Journal Se-
ries C Mechanical Systems, Machine Elements and Manufacturing, 45(4):1003–1012,
2002.

[590] D.F. Watson. Computing the n-dimensional Delaunay tessellation with application
to voronoi polytopes. Computer Journal, 24(2):167–172, 1981.

[591] N.P. Weatherill and O. Hassan. Efficient three-dimensional Delaunay triangulation

with automatic point creation and imposed boundary constraints. International
Journal for Numerical Methods in Engineering, 37(12):2005–2039, 1994.

[592] E.J. Weinberg and M.R. Kaazempur-Mofrad. A large-strain finite element formula-
tion for biological tissues with application to mitral valve leaflet tissue mechanics.
Journal of Biomechanics, 39(8):15571561, 2006.

[593] J.F. Wenk, L. Ge, Z. Zhang, M. Soleimani, D.D. Potter, A.W. Wallace, E. Tseng,
M.B. Ratcliffe, and J.M. Guccione. Mechanics of the mitral valve strut chordae
insertion region. Computer Methods in Biomechanics and Biomedical Engineering,
16(8):807–818, 2013.

[594] N. Westerhof, J.W. Lankhaar, and B.E. Westerhof. The arterial windkessel. Medical
& biological engineering & computing, 47(2):131–141, 2009.

[595] J. Wong, S. Goktepe, and E. Kuhl. Computational modeling of chemo-electro-

mechanical coupling: A novel implicit monolithic finite element approach. Interna-
tional Journal for Numerical Methods in Biomedical Engineering, 29(10):1104–1133,
2013.

[596] J. Wong and E. Kuhl. Generating fibre orientation maps in human heart models
using poisson interpolation. Computer Methods in Biomechanics and Biomedical
Engineering, 17(11):1217–1226, 2014.

[597] J. Xi, P. Lamata, J.L., P. Moireau, D. Chapelle, and N. Smith. Myocardial trans-
versely isotropic material parameter estimation from in-silico measurements based
on a reduced-order unscented kalman filter. J. Mech. Behav. Biomed. Materials,
4(7):1090–1102, 2011.

[598] F. Xie, Z. Qu, J. Yang, A. Baher, J.N. Weiss, and A. Garfinkel. A simulation study of
the effects of cardiac anatomy in ventricular fibrillation. J Clin Invest, 113:686–693,
2004.

[599] F. Xiong and C. Chong. A parametric numerical investigation on haemodynamics in

distal coronary anastomoses. Medical Engineering & Physics, 30(3):311–320, 2008.

[600] D. Xiu and J. S. Hesthaven. High-order collocation methods for differential equations
with random inputs. SIAM J. Sci. Comput., 27(3):1118–1139, 2005.

209
[601] D. Xiu and G. E. Karniadakis. Modeling uncertainty in steady state diffusion
problems via generalized polynomial chaos. Comput. Methods Appl. Mech. Engrg.,
191:4927–494, 2002.

[602] D. Xiu and G. E. Karniadakis. The Wiener-Askey polynomial chaos for stochastic
differential equations. SIAM J. Sci. Comput., 24:619–644, 2002.

[603] D. Xiu and S.J. Sherwin. Parametric uncertainty analysis of pulse wave propagation
in a model of a human arterial network. J. Comput. Phys., 226(2):1385–1407, 2007.

[604] Y. Yamashita. Theoretical studies on the inverse problem in electrocardiography

and the uniqueness of the solution. IEEE Trans. Biomed. Eng., BME-29:719?–725,
1982.

[605] K. Yanagihara, A. Noma, and H. Irisawa. Reconstruction of sino-atrial node pace-

maker potential based on the voltage clamp experiments. Jpn. J. Physiol, 30:841–
857, 1980.

[606] H. Yang and A. Veneziani. Estimation of cardiac conductivities in ventricular tissue

by a variational approach. Inverse Problems, 31:115001, 2015.

[607] P.J. Yim, J.J. Cebral, R. Mullick, H.B. Marcos, and P.L. Choyke. Vessel surface
reconstruction with a tubular deformable model. IEEE Transactions on Medical
Imaging, 20(12):1411–1421, 2001.

[608] M. Yin, X.Y. Luo, T.J. Wang, and P.N. Watton. Effects of flow vortex on a chorded
mitral valve in the left ventricle. International Journal for Numerical Methods in
Biomedical Engineering, 26(3-4):381–404, 2009.

[609] H.F. Younis, M.R. Kaazempur-Mofrad, R.C. Chan, A.G. Isasi, D.P. Hinton, A.H.
Chau, L.A. Kim, and R.D. Kamm. Hemodynamics and wall mechanics in human
carotid bifurcation and its consequences for atherogenesis: investigation of inter-
individual variation. Biomechanics and Modeling in Mechanobiology, 3(1):17–32,
2004.

[610] Y. Yu, H. Baek, and G.E. Karniadakis. Generalized fictitious methods for fluid-
structure interactions: Analysis and simulations. J. Comput. Physics, 245:317–346,
2013.

[611] S.Z. Zhao, X.Y. Xu, A.D. Hughes, S.A. Thom, A.V. Stanton, B. Ariff, and Q. Long.
Blood flow and vessel mechanics in a physiologically realistic model of a human
carotid arterial bifurcation. Journal of Biomechanics, 33(8):975–984, 2000.

[612] F. Zhu and J. Tian. Modified fast marching and level set method for medical image
segmentation. J Xray Sci Technol, 11(4):193–204, 2003.

[613] O.C. Zienkiewicz and R.L. Taylor. The Finite Element Method for Solid and Struc-
tural Mechanics. Butterworth-Heinemann, 6 edition, 2005.

210
[614] S. Zonca, L. Formaggia, and C. Vergara. An unfitted formulation for the interaction
of an incompressible fluid with a thick structure via an XFEM/DG approach. MOX-
Report n. 35-2016, Department of Mathematics, Politecnico di Milano, Italy, 2016.

[615] P. Zunino. Numerical approximation of incompressible flows with net flux defective
boundary conditions by means of penalty technique. Comput. Methods Appl. Mech.
Engrg., 198(37-40):3026–3038, 2009.

211
 MOX Technical Reports, last issues
Dipartimento di Matematica
Politecnico di Milano, Via Bonardi 9 - 20133 Milano (Italy)

36/2016 Mancini, L.; Paganoni, A.M.

Marked Point Process models for the admissions of heart failured patients

37/2016 Tugnoli, M; Abbà, A. ; Bonaventura, L.; Restelli, M.

A locally p-adaptive approach for Large Eddy Simulation of compressible
flows in a DG framework

34/2016 Menafoglio, A.; Secchi, P.

Statistical analysis of complex and spatially dependent data: a review of
Object Oriented Spatial Statistics

35/2016 Zonca, S.; Formaggia, L.; Vergara, C.

An unfitted formulation for the interaction of an incompressible fluid with a
thick structure via an XFEM/DG approach

33/2016 Antonietti, P. F.; Ferroni, A.; Mazzieri, I.; Quarteroni, A.

hp-version discontinuous Galerkin approximations of the elastodynamics
equation

32/2016 Tarabelloni, N.; Schenone, E.; Collin, A.; Ieva, F.; Paganoni, A.M.; Gerbeau, J.-F.
Statistical Assessment and Calibration of Numerical ECG Models

30/2016 Abramowicz, K.; Häger, C.; Pini, A.; Schelin, L.; Sjöstedt de Luna, S.; Vantini, S.
Nonparametric inference for functional-on-scalar linear models applied to
knee kinematic hop data after injury of the anterior cruciate ligament

31/2016 Antonietti, P.F.; Merlet, B.; Morgan, P.; Verani, M.

Convergence to equilibrium for a second-order time semi-discretization of the
Cahn-Hilliard equation

29/2016 Miglio, E.; Parolini, N.; Penati, M.; Porcù, R.

GPU parallelization of brownout simulations with a non-interacting particles
dynamic model

28/2016 Antonietti, P.F.; Dal Santo, N.; Mazzieri, I.; Quarteroni, A.

A high-order discontinuous Galerkin approximation to ordinary differential
equations with applications to elastodynamics