Volume 13

Mathematical Cardiac
Electrophysiology

Piero Colli Franzone • Luca F. Pavarino • Simone Scacchi

MS&A
Modeling, Simulation & Applications

ABC
MS&A

Volume 13

Editor-in-Chief

A. Quarteroni

Series Editors
T. Hou
C. Le Bris
A.T. Patera
E. Zuazua
More information about this series at
http://www.springer.com/series/8377
Piero Colli Franzone • Luca F. Pavarino •
Simone Scacchi

Mathematical Cardiac
Electrophysiology

123
Piero Colli Franzone Luca F. Pavarino
Dipartimento di Matematica Dipartimento di Matematica
UniversitJa degli Studi di Pavia UniversitJa degli Studi di Milano
Pavia Milano
Italy Italy

Simone Scacchi
Dipartimento di Matematica
UniversitJa degli Studi di Milano
Milano
Italy

ISSN 2037-5255 ISSN 2037-5263 (electronic)

MS&A – Modeling, Simulation & Applications
ISBN 978-3-319-04800-0 ISBN 978-3-319-04801-7 (eBook)
DOI 10.1007/978-3-319-04801-7
Springer Cham Heidelberg New York Dordrecht London
Library of Congress Control Number: 2014952788

© Springer International Publishing Switzerland 2014

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Preface

The goal of this book is to present some of the most successful and advanced
mathematical and numerical models used in the field of cardiac electrophysiology.
The bioelectric activity of the heart is the subject of a vast and still growing
interdisciplinary literature in medicine, physiology, bioengineering, mathematical
biology, chemistry, physics and bioinformatics. The long history and diversity of
this field of research is shown e.g. by the earlier monographs by Jack, Noble and
Tsien [250], Peskin [383], Nelson and Gezelowitz [346], Pilkington and Plonsey
[386], by the reference works by Pilkington et al. [387], Panfilov and Holden [369],
Keener and Sneyd [273], Gulrajani [215], Plonsey and Barr [393], Efimov et al.
[158], and by some of the periodic review books by Zipes and Jalife [588–590].
In the last decade, more specific monographs on computational electrocardiology
were published by Sachse [446], Pullan at el. [406], Sundnes et al. [502]; see
also Macfarlane et al. [312]. Since the emphasis of this book is on mathematical
and numerical aspects of the models and algorithms presented, we hope that this
monograph will present new elements and will complement the works above.
Electrocardiology deals with the description of both intracardiac bioelectric
phenomena and the extracardiac electric field generated in the animal or human
body. The practice of modern medicine relies on noninvasive imaging technologies,
such as CT, MRI and PET, for diagnostic purposes and for driving therapeutic
procedures. Even though cardiac arrhythmias are among the major causes of death
and disability, a noninvasive imaging technique yielding an accurate and reliable
diagnosis of the electrophysiological state of the heart is not yet available. Clinic
Electrocardiography deals with the detection and interpretation of noninvasive
potential measurements collected from the time course of the usual electrocardio-
grams (ECG) at a few points on the body surface or from the evolution of body
surface maps, i.e. potential distribution maps on the body surface reconstructed
from measures at numerous electrodes (100 or more, see the surveys by [504, 511]).
Since the electrode location of the ECG is centimeters away from the heart
surface and the current conduction from heart to thorax yields a strong signal

v
vi Preface

attenuation and smoothing, the information content of ECGs and body maps is
limited and it is a difficult task to extract from these signals detailed information
on pathological heart states associated with ischemia or sudden death. Indeed, the
origin of arrhythmogenic activity or the existence of abnormal electrophysiological
substrates in many cases may not be easily inferred from the sequence of cardiac
excitation.
The scientific base of Electrocardiology is the so-called Forward Problem of
Electrocardiology, i.e. modeling the bioelectric cardiac sources and the conducting
media in order to derive the potential field. Of considerable applicative interest are
also the so-called Inverse Problems of Electrocardiography in terms of potentials
(see e.g. the review [216, 442] and [68, 69, 415]) or in terms of the cardiac sources
(see e.g. [110, 439]).
In the past few decades, experimental electrophysiology has been increasingly
supported by the mathematical and numerical models of computational electro-
cardiology. The formulation of models at both cellular and tissue levels provide
essential tools in order to integrate the increasing knowledge of the bioelectrochem-
ical phenomena occurring through cardiac cellular membranes. Detailed cellular
phenomena are described in microscopic membrane models and the latter are then
inserted in macroscopic tissue models in order to investigate their effects at tissue
level. These coupled models are then validated by comparing simulated results with
experimental in vitro and in vivo data, generating a feedback loop that may lead to
improved and more detailed models and/or the redesign of new experiments. As a
further step, these electrophysiological models are being increasingly coupled and
integrated with mechanical models of tissue deformation, hemodynamical models
of cardiac blood flow and more in general with models of the cardiovascular
system. This complex integrative effort is the current focus of several research
projects, for example such as the Physiome Project (www.physiome.org.nz) and the
EC-sponsored Virtual Physiological Human (VPH) Initiative (www.vph-noe.eu).
Ultimately, the integration of these models should provide new tools enabling the
biomedical community to link genetic and proteomic databases to anatomy and to
functions at the cellular, tissue and organ level.
From a macroscopic point of view, the Forward Problem of Electrocardiology
is described by the so-called Bidomain model for the evolution of the intra,
extracellular and extracardiac potential fields. The two main components of the
Bidomain model are: (a) the dynamics of the ionic current flow through the cardiac
cellular membrane, modeled by a system of ordinary differential equations and
(b) a macroscopic representation of the cardiac tissue modeled as a bidomain
superposition of the intra and extra cellular media characterized by anisotropic
conductivity tensors associated with the fiber architecture of the myocardium.
The Bidomain model is computationally expensive because of the involvement of
different space and time scales. In fact, meaningful portions of cardiac tissue have
sizes on the order of centimeters, while the steep potential gradient is localized in
a thin layer about 1 mm thick, requiring discretizations on the order of a tenth of
millimeter. Moreover, a normal heartbeat can last on the order of 1 s, while the
time constants of the rapid kinetics involved range from 0.1 to 500 ms, requiring in
Preface vii

some phases time steps on the order of the hundredths of milliseconds (or less when
currents or shocks are applied). Therefore, in realistic three-dimensional models it is
possible to have discrete problems with more than O.107 / unknowns at every time
step and simulations have to be run for many thousands of time steps.
A simplified cardiac tissue model is the anisotropic Monodomain system, i.e. a
parabolic reaction-diffusion equation describing the evolution of the transmembrane
potential coupled with an ionic membrane model. This model has been widely used
for three-dimensional simulations due to its reduced computational costs.
Current large-scale simulations of whole heartbeats using Bidomain and Mon-
odomain models require adaptive and parallel tools in order to reduce their high
computational cost. While both tools can in principle be applied to both space and
time, most studies employ adaptive methods in time and parallel solvers in space,
since the other alternatives are still the subject of current research even for simpler
model problems in two dimensions. Therefore in this book, we present the main
numerical techniques for efficiently simulating cardiac reaction-diffusion models.
In particular, we focus on scalable parallel Bidomain solvers that are capable of
efficiently scaling their performance for increasing processor counts in current and
future multicore parallel computers.
Among the important aspects of cardiac modeling not covered in this book are
cardiac mechanics, blood flow, electro-mechanical and fluid-mechanical coupling,
and cardiac imaging. Research in these fields is also growing tremendously and a
separate book would be necessary to properly present the main mathematical and
numerical models available. For an overview of these related fields, we refer to e.g.
the monographs [47, 140, 242, 369, 446], the works [138, 232, 246, 384, 385, 491]
with the references therein, and the recent proceedings of the conferences FIMH
(Functional Imaging and Modeling of the Heart) [21, 180, 266, 316, 327, 366, 447],
CINC (Computing in Cardiology, http://cinc.org/archives/2013/, http://cinc.org/
archives/2012/), STACOM (Statistical Atlases and Computational Models of the
Heart) [74–76].
The book is structured in the following chapters.
In Chap. 1, we give a brief review of the basic physiology and anatomy
of the heart, including the specialized cells of the cardiac conduction system,
working cardiomyocytes, fibroblasts, extracellular matrix, collagen, gap junctions,
connexin, cardiac stem cells and the fiber and laminar architecture of the ventricular
myocardium. We then present the main phases of a cardiac action potential, its
spatial and temporal heterogeneity, and continue by describing the main features
of an electrocardiogram (ECG), with its leads, deflections, intervals and main
alterations. The chapter concludes with a review of the main cardiac imaging
techniques currently available.
Chapter 2, introduces the fundamental tools for modeling the bioelectric activity
of excitable cells: the Nernst – Planck equation, the Goldman-Hodgkin-Katz (GHK)
current-voltage relation, and the Nernst equilibrium potential, together with its ther-
modynamical derivation. Next, the Poisson-Nernst-Planck (PNP) electrodiffusion
model is derived and two classical current – voltage relations are obtained in the
short and long channel limits. With these tools, we can define the basic electrical
viii Preface

circuit model of the cellular membrane, where the transmembrane current, modeled
as the sum of the capacitive and ionic currents through the membrane, must balance
the given applied current. The ionic currents are then described by using the classical
ion channel gating models, allowing us to build cardiac action potential models. We
start with the celebrated Hodgkin-Huxley (H-H) model and briefly review some of
the historical ventricular models based on the H-H formalism, such as the Beeler-
Reuter, Luo-Rudy I and Luo-Rudy dynamic models, examining also how these ionic
models satisfy the principle of charge conservation and how to derive the so-called
restitution curve for the action potential duration of a given ionic model. Reduced
models, such as the minimal FitzHugh-Nagumo model are also presented, together
with their phase-plane analysis, bifurcation and frequency diagrams.
Chapter 3 presents mathematical models of periodic cardiac cells arrangements,
beginning with one-dimensional fibers, deriving the cable equation, showing a one-
dimensional homogenization technique and the main results on one-dimensional
traveling waves, namely traveling fronts for the bistable equation and traveling
pulses for the FitzHugh-Nagumo system with a diffusion term. We then move to
models of cardiac tissue in more dimensions, illustrating a two-scale homogeniza-
tion technique that allow us to derive an averaged Bidomain model, proving both
well-posedness results for the cellular and the averaged models and convergence
results based on  -convergence techniques. We then present an heuristic derivation
of the anisotropic Bidomain model in both parabolic-parabolic and parabolic-elliptic
forms. Well-posedness results are derived using different techniques, such as time
semi-discretization, Faedo-Galerkin techniques, and fixed point arguments.
Chapter 4 presents the main reduced macroscopic cardiac models: the linear
anisotropic Monodomain model, Eikonal models (both Eikonal-curvature and
Eikonal-diffusion models), and the relaxed non-linear anisotropic Monodomain
model. These model are then given in dimensional form and some well-posedness
results are summarized. The chapter is concluded by a numerical comparison
between activation time maps computed with these reduced models and the full
Bidomain model.
Chapter 5 is devoted to the modeling of anisotropic cardiac sources, presenting
both the differential and integral formulations of the potential field. Approximate
representations of cardiac sources such as the heart surface and oblique dipole
source models are given, as well as the cardiac sources splitting into axial and
conormal components in both axially symmetric and orthotropic media. The
chapter concludes with a numerical example illustrating this source splitting and
its comparison with experimental results.
Chapter 6 briefly reviews the Inverse problem of Electrocardiology, in terms of
cardiac sources, in terms of wavefront and in terms of potential alone, presenting
the mathematical models of the cardiac electric sources and their numerical
approximations.
Chapter 7 presents the main numerical techniques employed in the space
and time discretizations of the Monodomain and Bidomain cardiac models. In
particular, we apply the finite element method in space and finite difference
methods in time. The latter can be fully implicit or semi-implicit, and can employ
Preface ix

decoupling techniques and operator splitting methods between the ordinary and
partial differential equation components of the cardiac models. The chapter is
concluded by a review of numerical methods for the eikonal–diffusion equation.
Chapter 8 is devoted to the construction and analysis of parallel solvers for the
discrete Bidomain systems arising at each time step of an implicit or semi-implicit
time discretization. Our parallel solvers are based on domain decomposition meth-
ods, more specifically on overlapping Schwarz methods, which provide scalable
preconditioners accelerated with a Krylov space iterative method such as PCG or
GMRES. After recalling the main results of the abstract Schwarz theory, we derive
scalable convergence rate bounds for two-level and multilevel additive Schwarz
preconditioners for the Bidomain system. We then present the results of several
numerical tests with these Schwarz preconditioners in additive, multiplicative and
hybrid form, showing their scalability on different parallel machines and inves-
tigating their performance with respect to the different discretization parameters.
We also present how these Schwarz preconditioners can be combined with block-
diagonal and block-factorized preconditioners suggested by the 22 block structure
of discrete Bidomain systems.
Chapter 9 illustrates how to apply the Bidomain and Monodomain solvers
developed in the previous chapters to simulate and study some of the most
important phenomena in cardiac electrophysiology. More precisely, we present
detailed simulations of: (1) the genesis of cardiac excitation and virtual electrode
phenomena, in particular anode/cathode make/break and strength-interval (S-I)
curves; (2) the anisotropic propagation of excitation and recovery fronts in three
dimensional domains; (3) the effects of cardiac heterogeneities (transmural and
apico-basal) on fronts propagation and APD distribution; (4) the morphology of
electrograms, in particular of the QRS complex and the T wave; (5) the computation
of excitation and repolarization time markers; (6) the presence of ischemic regions
and effects such as ST-segment depression and elevation; (7) the simulation of
cardiac reentry phenomena.
An Appendix lists some of the main cardiac simulation research projects,
software libraries, some related monographs and tables of physical units and
constants used in the book.
The authors would like to thank Prof. Bruno Taccardi for introducing them
to the field of Mathematical Physiology and for many stimulating discussions,
Prof. Alfio Quarteroni for his encouragement throughout this project, Dr. Fabrizio
del Bianco and Dr. Lara Charawi for their help in proofreading the manuscript.

Pavia, Italy Piero Colli Franzone

Milano, Italy Luca F. Pavarino
Milano, Italy Simone Scacchi
April 2014
Contents

1 Basic Cardiac Anatomy and Electrocardiology . . . . . .. . . . . . . . . . . . . . . . . . . . 1

1.1 Conduction System: SA and AV Node, Purkinje Network .. . . . . . . . . . 2
1.2 Cardiac Tissue Organization . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 4
1.3 Fiber and Laminar Architecture of Ventricular Myocardium . . . . . . . . 7
1.4 Cardiac Action Potentials . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 8
1.4.1 Action Potential Phases . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 8
1.4.2 Action Potential Heterogeneity .. . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 10
1.5 The Electrocardiogram (ECG) .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 11
1.5.1 ECG Leads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 12
1.5.2 ECG Deflections and Intervals . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 13
1.5.3 ECG Diagnosis .. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 15
1.6 Cardiac Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 17
2 Mathematical Models of Cellular Bioelectrical Activity .. . . . . . . . . . . . . . . . 21
2.1 Excitable Cellular Membranes .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 21
2.2 The Nernst-Planck Equation . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 22
2.3 The Goldman-Hodgkin-Katz (GHK) Current-Voltage Relation . . . . . 23
2.4 Nernst Equilibrium Potential . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 24
2.5 Thermodynamical Derivation of the Nernst Potential . . . . . . . . . . . . . . . . 25
2.6 Electrodiffusion Models: The Poisson-Nernst-Planck
(PNP) Equation .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 28
2.6.1 PNP: The Short Channel or Low Concentrations Limit .. . . . . 31
2.6.2 PNP: The Long Channel or High Concentrations Limit . . . . . 33
2.6.3 Equilibrium Potential for Multi-ion Fluxes . . . . . . . . . . . . . . . . . . . 34
2.7 Electrical Circuit Model of the Cellular Membrane . . . . . . . . . . . . . . . . . . 35
2.8 Ion Channel Gating .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 36
2.9 Cardiac Action Potential Models . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 40
2.9.1 Hodgkin-Huxley Model.. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 41
2.9.2 General Structure of Cardiac Cellular Membrane Models . . . 42

xi
xii Contents

2.9.3 Ionic Models of Purkinje Fibers, Sinoatrial Node

(SAN), Atria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 43
2.9.4 Ventricular Models .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 43
2.9.5 Charge Conservation in Ionic Models .. . . .. . . . . . . . . . . . . . . . . . . . 55
2.9.6 Action Potential Duration Restitution Curve . . . . . . . . . . . . . . . . . 59
2.9.7 Reduced Ionic Models . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 59
2.9.8 Phase-Plane Analysis of the FitzHugh-Nagumo
(FHN) Model .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 62
2.9.9 Bifurcation Diagrams . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 70
3 Mathematical Models of Cardiac Cells Arrangements:
The Bidomain Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 77
3.1 Models of Cardiac Fibers . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 77
3.1.1 Cable Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 77
3.1.2 Homogenization .. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 80
3.1.3 Traveling Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 86
3.1.4 Conduction Velocity Restitution Curve . . .. . . . . . . . . . . . . . . . . . . . 92
3.2 Models of Cardiac Tissue . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 93
3.2.1 The Dimensionless Cellular Model P " . . .. . . . . . . . . . . . . . . . . . . . 97
3.2.2 Formal Two-Scale Homogenization .. . . . . .. . . . . . . . . . . . . . . . . . . . 98
3.2.3 The Dimensionless Averaged Model P . . .. . . . . . . . . . . . . . . . . . . . 100
3.2.4 Theoretical Results for the Cellular and Averaged
Models .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 101
3.2.5  -Convergence Result for the Averaged Model
with FHN Dynamics . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 103
3.3 The Macroscopic Anisotropic Bidomain Model ... . . . . . . . . . . . . . . . . . . . 106
3.4 Well-Posedness Results Based on Semi-discretization in Time . . . . . . 108
3.5 Well-Posedness Results Based on Faedo-Galerkin Techniques . . . . . . 112
3.6 Well-Posedness Results Based on Fixed Point Arguments .. . . . . . . . . . 116
3.7 Semi-discrete Approximation of the Bidomain Model
with FHN Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 120
4 Reduced Macroscopic Models: The Monodomain
and Eikonal Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 123
4.1 Linear Anisotropic Monodomain Model . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 123
4.2 Eikonal Models .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 125
4.3 Relaxed Non-linear Anisotropic Monodomain Model . . . . . . . . . . . . . . . 132
4.4 Dimensional Form of the Reduced Models . . . . . . .. . . . . . . . . . . . . . . . . . . . 133
4.4.1 Well-Posedness Results for Reduced Models.. . . . . . . . . . . . . . . . 136
4.4.2 Frank and Wulff Diagrams . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 136
4.5 Numerical Comparison.. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 143
5 Anisotropic Cardiac Sources . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 149
5.1 Differential Formulation of the Potential Field . . .. . . . . . . . . . . . . . . . . . . . 150
5.2 Integral Formulation of the Potential Field . . . . . . . .. . . . . . . . . . . . . . . . . . . . 152
Contents xiii

5.3 Approximate Representation of Cardiac Sources .. . . . . . . . . . . . . . . . . . . . 156

5.3.1 Heart Surface Source Model.. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 157
5.3.2 Oblique Dipole Source Model .. . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 159
5.4 Cardiac Source Splitting . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 162
5.4.1 Axially Symmetric Media . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 162
5.4.2 Orthotropic Media.. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 165
5.5 Interpretation of the Field Components . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 166
5.6 The Limit Case: Oblique Dipole Layer Model. . . .. . . . . . . . . . . . . . . . . . . . 168
5.7 Experimental and Simulation Results . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 170
6 The Inverse Problem of Electrocardiology .. . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 175
6.1 Inverse Problem in Terms of Potential Alone . . . . .. . . . . . . . . . . . . . . . . . . . 176
6.2 Macroscopic Equivalent Excitation Cardiac Sources .. . . . . . . . . . . . . . . . 180
6.3 Boundedness of the Potential . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 184
6.4 Numerical Approximation of the Integral Representation
of the Potential.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 185
6.5 Inverse Problem in Terms of Wavefront .. . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 187
7 Numerical Methods for the Bidomain and Reduced Models . . . . . . . . . . . . 191
7.1 Space Discretization of Monodomain and Bidomain Models .. . . . . . . 191
7.2 Time Discretization of Monodomain and Bidomain Models.. . . . . . . . 194
7.2.1 Fully Implicit Methods . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 194
7.2.2 Decoupled Implicit Methods . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 195
7.2.3 Decoupled Semi-implicit Methods .. . . . . . .. . . . . . . . . . . . . . . . . . . . 196
7.2.4 Operator Splitting Methods: Splitting ODEs and PDEs. . . . . . 203
7.3 Numerical Approximation of the Eikonal-Diffusion Equation .. . . . . . 204
8 Parallel Solvers for the Bidomain System . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 207
8.1 Bidomain Variational Setting . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 208
8.2 Abstract Convergence Theory for Schwarz Methods .. . . . . . . . . . . . . . . . 215
8.3 Two-Level Additive Schwarz Methods for the Bidomain System . . . 220
8.4 Multilevel Additive Schwarz Methods for the Bidomain System .. . . 223
8.5 Numerical Results for Multilevel Schwarz Preconditioners .. . . . . . . . . 227
8.5.1 Additive Preconditioner .. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 228
8.5.2 Multiplicative and Hybrid Preconditioners . . . . . . . . . . . . . . . . . . . 234
8.6 Block Preconditioners for the Bidomain System .. . . . . . . . . . . . . . . . . . . . 241
8.6.1 Block-Diagonal and Block-Factorized Bidomain
Preconditioners .. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 241
8.6.2 Numerical Results with Block Preconditioners .. . . . . . . . . . . . . . 244
9 Simulation Studies of Cardiac Bioelectrical Activity . . . . . . . . . . . . . . . . . . . . 249
9.1 Cardiac Excitation and Virtual Electrode Phenomena .. . . . . . . . . . . . . . . 250
9.1.1 Methods and Parameter Calibration .. . . . . .. . . . . . . . . . . . . . . . . . . . 252
9.1.2 Anode and Cathode Make Mechanisms.. .. . . . . . . . . . . . . . . . . . . . 254
9.1.3 Anode and Cathode Break Mechanisms . .. . . . . . . . . . . . . . . . . . . . 262
9.1.4 Cathodal and Anodal Strength-Interval S-I Curves.. . . . . . . . . . 268
xiv Contents

9.2 Anisotropic Propagation of Excitation and Recovery Fronts.. . . . . . . . 270

9.2.1 Excitation and Repolarization Sequences .. . . . . . . . . . . . . . . . . . . . 271
9.2.2 Discussion on APD Distribution and Dispersion .. . . . . . . . . . . . 279
9.3 Heterogeneous Cardiac Tissue . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 284
9.3.1 Transmural Heterogeneity in 3-D Cardiac Slabs . . . . . . . . . . . . . 287
9.3.2 Transmural Heterogeneity in 3-D Ellipsoids.. . . . . . . . . . . . . . . . . 298
9.3.3 Transmural and Apex-Base Heterogeneity in 3-D
Ellipsoids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 309
9.4 QRS Complex and T Wave Morphology in Electrograms.. . . . . . . . . . . 315
9.4.1 Methods and Parameter Calibration .. . . . . .. . . . . . . . . . . . . . . . . . . . 316
9.4.2 Unipolar and Bipolar ECG Simulations . .. . . . . . . . . . . . . . . . . . . . 318
9.5 Extracellular Markers of Excitation and Repolarization Times . . . . . . 323
9.5.1 Waveform Postprocessing and Repolarization
Time Markers.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 326
9.5.2 Parameter Calibrations for the Model Simulations .. . . . . . . . . . 329
9.5.3 Global Quantitative Analysis of RT Markers . . . . . . . . . . . . . . . . . 332
9.6 Subendocardial Ischemia, ST Depression and Elevation .. . . . . . . . . . . . 339
9.6.1 Mechanisms for the ST Segment Potential Patterns .. . . . . . . . . 339
9.6.2 Ischemic Simulations . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 343
9.7 Reentry Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 348
9.7.1 Stable Scroll Waves . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 352
9.7.2 Scroll Waves Breakup .. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 353
9.7.3 Scroll Waves in Ellipsoidal Geometry . . . .. . . . . . . . . . . . . . . . . . . . 357

A Cardiac Simulation Projects, Software, Libraries . . .. . . . . . . . . . . . . . . . . . . . 361

A.1 IUPS Physiome Project . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 361
A.2 Virtual Physiological Human (VPH) . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 362
A.3 NSR Physiome.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 362
A.4 Other Simulation Software and Modeling Environments . . . . . . . . . . . . 363
A.5 Some Related Monographs . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 364
A.6 Physical Units and Constants . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 366

References .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 367

Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 395
Chapter 1
Basic Cardiac Anatomy and Electrocardiology

In this chapter, we briefly review the basic physiology and anatomy of the heart; a
more complete treatment with many more details can be found in standard textbooks
such as [189,267]. The heart is a double pump consisting of four chambers, two atria
in the upper part, separated by the interatrial septum and two ventricles in the lower
part, separated by the interventricular septum. Atria and ventricles are separated by
the atrioventricular septum, which contains the tricuspid valve in the right heart and
the mitral valve in the left heart. The right ventricle is connected to the pulmonary
artery via the pulmonary valve and the left ventricle is connected to the aorta via
the aortic valve, see Fig. 1.1 for a schematic view and [189] for more details. The
ventricular walls are externally lined by a thin connective layer called epicardium
and internally by the endocardium, with the cardiac muscle called myocardium in
between. The left ventricular wall is about three times thicker than the right one,
while the atrial walls are considerably thinner. The right heart functions as a pump
driving blood through the pulmonary circulation, to the lungs and then back to the
heart, while the left heart functions as another pump driving the oxygenated blood
through the systemic circulation around the body. These two main cardiac functions,
mechanical and haemodynamical, are driven and coordinated by the electrical
activity of the heart, which is the main focus of this monograph. In this chapter,
we will briefly review the essential features of the cardiac bioelectrical activity. The
main events in a normal cardiac cycle are often described by a Wiggers diagram,
see Fig. 1.2, showing the simultaneous time course of blood pressure (aortic, atrial,
ventricular), ventricular volume, electrocardiogram, and phonocardiogram during a
normal cardiac cycle.

© Springer International Publishing Switzerland 2014 1

P. Colli Franzone et al., Mathematical Cardiac Electrophysiology, MS&A 13,
DOI 10.1007/978-3-319-04801-7__1
2 1 Basic Cardiac Anatomy and Electrocardiology

Fig. 1.1 Schematic diagram of the heart anatomy (From [240])

1.1 Conduction System: SA and AV Node, Purkinje Network

As in other muscle cells, the contraction of cardiac cells is initiated by an electrical

activation due to an action potential, a depolarizing transitory membrane current
that raises the transmembrane potential of an excitable cell from its resting value
ranging between 90 and 80 mV to slightly positive values, followed by a
repolarizing current that returns the transmembrane potential to its resting value.
Differently from skeletal muscle cells, cardiac cells are able to autonomously
activate, independently of a nervous stimulus. The electrical activity of the heart
originates at the sinoatrial node (SAN), a group of cardiac pacemaker cells located
on the right atrium, see Fig. 1.3 and [444] for more details. SAN cells have the
fastest spontaneous depolarization speed of the cardiac conduction system and
therefore they control the cardiac frequency in normal conditions. The autonomic
nervous system and the endocrine system have direct influence over the SAN
1.1 Conduction System: SA and AV Node, Purkinje Network 3

Fig. 1.2 Wiggers diagram of a normal cardiac cycle (From [239])

Fig. 1.3 Schematic diagram of the heart conduction system

cells and they can modulate both beat-to-beat and long-term variations of the
cardiac frequency. In normal conditions, SAN cells generate an action potential
that propagates throughout the right atrium and through Bachmann’s bundle to the
left atrium, stimulating the myocardium of both atria to contract. The activation
front reaches the atrioventricular node (AVN) located at the base of the atria
4 1 Basic Cardiac Anatomy and Electrocardiology

where the interatrial septum and interventricular septum meet. AVN cells have a
relatively slow conduction velocity and they are responsible for the major part of
normal conduction delays between atrial and ventricular contractions. Such delays
are properly timed in order to optimize the atrial pump activity and to protect
the ventricles from too early stimulation. The AVN conducts the action potential
through the nonexcitable atrioventricular septum, activating the specialized fibers
of the bundle of His and the Purkinje network that spread as a tree-like left and right
bundle branches ending on the endocardial surface of the ventricles. These Purkinje
terminations transmit the action potentials to the ventricular walls and cardiac
excitation then propagates throughout the ventricles. The electrical activation of the
left ventricle, the largest cardiac chamber, normally starts at the interventricular
septum, propagates towards the anteroapical region and reaches the posterobasal
region. The electrical activation of the right ventricle starts shortly after the left
ventricle activation; see e.g. [462].

1.2 Cardiac Tissue Organization

Cardiac muscle presents specific features different from the other two main types
of muscle, skeletal and smooth muscles. The two fundamental features of cardiac
muscle are its cyclical activity of contraction/relaxation, where the individual
myocytes contract and relax according to a precise sequence and its homeostatic
regulation in order to maintain a relatively constant circulatory pressure in spite
of large flow variations. While skeletal myocytes are functionally separated and
individually innervated, cardiac myocytes are electrically connected in order to
form a single functional unit, known as cardiac syncytium. While skeletal muscle
strength is modulated mostly by recruiting a variable number of motor units, in the
cardiac syncytium recruitment is not possible and strength modulation relies on the
myocytes strength modulation. Moreover, while in skeletal muscle cells a sequence
of action potentials can generate an increasing force just by a temporal sum of single
pulses, in cardiac myocytes each cycle of contraction/relaxation is activated by a
single action potential that has a duration about 100 times longer.
In cardiac tissue are present three fundamental cell types with different structural
and functional characteristics: working, nodal, and conduction cells.
Working cardiomyocytes. These form the mass of atrial and ventricular tissues
and are responsible for force development. Their morphology is tubular (with
possible branches), with a length of about 50–150 m and a diameter of about
10–20 m. Myocytes volume and shape can be complex and variable, according
to the tissue region, species, developmental stage and disease processes. They are
enclosed by a lipid membrane, the sarcolemma, and contain one or more nuclei,
mitochondria, myofibrils, the sarcoplasmic reticulum, sarcomeres, the cytoskele-
ton anchoring the different organelles and an aqueous solution, the sarcoplasm,
filling the intracellular space. The sarcolemma is a semi-permeable barrier and
contains ion channels, pumps and exchangers that allow the inward and outward
1.2 Cardiac Tissue Organization 5

currents involved in the action potential, as well as other proteins involved in cell
adhesion and signalling. Cardiomyocytes exhibit a periodic structure with cross
striations formed by alternating segments of thick and thin protein filaments. Their
sarcolemma presents deep invagination along the Z-discs known as T-tubules, that
allow depolarization of the membrane to quickly penetrate to the interior of the cell
and play a critical role in the excitation-contraction coupling. Cardiomyocytes are
mechanically coupled at specialized region of the membrane known as intercalated
disks, which also include gap junction channels providing electrical coupling (see
below).
Nodal cells. These cells constitute the SA and AV nodes and are able to
autonomously activate due to the presence of specific ionic channels. Compared
with working myocytes, they are smaller, more tortuous, have limited contractile
activity and lack T-tubules. Nodal cells are richly innervated by parasympathetic
and sympathetic nervous system fibers, that make them susceptible to autonomic
influences. The SA and AV nodes also contain some transitional cells with
intermediate features between nodal and atrial myocytes.
Conduction cells. These are also called Purkinje cells and have a morphology
similar to working myocytes but with a larger diameter (up to 80 m), reduced
contractile proteins and a few T-tubules. They also have the ability of automaticity,
but at a slower rate than SA or AV node cells.
Table 1.1 reports the average values of diastolic potential, upstroke velocity,
overshoot, action potential duration (APD) and propagation velocity for different
mammalian cardiomyocytes. For detailed measurements of conduction velocities in
canine cardiac tissue, see [586, Table 16-1, p. 148].
Fibroblasts. Cardiac tissue is composed by myocytes that are arranged in
a complex three-dimensional fiber and laminar architecture, described below. In
areas between and surrounding cardiac myocytes there are cardiac fibroblasts, that
constitute the major non-myocyte cell population in the ventricles [206]. Although
cardiomyocytes occupy the largest volume fraction of the normal heart, they are
outnumbered by the smaller fibroblasts. These are responsible for the synthesis of
extracellular matrix proteins such as different types of collagen. Fibroblasts can

Table 1.1 Average values of diastolic potential, upstroke velocity, overshoot, action potential
duration (APD) and propagation velocity for different mammalian cardiomyocytes
Diastolic Upstroke Propagation
potential velocity Overshoot APD velocity
Cell types (mV) (mV/ms) (mV) (ms) (mm/ms)
SAN 50, 60 1–10 30 100–200 0.03–0.05
Atria 80 100–200 30 100–200 0.3–0.7
AVN 60, 70 5–15 20 100–300 0.1
His budle/Purkinje 80, 90 800 40 200–500 2–4
Ventricles 80 100–200 40 200–500 0.5–0.8 along
0.15–0.25 across
6 1 Basic Cardiac Anatomy and Electrocardiology

develop into myofibroblasts and both cell types are involved in the development of
fibrosis in injured cardiac tissue. The structural arrangement of fibroblasts is still not
well understood, but some authors have suggested that they are organized in sheets
that follow closely the myocardial sheets, with some fibroblasts forming strands that
bridge cleavage planes between sheets [206]. Cardiac fibroblasts are electrically
non-excitable cells, but they are efficient mechano-electrical transducers. The
possibility that fibroblasts may actively contribute to cardiac electrophysiology has
been considered only recently, see [77, 192, 330, 424].
Extracellular matrix and collagen. The extracellular matrix (ECM) is an
extensive and highly organized network of fibrous proteins, collagen and elastin,
and is the major determinant of passive mechanical properties of cardiac tissue. The
ECM plays important roles in muscle development, myoblast differentiation and
in maintaining functional integrity of the myocardium [576]. Collagen and elastin
have relative densities that can vary with tissue type, species and age. The ECM
structure differs at different cardiac locations. In the endomysium, the collagen
network surrounds single myocytes and interconnects neighboring myocytes. In
the perimysium, bundles of collagen fibers envelope groups of adjacent myocytes,
providing the laminar cardiac structure and long perimysal collagenous tendons can
link adjacent laminae. In the epimysium, layers of collagen and elastin are found at
the epicardium and endocardium.
Gap junctions and connexin. Cardiac myocytes are mechanically connected
by junctions anchored to the cytoskeleton, known as intercalated disks, and are
electrically connected by electrical synapses, known as gap junctions, connect-
ing myocytes mostly end-to-end (longitudinal gap junctions) but also laterally
(transverse gap junctions) [423]. Gap junctions are small ionic channels with low
selectivity, with a diameter of about 2 nm and length of about 2–12 nm. They are
composed of two adjoint connexons, located in the membranes of the two coupled
cells, and each connexon is formed by six channel proteins known as connexins
(Cx40, Cx43, Cx45). Cx43 is the main cardiac connexin and is mostly present in
ventricular tissue. Cx40 is present in atrial tissue and in the conduction system,
while Cx45 is present only in the conduction system. Longitudinal gap junctions
are prevalent in ventricular myocardium, yielding a macroscopic electrical coupling
that is anisotropic.
Cardiac stem cells. The human heart has been traditionally viewed as a
terminally differentiated postmitotic organ, with cardiomyocytes established in
number at birth and persisting throughout the lifespan of the organism. This view has
been recently changed by the discovery that cardiac stem cells live in the heart and
differentiate into the various cardiac cell lineages; see e.g. [41,46,190,289,296] and
the references therein. Cardiac stem cells regulate myocyte turnover and condition
myocardial recovery after injury. The initial enthusiasm triggered by the discovery
of endogenous cardiac stem cells was then followed by some controversy over
the nature of myocyte renewal in the human heart. Intense research studies are
underway to explore this new field of cardiac regeneration. Several categories of
cardiac progenitors have been described and diverse experimental strategies to
remuscularize the injured heart using adult stem cells and pluripotent stem cells,
1.3 Fiber and Laminar Architecture of Ventricular Myocardium 7

cellular reprogramming and tissue engineering are in progress. The novel results that
are rapidly appearing in the literature impose a reconsideration of the mechanisms
involved in myocardial aging and the progression of cardiac hypertrophy to heart
failure. Even if controversial experimental issues are still present, cardiac regener-
ation studies have great potentials and could soon lead to new understanding and
treatments of infarcted hearts and to restoration of myocardial mass and ventricular
function; see e.g. [190].

1.3 Fiber and Laminar Architecture of Ventricular

Myocardium

In addition to the complex extracellular network of fibroblasts and collagen

described above, the ventricular myocardium exhibits a complex three-dimensional
spatial organization where myocytes are connected mostly end-to-end to form
cardiac fibers with varying orientation. This ventricular fiber orientation rotates
smoothly between endocardium and epicardium, see Streeter [498]. Cardiac fibers
are also known to have an additional laminar organization consisting of muscle
sheets, typically 4–6 myocytes thick, running radially from epi- to endocardium,
with surface orientation varying throughout the ventricles. These sheets are sepa-
rated by gaps called cleavage planes and by layers of connective tissue, see e.g.
[138, 234, 292–294, 465, 530].
Diffusion tensor magnetic resonance imaging (DT-MRI) has been used to
measure fiber orientation as the principal eigenvector of the diffusion tensor
measured at each image voxel in formaldehyde-fixed hearts [465]. In DT-MRI,
the pattern of diffusion is represented mathematically by a symmetric second-rank
tensor in three-dimensional space, which can be written as a 3  3 matrix. The
three orthogonal eigenvectors of this tensor (ranked in order of the magnitudes of
their corresponding eigenvalues) have been related to cardiac structure, with the
eigenvalues representing the diffusion along three principal axes. Theory suggests
that the direction of greatest proton diffusion, i.e. the eigenvector with the largest
eigenvalue (primary eigenvector) will be along the fiber long axis, intermediate
diffusion (the secondary eigenvector) will lie in the myolaminar plane, orthogonal
to the fiber long axis. The third and minor direction of diffusion (the tertiary
eigenvector) is by definition orthogonal to the primary and secondary eigenvectors,
so it is normal to the myolaminar plane. The correspondence between the primary
eigenvector and fiber orientation, and the secondary and tertiary eigenvectors and
sheet orientation has been validated by combined DT-MRI and three-dimensional
histological reconstruction of fiber and sheet structure.
Therefore, at any myocardial point x, it is possible to identify a triplet of
orthonormal principal axes al .x/; at .x/; an .x/, with al .x/ parallel to the local fiber
direction, at .x/ and an .x/ tangent and orthogonal to the radial laminae, respectively,
and both being transversal to the fiber axis [34, 137, 292, 293]; see Fig. 1.4 for a
8 1 Basic Cardiac Anatomy and Electrocardiology

Fig. 1.4 Left: schematic representation of ventricular fiber architecture with orthonormal triplets
of vectors along fiber (red), and two orthogonal vectors (blue and black) on intramural, epi-, mid-,
endocardial sections. Right: schematic view of a few laminae (blue) on selected sections (outlined
in green) of the ventricular wall (outlined in red)

schematic representation of ventricular fiber architecture of triplet of orthonormal

principal axes (left) and laminar organization (right).

1.4 Cardiac Action Potentials

Cardiac cells are excitable cells, some autonomously (SAN pacemaker cells) and
some after a proper electrical stimulus. The excitation of a cardiac cell causes a
rapid variation of its potential difference across the cell membrane, the so-called
transmembrane potential. If the stimulus is below a certain threshold value, the
transmembrane potential quickly returns to its resting value after the stimulus ends.
On the other hand, if the stimulus is above threshold, the cell membrane depolarizes
and the transmembrane potential increases from the negative resting value to a value
around or above zero, remains around this value for a certain interval and then
returns to the resting value after the membrane repolarizes. This event is called an
action potential.

1.4.1 Action Potential Phases

The main phases of a typical ventricular action potential are displayed in Fig. 1.5.
Phase 0. In this phase, myocytes undergo a rapid depolarization due to the
opening of the fast NaC channels. This causes a rapid increase in the membrane
1.4 Cardiac Action Potentials 9

K+, Cl− (out) 1 Ca

2+ +
(in), K (out)

Ito1,2 ICa−L, IKs

2

0 + K+ (out)
Na (in)
IKs, IKr, IK1
I 3
Na
0

4
4
−84 mV

0 300 msec

Fig. 1.5 Schematic plot of a cardiac action potential with its phases 0 (depolarization or upstroke),
1 (peak or notch), 2 (plateau), 3 (repolarization or recovery), 4 (resting) and main ions and currents
involved in each phase

conductance and thus a rapid influx of NaC ions into the cell, the INa current,
through sodium channels formed by alpha-subunits with channel proteins of the
NAV 1:# family encoded by SCN genes (mostly NaV 1:5 channel protein encoded by
the SCN5A gene associated with Long QT and Brugada syndrome channelopathies)
and by beta-subunits with NaV ˇ# channel proteins encoded by SCN#B genes.
Phase 1 occurs with the inactivation of the fast NaC channels. The transient net
outward current causing the small downward deflection of the action potential is due
to K C and Cl ions carried by the Ito1 and Ito2 currents, respectively.
Phase 2 is the plateau phase sustained by a balance between inward movement
of Ca2C , the ICa current through L-type calcium channels (CaV 1:2 channel protein
encoded by CACNA1C gene) and outward movement of potassium ions K C , the IKs
current through the slow delayed rectifier potassium channels (KvLQT1 channel
protein encoded by the KCNQ1 gene). The sodium-calcium exchanger current
INa;Ca and the sodium/potassium pump current INa;K also play minor roles during
phase 2.
Phase 3. This is the “rapid repolarization” phase of the action potential, when
the L-type Ca2C channels close while the slow delayed rectifier (IKs ) K C channels
are still open. This ensures a net outward current, corresponding to negative change
in membrane potential, thus allowing more types of K C channels to open. These
are primarily the rapid delayed rectifier K C channels (IKr current, KV 11:1 channel
protein encoded by hERG or KCNH2 genes) and the inwardly rectifying K C current
IK1 . This net outward, positive current causes the cell to repolarize. The delayed
10 1 Basic Cardiac Anatomy and Electrocardiology

rectifier K C channels close when the membrane potential is restored to about 80 to
85 mV, while IK1 is maintained throughout phase 4, contributing to set the resting
membrane potential.
Phase 4 is the resting phase, during which the transmembrane potential remains
at the resting value of about 84 mV until it is stimulated by an external electrical
stimulus.

1.4.2 Action Potential Heterogeneity

Cardiac action potentials differ significantly in different regions of the heart. Indeed,
action potentials from atrial, ventricular and specialized conduction myocytes
exhibit different waveforms, see Fig. 1.6. Moreover, isolated myocytes with differ-
ent action potential morphologies have been observed in the epicardial, midmyocar-
dial (M-cells), and endocardial regions (transmural heterogeneity) and also along
the apex-to-base direction of the ventricle (apex-to-base heterogeneity). When these
myocytes are embedded in the ventricular wall, their heterogeneous properties affect
the sequence of repolarization and the actual APD distribution in the entire wall
(spatial dispersion of APD).
Right and left ventricle heterogeneity. Right ventricular myocytes from dog
hearts show action potentials with shorter APD and deeper notches than in left
ventricular myocytes. Some authors have related these features to larger Ito and
IKs currents in the right ventricle and have suggested that arrhythmogenic gradients
could arise at the interventricular septum.

Fig. 1.6 Heterogeneity of cardiac action potentials from different regions of the heart
1.5 The Electrocardiogram (ECG) 11

Apex-to-base dispersion. Rabbit ventricular myocytes from the basal region

show action potential with shorter APD than myocytes from the apex region.
Moreover, IKr and IKs currents are smaller in apical than in basal myocytes,
with IKr the largest in apical cells and IKs the largest in basal cells. The shorter
basal APD causes ventricular repolarization to proceed from the basal toward the
apical region. Transmural and apex-to-base APD dispersion have been shown to be
arrhythmogenic in many heart preparations.
Transmural dispersion. Studies in isolated tissues and myocytes in several ani-
mal species have found differences in action potentials shapes and APD across the
ventricular wall, showing a transmural dispersion with shorter APD in subepicardial
than in subendocardial cells and with the longest APD in midmyocardial cells, the
so-called M-cells; see e.g. [7–10, 553, 572]. Studies have found transmural hetero-
geneity in the expression of ionic channels, in particular in the INa ; IKr ; IKs ; Ito1
currents and in the NaC =Ca2C exchanger. There has been some controversy whether
transmural repolarization gradient are present in vivo and in particular in human
heart, since cell coupling and electrotonic currents tend to mask APD heterogeneity.
Since the T wave in the ECG (described in the next section) marks ventricular
repolarization and the APD determines the QT interval, transmural dispersion of
repolarization contributes to amplitude, polarity and duration of the T wave and
to the QT interval duration. It can also contribute to ST segment elevation, QT
dispersion and pathological U waves.
Temporal dispersion of APD. In addition to spatial dispersion, APD can also
show temporal dispersion due to beat-to-beat changes alternating between short and
long action potentials. For example, T wave alternans (beat-to-beat changes of the
T wave amplitude) are associated with reentrant ventricular tachycardia. Moreover,
spatial and temporal dispersion together seem to be associated with a broad range
of cardiac arrhythmias. For example, spatial dispersion of APD and refractoriness
can lead to temporal dispersion of repolarization after a premature stimulus, and
the resulting spatio-temporal dispersion can generate a functional conduction block
initiating reentry.

1.5 The Electrocardiogram (ECG)

The ECG is the registration of the extracellular potential on the body surface due to
the propagation of cardiac action potentials. The registrating electrodes are usually
placed at standard positions described below.
12 1 Basic Cardiac Anatomy and Electrocardiology

1.5.1 ECG Leads

Einthoven leads. Given the potentials at the left arm ˚L , at the right arm ˚R and at
the left foot ˚F , Einthoven standard limb leads are defined as

VI D ˚L  ˚R
VII D ˚F  ˚R
VIII D ˚F  ˚L :

By Kirchhoff’s law, we have VI C VIII D VII . Einthoven assumed that the heart
is located at the center of a homogeneous spherical conductor representing the
torso, and that the three vectors associated with these three leads form an equilateral
triangle in the vertical frontal plane of the body, see Fig. 1.7.
Augmented leads. In addition to the three Einthoven leads, three additional
leads, denoted by aV R ; aV L ; aV F and known as augmented leads, are defined on
the frontal plane in the directions that bisect the sectors associated with each pair of
Einthoven leads. For example, the augmented aV F lead is defined by

˚L C ˚R
VaV F D ˚F  :
2

Fig. 1.7 Schematic view of Einthoven standard limb leads (left) and augmented leads (right)
1.5 The Electrocardiogram (ECG) 13

Fig. 1.8 The six precordial leads V1 ; : : : ; V6 in a frontal view (left) and horizontal cross-section
(right) of the thorax

In this way, we obtain six leads that divide the frontal plane into equal 30ı sectors,
see Fig. 1.7. In order to measure potentials close to the heart, six additional leads are
defined on a horizontal plane by using the Wilson central terminal

˚R C ˚L C ˚F
˚CT D :
3
Precordial leads. The six precordial (or chest) leads, denoted by V1 ; : : : ; V6 ,
are obtained by comparing the Wilson central terminal with unipolar electrode
readings taken from six different locations on the chest, see Fig. 1.8 and [444]
for more details. V1 and V2 are located at the fourth intercostal space on the
right and left side of the sternum, and the other four proceed around the left
chest just below the fourth rib, ending with V6 under the armpit. The 12 leads
I; II; III; aV R ; aV L ; aV F ; V1 ; : : : ; V6 form the most commonly used clinical ECG
system.

1.5.2 ECG Deflections and Intervals

A schematic view of a normal electrocardiogram is shown in Fig. 1.9, while an

example of a 12-lead recording is shown in Fig. 1.10. The various deflections in
the ECG signal are commonly denoted in alphabetic order as P, Q, R, S, T, U. The
temporal sequence of these waves remains constant, but their amplitude and sign
can vary depending on the recording electrode position on the thorax.
P wave. The P wave is associated with atrial depolarization, that continues during
the subsequent PQ segment. The completion of atrial repolarization generates an
atrial Ta wave not visible in normal ECGs because hidden by the QRS complex.
The PQ segment is also associated with the delay between the beginning of atrial
and ventricular activation (atrioventricular conduction time).
14 1 Basic Cardiac Anatomy and Electrocardiology

Fig. 1.9 Schematic view of a normal ECG waveform

Fig. 1.10 Sample of actual ECG waveforms

QRS complex. The subsequent QRS complex, usually the largest deflection
in the ECG, is associated with the complex activation sequence of the ventricles,
resulting in upward and downward deflections and multiple peaks in most of the
leads. The Q wave corresponds to the activation of the interventricular septum, the R
wave corresponds to the septum activation completion and the apex activation, the S
wave corresponds to the activation of the ventricular free walls and the basal region.
These three waves are not always detectable and can sometimes merge, yielding a
quite variable morphology of the QRS complex.
ST segment. The QRS complex in normal ventricular myocytes is followed by a
rather long period where all cells are depolarized during the plateau of the action
potential. This isoelectric interval, known as ST segment, lasts a few hundreds
1.5 The Electrocardiogram (ECG) 15

milliseconds, starting at the end of the QRS complex (the so-called J point) and
ending at the beginning of the T wave.
T wave. The T wave signals ventricular repolarization and its morphology
depends on the repolarization sequence through the ventricles. In normal hearts, the
start of the T wave is thought to correlate with the plateau ending of the epicardial
cells, the T wave peak with the full repolarization of the epicardium, the end of the
T wave with the repolarization of the midmyocardial M-cells.
QT interval. The QT interval is measured from the beginning of the QRS
complex to the end of the T wave. Since the QT interval is strongly frequency
dependent, a corrected QT interval
p QTc is sometimes defined, for example by
normalizing it as QTc D QT= RR.
U wave. After the T wave, it is sometimes present a smaller diastolic deflection
known as U wave. Some authors explain the U wave as due to repolarization of
the papillary muscles or Purkinje fibers, while others explain it as due to late
repolarization of M-cells or yet to afterpotentials caused by ventricular stretch.
Other important cardiac events too weak to be detected on the ECG include atrial
repolarization, SA nodal activation, AV nodal conduction and Purkinje network
propagation. An example of a normal 12-lead ECG recording is shown in Fig. 1.10.

1.5.3 ECG Diagnosis

We briefly review here some of the main ECG alterations. For a more complete
treatment see [189] and [317].
Normal sinus rhythm. The normal sinus rhythm of a healthy normal heart has
an ECG with all three P, QRS, T complexes normal, distinguishable and with a
frequency between 60 and 100 bpm (beats per minute).
Sinus bradycardia and tachycardia. A sinus rhythm of less than 60 bpm
is called sinus bradycardia and may be a consequence of increased vagal or
parasympathetic tone. A sinus rhythm of higher than 100 bpm is called sinus
tachycardia; most often it is due to physiological response to physical exercise or
psychical stress, but may also result from congestive heart failure.
Sinus arrhythmia. If the sinus rhythm is irregular such that the longest PP or RR
interval exceeds the shortest interval by 0.16 s, then it is called sinus arrhythmia, a
very common situation in all age groups and not considered a heart disease in young
people.
Nonsinus atrial rhythm and wandering pacemaker. If the origin of atrial
contraction is located somewhere else in the atria other than the sinus node (nonsinus
atrial rhythm) or wanders in the atria (wandering pacemaker), then the P waves can
have variable polarity and the PQ interval variable duration.
Paroxysmal atrial tachycardia (PAT). Paroxysmal atrial tachycardia describes
the condition when the P waves are a result of a reentrant activation front (circus
movement) in the atria, usually involving the AV node. This leads to a high rate of
activation, usually between 160 and 220 bpm. In the ECG the P wave is regularly
16 1 Basic Cardiac Anatomy and Electrocardiology

followed by the QRS complex. The isoelectric baseline may be seen between the T
wave and the next P wave.
Atrial flutter. When the heart rate is sufficiently elevated so that the isoelectric
interval between the T wave ending and the P wave beginning disappears, the
arrhythmia is called atrial flutter. The origin is also believed to involve a reentrant
atrial pathway. The frequency of these fluctuations is between 220 and 300 bpm.
The AV node and, thereafter, the ventricles are generally activated by every second
or every third atrial impulse (2:1 or 3:1 heart block).
Atrial fibrillation. The activation in the atria may also be fully irregular and
chaotic, producing irregular fluctuations in the baseline called atrial fibrillation.
A consequence is that the ventricular rate is rapid and irregular, though the QRS
contour is usually normal.
Ventricular tachycardia. Slower conduction in ischemic ventricular muscle that
leads to circular activation (reentry) causes activation of the ventricular muscle at a
high rate (over 120 bpm) and generates rapid, bizarre, and wide QRS complexes.
This arrythmia is called ventricular tachycardia and is often a consequence of
ischemia and myocardial infarction.
Ventricular fibrillation. When ventricular depolarization occurs chaotically, the
situation is called ventricular fibrillation. The ECG has coarse irregular deflections
without QRS-complexes. The cause of fibrillation is the establishment of multiple
re-entry loops usually involving diseased heart muscle. The contraction of the
ventricular muscle is irregular and is ineffective at pumping blood, with a result
of almost immediate loss of consciousness and death within minutes.
Atrioventricular blocks. In normal sinus rhythm, the P waves always precede
the QRS-complex with a PR-interval of 0.12–0.2 s. When the P wave always
precedes the QRS complex but the PR interval is prolonged over 0.2 s, first-degree
atrioventricular block is diagnosed. If the PQ interval is longer than normal and
the QRS complex sometimes does not follow the P wave, the atrioventricular block
is of second-degree. If the PR interval progressively lengthens, leading finally to
the dropout of a QRS complex, the second degree block is called a Wenkebach
phenomenon. Complete lack of synchronism between the P wave and the QRS
complex is diagnosed as third-degree (or total) atrioventricular block.
Bundle-branch blocks. In right bundle-branch block (RBBB), the right bundle-
branch is defective so that the electrical impulse cannot travel through it to the
right ventricle, but activation reaches the right ventricle by proceeding from the
left ventricle. This can be seen in the ECG as a QRS complex wider than 0.1 s, or
a broad terminal S wave in lead I, or a double R wave in lead V1. The situation in
left bundle-branch block (LBBB) is similar, but activation proceeds in a direction
opposite to RBBB. The ECG has a broad and tall R wave, usually in leads I, aVL,
V5 or V6.
Wolff-Parkinson-White (WPW) syndrome. In this syndrome, the QRS com-
plex initially exhibits an early upstroke called the delta wave. The interval from the
P wave to the R spike is normal, but the early ventricular excitation forming the
delta wave shortens the PQ time. The cause of the WPW syndrome is the passage of
activation from the atrium directly to the ventricular muscle via an abnormal route,
1.6 Cardiac Imaging 17

called the bundle of Kent, which bypasses the AV junctions. This activates part of
the ventricular muscle before normal activation reaches it via the conduction system
(after a delay in the AV junction), a process called pre-excitation.
Atrial hypertrophy. Right atrial hypertrophy is a consequence of right atrial
overload, due to tricuspid valve disease (stenosis or insufficiency), pulmonary valve
disease, or pulmonary hypertension. An unusually large (i.e. 0.25 mV) P wave is
seen in leads aVF and III. Left atrial hypertrophy is a consequence of left atrial
overload, due to mitral valve disease (stenosis or insufficiency), aortic valve disease,
or hypertension in the systemic circulation. The P wave exhibits two phases with the
same polarities in lead I but opposite polarities in V1. This typical P wave form is
called the mitral P wave.
Ventricular hypertrophy. Right ventricular hypertrophy is a consequence of
right ventricular overload, due to pulmonary valve stenosis, tricuspid insufficiency,
pulmonary hypertension or many congenital cardiac abnormalities, such as a
ventricular septal defect. The ECG shows a tall R wave of 0.7 mV in lead V1. Left
ventricular hypertrophy is a consequence of left ventricular overload, due to mitral
valve disease, aortic valve disease, or systemic hypertension. The ECG has a tall R
wave in leads I, III, V5, V6 and has a tall S wave in leads III, V1 and V2.
Myocardial ischemia and infarction. The decreased transport of oxygen to the
cardiac muscle due to occlusion of a coronary artery is called ischemia. Ischemia
causes changes in the resting potential and in the repolarization of the muscle cells,
which are seen as changes in the T wave. If the oxygen transport is terminated in a
certain area, the heart muscle dies in that region. This is called an infarction.

1.6 Cardiac Imaging

As part of the rapidly evolving fields of medical and biological imaging, also the
field of cardiac imaging is currently experiencing tremendous progresses that are
producing better imaging techniques seeking to reveal, diagnose, or examine both
cardiac diseases and the normal cardiac anatomy and physiology. Among the many
techniques available, we mention the following ones and refer the interested readers
to some of the recent monographs [83,291,475,554] and reviews [2,83] in the field.
Angiocardiography is a technique for radiographic examination of the heart
chambers and thoracic veins and arteries. A liquid radiocontrast agent, typically
containing iodine, is injected into the bloodstream, then the tissues are examined
using X-rays. To avoid dilution, the radiopaque material is typically introduced with
a catheter, a process known as selective angiocardiography. The X-ray image is
normally captured on high speed serial media that allow the motion to be observed,
such as a 35 mm film. The process requires fasting before the test, with a sedative
and an antihistamine being administered before the test. Angiocardiography can be
used to detect and diagnose congenital defects in the heart and adjacent vessels. The
use of angiocardiography has declined with the introduction of echocardiography.
18 1 Basic Cardiac Anatomy and Electrocardiology

However, angiocardiography is still in use for selected cases as it provides a higher

level of anatomical detail than echocardiography.
Coronary catheterization uses pressure monitoring and blood sampling through
a catheter inserted into the heart through blood vessels in the leg to determine the
functioning of the heart, and, following injections of radiocontrast dye, uses X-ray
fluoroscopy, typically at 30 frames per second, to visualize the position and size
of blood within the heart chambers and arteries. Coronary angiography is used to
determine the patency and configuration of the coronary artery lumens.
Echocardiogram. Transthoracic echocardiogram uses ultrasonic waves for con-
tinuous heart chamber and blood movement visualization. In recent times, it has
become one of the most commonly used tools in diagnosis of heart problems, as it
allows non-invasive visualization of the heart and the blood flow through the heart,
using a technique known as Doppler.
Transoesophageal echocardiogram uses a specialized probe containing an ultra-
sound transducer at its tip, that is passed into the patient’s esophagus. It is used in
diagnosis of various thoracic defects or damage, i.e. heart and lung imaging. It has
some advantages and disadvantages over thoracic or intravascular ultrasound.
Percutaneous echocardiogram, also known as intravascular ultrasound, is an
imaging methodology using specially designed, long, thin, complex manufactured
catheters attached to computerized ultrasound equipment to visualize the lumen and
the interior wall of blood vessels.
3D echocardiography (also known as 4D echocardiography when the pic-
ture is moving) is now possible, using a matrix array ultrasound probe and an
appropriate processing system. This enables detailed anatomical assessment of
cardiac pathology, particularly valvular defects, and cardiomyopathies. The ability
to slice the virtual heart in infinite planes in an anatomically appropriate manner
and to reconstruct three-dimensional images of anatomic structures makes 3D
echocardiography unique for the understanding of the congenitally malformed heart.
Real Time 3-Dimensional echocardiography can be used to guide the location of
bioptomes during right ventricular endomyocardial biopsies, placement of catheter
delivered valvular devices, and in many other intraoperative assessments. The 3D
Echo Box developed by the European Association of Echocardiography offers a
complete review of Three Dimensional Echocardiography; see http://www.escardio.
org/communities/EACVI/education/echo-box/3d-echobox-2/Pages/welcome.aspx.
Computed Tomography Angiography (CTA), an imaging methodology using
a ring-shaped machine with an X-Ray source spinning around the circular path so
as to bathe the inner circle with a uniform and known X-Ray density. Cardiology
uses are growing with the incredible developments in CT technology. Currently,
multidetector CT, specially the 64 detector-CT are allowing to make cardiac studies
in just a few seconds (less than 10 s, depending on the equipment and protocol used).
These images are reconstructed using algorithms and software. Great development
and growth will be seen in the short term, allowing radiologists to diagnose cardiac
artery disease without anesthesia and in a non-invasive way.
1.6 Cardiac Imaging 19

Magnetic Resonance Imaging (MRI) (originally called nuclear magnetic res-

onance imaging), an imaging methodology based on aligning the spin axis of
nuclei within molecules of the object being visualized using both powerful super-
conducting magnets and radio frequency signals and detectors. Cardiology uses
are growing, especially since MRI differentiates soft tissues better than CT and
allows for comprehensive exams including the quantitative assessment of size,
morphology, function, and tissue characteristics in one single session. Current
implementations for cardiology uses are sometimes limited by lengthy protocols,
claustrophobia and contraindications based on some complex metallic implants
(pacemakers, defibrillators, insulin pumps), while artificial valves and coronary
stents are generally not problematic. Image quality can be reduced by the continuous
movement of heart structures. There is a promising future in cardiac MRI by more
efficient scans, increasing availability of scanners and more widespread knowledge
about its clinical application.
Optical mapping. Traditionally, surface electrodes have been, and continue to
be used to measure extracellular cardiac electric potentials. Arrays of electrodes are
placed on the heart’s surface and electrical activities are recorded simultaneously
from a limited number (a few hundred) of sites. These surface contact mapping
techniques may suffer from low spatial and temporal resolution, low depth field,
and far-field effects.
Rapid technical innovations during the last decades of the twentieth century
have led to the development of sophisticated optical mapping techniques, see
e.g. [18, 227]. Optical mapping is performed by using voltage sensitive dyes
and imaging systems with high temporal and spatial resolutions (i.e., acquiring
thousands of pixels in a few milliseconds) and can be used in a variety of settings
from the subcellular level in vitro to the whole heart in vivo. Both activation and
repolarization times can be measured directly from different sites in both normal
and diseased heart (e.g., ischemic heart).
Numerical studies of optical mapping techniques have been carried out using the
Bidomain equations coupled with a photon diffusion equation, see e.g. [50,404,433]
and the references therein.
Chapter 2
Mathematical Models of Cellular Bioelectrical
Activity

In this chapter, we briefly review the main mathematical models used in cellular
electrophysiology, and we refer to [212, 273] for a more complete treatment and
additional topics.

2.1 Excitable Cellular Membranes

A fundamental topic in molecular biophysics is the study of ionic currents through

protein channels of excitable cellular membranes as a function of the ionic
concentrations, applied voltage and channel structure.
The cellular membrane is a lipid bilayer in which are immersed proteins. The
membrane contains water-filled pores and protein-lined pores, called ionic channels,
which allow the flow of specific ions, primarily NaC ; K C ; Cl ; and Ca2C . The
membrane acts as a barrier to the free flow of ions and maintains concentration
differences of these ions. The concentration gradients produce a potential difference
across the membrane, the transmembrane potential, that drives the ionic currents.
Concentration differences are set up and maintained by active transport mecha-
nisms that use energy to pump ions against their concentration gradients.
One of the most important of these pumps is the NaC  K C ATPase sodium-
potassium pump: it uses energy stored in ATP (adenosine triphosphate) molecules
to pump NaC out of the cell and K C in

ATP C 3ŒNaC i C 2ŒK C e ! ADP C Pi C 3ŒNaC e C 2ŒK C i ;

releasing ADP (adenosine diphosphate) and inorganic phosphate Pi . For example,

in the squid giant axon: ŒNaC i D 50 mM < ŒNaC e D 437 mM, ŒK C i D
397 mM > ŒK C e D 20 mM. At the equilibrium, this biochemical pump maintains
a potential jump between the intra- and extracellular space of vrest D 65 mV.

© Springer International Publishing Switzerland 2014 21

P. Colli Franzone et al., Mathematical Cardiac Electrophysiology, MS&A 13,
DOI 10.1007/978-3-319-04801-7__2
22 2 Mathematical Models of Cellular Bioelectrical Activity

2.2 The Nernst-Planck Equation

In general, the flux across the membrane of the generic ion K with valence z is
the sum of two contributions, the diffusion flux Jdiff and the electric flux Jelect , all
measured in mol cm2 s1 . The constitutive law describing this total flux is known
as the Nernst-Planck equation.
+
[K ]i [K+]
e
intracellular extracellular
media J media
K,diff

J
K,elect

membrane

Due to the ion motion by diffusion, down the concentration gradient, the
diffusion flux Jdiff satisfies Fick law:

Jdiff D Drc;

where D is the diffusion coefficient (cm2 s1 ) and c is the concentration (mol cm3).
Due to the ion motion by electric field, the electric flux Jelect satisfies the Planck
equation:
z
Jelect D  cru;
jzj

where u is the electric potential (V ),  isthe mobility of the ion (cm2 V1 s1 ), z is
C1 for positive ions
jzj D
z
the valence of the ion, so that
1 for negative ions:
We note that if the concentration gradient is zero then Jdiff D 0, while if the
potential gradient is zero then Jelect D 0.
In both the electric and the diffusion fluxes, the ions collide with the solvent
and therefore the diffusion coefficient D and the ion mobility  should be related.
Indeed, Einstein [159] found that this relationship is given by

RT
DD ; (2.1)
jzjF

where R is the gas constant .8:314 J=.K  mol//, T the absolute temperature .K/,
F D e NA the Faraday’s constant .9:6485e C 4 C=mol/, e the elementary charge
and NA the Avogadro’s number, see Tables A.1 and A.2. Then it follows that the
total ionic flux J is given by
z
J D Jdiff C Jelect D Drc  cru;
jzj
2.3 The Goldman-Hodgkin-Katz (GHK) Current-Voltage Relation 23

hence, using (2.1), we obtain the Nernst-Planck equation

 
zF
J D D rc C cru : (2.2)
RT

Let us assume now that the ion flux J and the electric potential u are transverse to
the cell membrane, which extends from x D 0 (inside) to x D L (outside), and the
diffusion coefficient D is constant. Moreover, if we are at the steady state @t c D 0,
thus the conservation law

@t c C @x J D 0

implies that J is constant. Hence, we can reduce Eq. (2.2) to the one-dimensional
problem
dc zF du J
.x/ C .x/c.x/ C D 0: (2.3)
dx RT dx D
Equation (2.3) is clearly a linear differential equation in c.x/, whose solution can
be obtained in terms of u.x/ and J by multiplying by
 Z 
zF x du
exp .s/ds
RT 0 ds
and integrating on .0; x/, yielding
  Z x   
zF J zF
c.x/ D exp  .u.x/  u.0// c.0/  exp .u.s/  u.0// ds :
RT D 0 RT
(2.4)

2.3 The Goldman-Hodgkin-Katz (GHK) Current-Voltage

Relation

If we suppose the electric field constant across the membrane, then

du v
D ;
dx L
where v D u.0/  u.L/ D ui  ue and we have adopted the standard notation
with indexes i; e denoting intra- and extracellular quantities. In this case, setting
c.0/ D c i , the solution (2.4) of the Nernst-Planck equation is
    
JRTL zvF zvF
c.x/ D 1  exp x C c i exp x :
DzvF RTL RTL
24 2 Mathematical Models of Cellular Bioelectrical Activity

In order to satisfy c.L/ D c e , it must be that

 
DzFv c i  c e exp zvF
J D  RT
 :
LRT 1  exp zvF RT

The electrical current density in C s1 cm2 D A cm2 is defined by I WD zFJ,

hence we obtain the Goldman-Hodgkin-Katz (GHK) current-voltage relation
 
z2 F 2 c i  c e exp zvF
I DP v  RT
 ; (2.5)
RT 1  exp zvFRT

where P D D L is the permeability of the membrane to the ion K.

The GHK flux equation is applied in several fields, one of the most important
being in modeling the L-type channels in calcium dynamics, see e.g. [47].

2.4 Nernst Equilibrium Potential

We consider a membrane separating two neutral solutions with different ionic

concentrations, e.g. the salt KCl composed by the ions potassium K C and chloride
Cl . Moreover, we suppose that the membrane is permeable only to one of the two
ions, that we denote by K, with generic valence z.
At thermodynamical equilibrium, each local process and its reverse proceed at
the same rate, hence J D 0 and the solution (2.4) of the Nernst-Planck equation
becomes
 
zF
c.x/ D exp  .u.x/  u.0// c.0/;
RT

from which it follows that in x D L

 
c.L/ zF
log D .u.L/  u.0// :
c.0/ RT

Using the standard notation with indexes i; e denoting intra- and extracellular
quantities, respectively, we then obtain the Nernst equation for the equilibrium
(reversal) potential
 i
RT c
vK WD ui  ue D  log e :
zF c
2.5 Thermodynamical Derivation of the Nernst Potential 25

Table 2.1 Intra- and Extracellular Intracellular Nernst

extracellular concentration
concentration (mM) concentration (mM) potential (mV)
and Nernst potential values
for ventricular myocytes NaC 145 15 60
Cl 100 5 80
KC 4.5 160 95
Ca2C 1.8 1e4 130
HC 1e4 2e4 18

This derivation is based on the constitutive Nernst-Planck law for the total flux.
For a thermodynamical derivation from electrochemical potentials that shows the
universal character of the Nernst potential, see the next Section.
For ventricular myocytes, the typical intra- and extracellular concentrations and
Nernst potential values are given in Table 2.1.

2.5 Thermodynamical Derivation of the Nernst Potential

We recall that the first law of Thermodynamics states that for a closed system the
change dU of internal energy equals the difference between the amount of heat
dQ supplied to the system and the amount of external work dW performed by the
system,

dU D dQ  dW:

Since the work performed by the system can be written in terms of the increase dV
in its volume V and the external pressure p as dW D pdV, the first law can also be
written as

dU D dQ  pdV:

The second law of Thermodynamics states that the change dS of the entropy S of
the system is

dS D dSrev C dSirr ;

where dSrev is the change due to the reversible interaction of the system with its
environment and dSirr is the irreversible internal change within the system that is
always  0. Since dSrev D dQ=T with T the absolute temperature, this law can also
be written as

dSirr D dS  dQ=T:
26 2 Mathematical Models of Cellular Bioelectrical Activity

The Gibbs free energy is defined as

G D U C pV  TS;

i.e. as the difference between the enthalpy H D U C pV and the product TS.
Differentiating and using the first and second laws of Thermodynamics, we obtain

dG D Vdp  SdT  TdSirr :

For reversible processes, dSirr D 0, hence the second law implies dQ D TdS and
the first law

dU D TdS  pdV;

which is a fundamental relationship expressing the change in internal energy dU

of a closed system in terms of reversible changes in its entropy S and volume V .
On the other hand, for open systems the internal energy can also increase due to an
increase dni of its i -th component
 @U 
dU D dni ;
@ni S;V;nj

where S; V and the other components nj , with j ¤ i , are kept constant. Hence,
the first law of Thermodynamics becomes
X
dU D TdS  pdV C i dni ;
i

 @U 
where i D is the chemical potential of the i -th component, and the
@ni S;V;nj
Gibbs free energy becomes
X
dG D Vdp  SdT C i dni : (2.6)
i

This last equation shows that the chemical potential for the i -th component can also
 @G 
be defined as i D .
@ni T;p;nj
For charged ions with valence zi , the electrochemical potential is defined as

Q i D i C zi Fu; (2.7)

where F is the Faraday constant and u the electric potential.

2.5 Thermodynamical Derivation of the Nernst Potential 27

The chemical potential g for an ideal gas can be derived using the equation
of state pV D nRT, relating the pressure p, volume V , number of moles n, gas
constant R, temperature T . If we consider n D 1 mol of ideal gas expanding at
constant temperature, then Eq. (2.6) gives us
dp
dG D Vdp D RT ;
p
and by integration
p
G D G0 C RT log ;
p0
where p0 is a reference pressure and G0 the associated Gibbs free energy. Since for
a mole of pure substance the chemical potential is given by the Gibbs free energy,
the last equation can be rewritten as
p
g D 0g C RT log ; (2.8)
p0

with 0g the chemical potential of the gas at the reference pressure p0 . If we take the
reference pressure equal to 1 atmosphere, Eq. (2.8) becomes

g D 0g C RT log p: (2.9)

The chemical potential of a solute can be derived by considering a solute

dissolved in a solvent to be similar to gas molecules present in free space. Since
the ideal gas equation can be written as p D Vn RT D cRT, with c the molar
concentration of the solute, we see that, at constant temperature, the pressure is
proportional to the concentration c. Therefore, by the analogy with a gas, the
chemical potential of a solute s dissolved in a solvent is given by Eq. (2.9) with
the pressure replaced by the concentration cs of the solute:

s D 0s C RT log cs : (2.10)

For an ion species in solution, Eq. (2.10) generalizes to the case of an electrochemi-
cal potential Q s as in (2.7)

Q s D s C zs Fu D 0s C RT log cs C zs Fu: (2.11)

The Nernst equilibrium potential can be derived from this equation by considering
an ion species s in equilibrium across a membrane and denoting by csi ; cse the
intracellular and extracellular concentrations, respectively. Since at equilibrium we
must have Q is D Q es , Eq. (2.11) yields

0s C RT log csi C zs Fusi D 0s C RT log cse C zs Fuse ;

28 2 Mathematical Models of Cellular Bioelectrical Activity

so that

csi
zs F .usi  use / D RT log ;
cse

RT ci
i.e. the Nernst equilibrium potential vs D usi  use D  log se :
zs F cs

2.6 Electrodiffusion Models: The Poisson-Nernst-Planck

(PNP) Equation

We now suppose to have two types of ions, a cation S1 and an anion S2 , with
concentrations c1 and c2 , valences z1 D z > 0 and z2 D z < 0. We consider
the flux through a channel ˝ connecting the intra- and extracellular neutral spaces.
The Nernst-Planck equations for the fluxes J1 ; J2 of S1 and S2 state that
 
zF
J1 D D1 rc1 C c1 ru ;
 RT 
zF
J2 D D2 rc2  c2 ru ;
RT

and the conservation law (continuity equation)

@t c1 C div J1 D 0 in ˝;
@t c2 C div J2 D 0 in ˝:

Furthermore, by Gauss law

div D D  in ˝;

where D D a E is the dielectric displacement associated to the electric field E, a

the dielectric constant in the ionic solution,  the charge density. Since E D ru
with u the electric potential, we have D D a ru and since in one mole there are
qN A z D Fz charges, we have  D Na q.z1 c1 C z2 c2 / D Fz.c1  c2 /. Hence, inside
the channel ˝

Fz
u D .c1  c2 /:
a
2.6 Electrodiffusion Models: The Poisson-Nernst-Planck (PNP) Equation 29

We denote by S the lateral surface of the channel ˝ and by @˝i ; @˝e the two
channel openings in contact with the intra and extracellular media, respectively.
Then, we make the following additional assumptions, yielding boundary conditions
for the ionic concentrations, fluxes and the electric potential:
(a) The cation and anion concentrations on both sides of the membrane are
constant, so that the intra- and extracellular spaces are in electrical equilibrium

c1 .x; t/j@˝i D c2 .x; t/j@˝i D c i ; c1 .x; t/j@˝e D c2 .x; t/j@˝e D c e I

(b) Since the dielectric constants a of the channel’s solution and m of the
membrane satisfy m  a , we assume that the channel wall is electrically
insulated,

nT EjS D 0; i.e. nT rujS D 0I

(c) The channel wall prevents the movement of ions toward the lipidic regions,

nT Ji jS D 0; nT Je jS D 0I

(d) The electric potentials at the channel openings in contact with the extracellular
and intracellular media are constant and equal to ui and ue , respectively, but,
since the potential is given up to an additive constant, we set

uj@˝i D ui  ue D v; uj@˝e D 0:

The Poisson-Nernst-Planck (PNP) equation treats discrete ions in a channel

as a continuum charge distribution, representing the average ion fluxes in terms
of densities and potential gradients. For a derivation of the PNP system from
the Langevin model of ionic motion, see Schuss et al. [464]. For a singular
perturbation analysis of the steady-state PNP system, see e.g. [26, 212, 438, 484]
and the references therein. In addition to continuum PNP equations, more detailed
approaches have been studied, such as all-atom Molecular Dynamics (MD), non-
linear Poisson-Boltzmann (PB) equation, Brownian Dynamics (BD), see Roux et al.
[437]. For a multiscale model linking MD and electrophysiology for the cardiac IKs
channel, see Silva et al. [482].
Since the channel has a diameter of a few Angstroms, it can be assumed cylindri-
cal and the membrane around the channel planar. Then the ionic concentrations
c1 ; c2 , fluxes J1 ; J2 and potential u are constant on each cross section of the
cylindrical channel ˝ and depend only on the position x across the cylindrical
channel ˝, that we assume to range from 0 (intracellular media) to L (extracellular
media).
30 2 Mathematical Models of Cellular Bioelectrical Activity

We now consider the steady-state of the 1-D model, assuming that the ion
concentrations are constant across the channel section. Collecting the previous
equations and boundary conditions, we find the one-dimensional Poisson-Nernst-
Planck (PNP) system:
8  
ˆ zF
ˆ
ˆ @t c1 C @x J1 D 0 with J1 D D1 @x c1 C c1 @x u
ˆ
ˆ  RT 
ˆ
ˆ
ˆ
ˆ zF
ˆ
ˆ @t c2 C @x J2 D 0 with J2 D D2 @x c2  c2 @x u
ˆ
ˆ RT
ˆ
ˆ
ˆ
< @2 u D  zF .c1  c2 /
xx
a
ˆ
ˆ with boundary conditions:
ˆ
ˆ
ˆ
ˆ D t/ D c i c1 .L; t/ D c2 .L; t/ D c e
ˆ
ˆ
c1 .0; t/ c2 .0;
ˆ
ˆ u.0; t/ D v u.L; t/ D 0
ˆ
ˆ
ˆ
ˆ
ˆ and initial conditions:
:̂
u.x; 0/ D 0 c1 .x; 0/ D c2 .x; 0/ D 0:

At steady state @t c1 D @t c2 D 0, hence the stationary PNP system becomes

8  
ˆ d dc1 zF du
ˆ
ˆ D1 C c1 D 0;
ˆ
ˆ dx  dx RT dx 
ˆ
ˆ
ˆ
ˆ d dc2 zF du
ˆ
ˆ D2  c2 D 0;
ˆ
< dx dx RT dx
2
d u zF (2.12)
ˆ
ˆ D  .c1  c2 /
ˆ
ˆ dx
2 a
ˆ
ˆ D .0/ D c i
ˆ
ˆ c1 .0/ c2
ˆ
ˆ
ˆ c1 .L/ D c2 .L/ D c e
:̂
u.0/ D v u.L/ D 0:

This one-dimensional non-linear boundary value system admits a unique solution

c1 .x/; c2 .x/; u.x/, see [370] and for a perturbation analysis [26]. Multidimensional
existence results follow from the theory of semiconductor equations, see e.g. [320],
and exploiting the gradient flow structure of the PNP system, see e.g. [6]. The system
can be written in dimensionless form by rescaling the variables and functions as

c1 c2
y D x=L; D uzF=RT; pD ; nD i ;
Cc
ci e c C ce
L L c i;e vzF
Jp D J1 ; Jn D J2 i ; O
c i;e
D ; vO D ;
.c i C c e /D1 .c C c e /D2 ci C ce RT
2.6 Electrodiffusion Models: The Poisson-Nernst-Planck (PNP) Equation 31

so that

d2 zFL2 d 2 u z2 F 2 L2 i
D D  .c C c e /.p  n/;
dy2 RT dx2 RTa
dp d
Jp D Cp ;
dy dy
dn d
Jn D n :
dy dy

dJ p dJ n
The first two equations of (2.12) require that D D 0, i.e. that Jp and Jn
dy dy
are constant. Therefore, the dimensionless one-dimensional stationary PNP system
becomes:
find fp.x/; n.x/; .x/g and constants fJp ; Jn g such that
8
ˆ dp d
ˆ
ˆ Jp D Cp ;
ˆ
ˆ dy dy
ˆ
ˆ
ˆ
ˆ dn d
ˆ
< Jn D n
dy dy
d 2 (2.13)
ˆ
ˆ D  2 .p  n/;
ˆ
ˆ
ˆ
ˆ dy2
ˆ
ˆ
ˆ with boundary conditions:
:̂
p.0/ D n.0/ D cOi ; p.1/ D n.1/ D cOe ; .0/ D vO ; .1/ D 0;

z2 F 2 L2 i
with 2
D .c C c e /.
a RT
Since no analytical solutions of this system are known, we will look for
approximate solutions obtained by perturbation techniques with respect to the
parameter .

2.6.1 PNP: The Short Channel or Low Concentrations Limit

zF p
We look first for approximate solutions when the parameter D p L ci C ce
a RT
is small, corresponding to short channels or low ionic concentrations. This is a
regular perturbation problem and in order to obtain an approximation with error of
order O. / we solve the reduced problem in the limit case of D 0. Thus

d2
D0
dy2
32 2 Mathematical Models of Cellular Bioelectrical Activity

d
implies that the electric field E D  is constant and the potential .y/ is linear,
dy
so from the boundary conditions we obtain

d
.y/ D vO  vO y; D Ov:
dy

Then the differential equation for p becomes

dp
Jp D  vO p:
dy

Integrating between 0 and y and using the left boundary condition p.0/ D cOi , we
obtain

1  e vOy
p.y/ D cOi e vOy C Jp :
vO

The right boundary condition p.1/ D cOe then determines

cOi  cOe e Ov

Jp D vO :
1  e Ov
Analogously, the differential equation for n becomes

dn
Jn D C vO n;
dy

so, by integrating and imposing the boundary conditions, we obtain

1  e Ovy
n.y/ D cOi e Ovy  Jn ;
vO
cOi  cOe e vO
Jn D Ov :
1  e vO
Returning to the original variables, we then determine the dimensional fluxes J1 ; J2 ,
hence the ionic density currents
zF
z2 F 2 D1 c i  c e e  RT v
I1 D zFJ 1 D v zF
;
RT L 1  e  RT v zF
z2 F 2 D2 c c e
i e RT v
I2 D zFJ 2 D .v/ zF :
RT L 1  e RT v

These are GHK current-voltage relations (2.5), with PK D DK =L; K D 1; 2.

2.6 Electrodiffusion Models: The Poisson-Nernst-Planck (PNP) Equation 33

2.6.2 PNP: The Long Channel or High Concentrations Limit

We now look for approximate solutions

p when the parameter is large. We define
a RT 1
the small parameter D 1= D p and consider the singular
zF L c i C c e
perturbation problem for the potential
2
2d
D n  p:
dy2

The reduced problem obtained in the limit case for D 0 implies that p.y/ D n.y/,
so by adding or subtracting the equations for Jp and Jn in (2.13) we have
8
ˆ
ˆ dp dn dp
ˆ
ˆ Jp  Jn D C D2 ;
< dy dy dy
d d d (2.14)
ˆ
ˆ Jp C Jn D p Cn D 2p
ˆ dy dy dy
:̂
p.0/ D n.0/ D cOi ; p.1/ D n.1/ D cOe ; .0/ D vO ; .1/ D 0:

dp
The first equation requires to be constant and from the boundary conditions we
dy
obtain p.y/ D cO C .cO  cO /y, therefore system (2.14) becomes
i e i

8
ˆ
ˆ Jp C Jn
< D .cOe  cOi /
2 (2.15)
ˆd JO JO
:̂ D D i ;
dy p.y/ cO C .cOe  cOi /y

with JO D .Jn  Jp /=2, yielding

JO cOi C .cOe  cOi /y

.y/ D vO C log :
cOe  cOi cOi

JO cOe
Then the boundary condition .1/ D 0 gives us the condition vO C log D
cOe  cOi cOi
Ov
0, that determines JO D .cOe  cOi / , which coupled with (2.15) implies
log ccOOi
e

8
ˆ
ˆ Ov
ˆ Jp D .cOe  cOi /.
< e C 1/;
log ccOOi
ˆ Jn D .cOe  cOi /. Ov e  1/:
ˆ
:̂
log ccOOi
34 2 Mathematical Models of Cellular Bioelectrical Activity

ci C ce ci C ce
Returning to the dimensional fluxes J1 D D1 Jp ; J2 D D2 Jn , we
L L
c Cc
i e
c C ce
i
find that the dimensional currents I1 D zF D1 Jp ; I2 D zF D2 Jn
are L L
!
zF e v zFD1 c e  c i
I1 D .c  c i /D1
RT ce
1 D .v  v1 /;
L zF log c i
L v1
! (2.16)
zF e v zFD2 c e  c i
I2 D  .c  c /D2 RT
i
c e  1 D .v  v2 /;
L zF log c i
L v2

RT ce
where vk D log i ; k D 1; 2 are the Nernst potentials of the two ions
zk F c
considered.
These linear current-voltage relations are derived using the approximation of a
constant diffusive gradient.

2.6.3 Equilibrium Potential for Multi-ion Fluxes

The Nernst equilibrium potential has a derivation based on universal thermodynamic

principles only in the simple case of a single-ion flux. If we consider two or more
ions, the situation is more complex and in general there is no potential that zeros
each individual ionic current. However, we can still compute a reverse potential that
zeros the total ionic current.
For example, if we consider the two ions NaC ; K C and we assume that their
channels obey GHK current-voltage relations
F
e  RT v
F2 c i  cNa e
INaC D vPNa Na F
;
 RT
RT 1  e Fv
e  RT v
F2 c i  cK e
IK C D vPK K F
;
 RT
RT 1e v

with PNa D DNa

L ; PK D DK
L , then the reverse potential at which INaC C IK C D 0 is

RT i
PNa cNa C PK cK
i
vr D  log e :
F e
PNa cNa C PK cK

If instead the channels obey linear current-voltage relations

INaC D GNa .v  vNa /; IK C D GK .v  vK /;

2.7 Electrical Circuit Model of the Cellular Membrane 35

then the reverse potential that zeros the total current is

GNa vNa C GK vK
vr D :
GNa C GK

Analogously, if we consider the three ions NaC ; K C ; Cl with GHK channels, we
have

RT i
PNa cNa C PK cK
i
C PCl cCl
i
vr D  log e ;
F e
PNa cNa C PK cK
e
C PCl cCl

while with linear channels we have

GNa vNa C GK vK C GCl vCl
vr D :
GNa C GK C GCl

2.7 Electrical Circuit Model of the Cellular Membrane

Since the cell membrane separates charges that accumulate at its intra- and
extracellular surfaces, it can be viewed as a capacitor. The capacitance is defined
as the ratio between the charge Q across the capacitor and the voltage potential
drop v necessary to hold the charge

Q
Cm D :
v

Since the capacitive current is Icap D dQ=dt, if Cm is constant, we have

dQ dv
Icap D D Cm :
dt dt
The cellular membrane is modeled as a capacitor in parallel with a resistor (ionic
current), as shown in Fig. 2.1, so by the current conservation law, the transmembrane
current given by the sum of the capacitative and ionic currents, must be equal to the
applied current Iapp

dv
Cm C Iion D Iapp : (2.17)
dt
The structure of the total ionic current will be described by the specific ionic
membrane model adopted. In order to discuss these models, we need to introduce
the formalism for modeling ion channel gating. For more properties about ionic
channels of excitable membranes see [228].
36 2 Mathematical Models of Cellular Bioelectrical Activity

Fig. 2.1 Electrical circuit model of the cellular membrane

2.8 Ion Channel Gating

In general, for a given transmembrane potential drop, the ionic current through a
population of ionic channels is given in a unit area of membrane surface and can be
modeled as a product

Iion D g.v; t/ .v/;

where g.v; t/ is the proportion of open channels in a unit area of the membrane
surface and .v/ is the current-voltage (I-V) relation of a single open channel. In
the previous sections on electrodiffusion models, we have seen different models for
.v/. For example, in the long channel limit (2.16), we have .v/ D gc .v  vr /,
where gc and vr are the channel conductance and resting potential, respectively.
The proportion of open channels in a unit area of the membrane surface can be
N Ntot N
written as g.v; t/ D D , where N is the number of open channels
S S Ntot
on the membrane surface, Ntot is the total number of membrane channels, S is
the membrane surface area. Hence, in the long channel limit we have Iion D
Ntot gc
G c w.v  vr /, where G c D is the maximal channel conductance per unit
S
2.8 Ion Channel Gating 37

N
area of the membrane surface and w D is the percentage of open channels. We
Ntot
now show how to model this percentage w, also known as gating variable, starting
with the simplest case of a single-unit two-state channel.
One unit protein with two states. Consider a membrane portion of unit area
containing a given type of ionic channels that behave independently and that assume
two states S0 (close) or S1 (open), with transition rate constants ˛; ˇ (in general
dependent on the transmembrane potential v) as indicated in the first order kinetic
diagram

˛
close state D S0  S1 D open state :
ˇ

If S0 .t/; S1 .t/ are the average numbers of channels that at time t are in the state
S0 ; S1 , respectively, then
8
< dS1 .t/
D ˛ S0 .t/  ˇ S1 .t/
: S dt
1 .t/ C S0 .t/ D S:

By setting s0 D S0 .t/=S; s1 D S1 .t/=S , the percentages of close and open channels

per unit area of the surface membrane, we then have
8
< ds1
D ˛ s0  ˇ s1
dt
: s C s D 1:
0 1

Eliminating s0 and denoting by w D s1 the gating variable, we have

dw
D ˛ .1  w/  ˇ w :
dt

By setting w1 D ˛
˛Cˇ
and w D 1
˛Cˇ
, this equation can be rewritten as

dw w1  w
D ; (2.18)
dt w

where w1 and w are the equilibrium state and the time constant, respectively.
By voltage-clamp or patch-clamp techniques, it is possible to apply an external
current that balances the membrane current in such a way that the transmembrane
potential v is kept at a fixed value in time. In this case, the capacitative current
38 2 Mathematical Models of Cellular Bioelectrical Activity

Icap D Cm dv
dt
is zero and in the circuit model equation (2.17) in the long-channel
limit we have Iion .t; v/ D G w.t; v/ .v  vr / D Iapp ; hence the membrane
conductance can be obtained as

Iapp .t; v/
G w.t; v/ D :
v  vr

For each fixed value of v, these experimental curves can be plotted in time and
compared with the theoretical curves predicted by solving the gating equation (2.18)
in the voltage-clamp setting: if at time t D 0 v is increased from 0 to v0 and held
fixed at such value, the time behavior is given by

w.t/ D w1 .v0 /.1  e t = w .v0 /

/; (2.19)

while if at time t D T we return v back to 0, we obtain the evolution

w.t/ D w1 .v0 /e .t T /= w .v0 /

/; (2.20)

see Fig. 2.2. If the experimental conductance curves do not agree with these
exponential curves, then the single-unit two-state channel model is not adequate and
more complex models must be considered. Such models are introduced in the next
sections and indeed they will be needed to correctly describe the ionic currents in the
membrane models developed in the literature, starting with the classical Hodgkin-
Huxley model presented in Sect. 2.9.1.

0.9 v = +15
0

0.8
v = −25
0
0.7

0.6 v = −45
0
W

0.5

0.4

0.3 v0 = −65
0.2

0.1

0
0 20 40 60 80 100
TIME

Fig. 2.2 Plots of voltage-clamp solutions (2.19) on the interval Œ0 T  and (2.20) on ŒT 100, with
T D 50 ms, for v0 D 65; 45; 25; 15 mV
2.8 Ion Channel Gating 39

Two subunits with two states. Analogously, if we consider ionic channels with
two equal and independent subunits, which can open and close according to the
following kinetic diagram

2˛ ˛
close state S0  S1  S2 D open state ;
ˇ 2ˇ

we have
8 ds
ˆ
ˆ
0
D ˇs1  2˛s0
ˆ
ˆ dt
ˆ
ˆ
ˆ
ˆ
<
ds1
D ˇs1 C 2˛s0  ˛s1 C 2ˇs2
ˆ
ˆ dt
ˆ
ˆ
ˆ
ˆ
ˆ ds2
:̂ D ˛s1  2ˇs2 :
dt

From the condition d

dt .s0 C s1 C s2 / D 0 it follows s0 C s1 C s2 D 1; hence
8
ˆ ds0
ˆ
ˆ D ˇs1  2˛s0
ˆ
ˆ dt
ˆ
ˆ
<
ds2
ˆ
ˆ D ˛s1  2ˇs2
ˆ
ˆ dt
ˆ
ˆ
:̂
s1 D 1  s0  s2 ;

and after a change of variable

s2 D n2 ; s1 D 2n.1  n/; s0 D .1  n/2

dn
D ˛.1  n/  ˇn:
dt
k subunits with two states. Consider now an ionic channel with k equal and
independent subunits, which can open and close according to the following kinetic
diagram
kα (k−1) α α
S0 S1 S .... Sk−1 Sk
β 2
2β kβ
s s1 s2 sk−1 sk
0
40 2 Mathematical Models of Cellular Bioelectrical Activity

As in the case with two subunits, after a change of variable, we obtain

 
k
s2 D nk ; sj D n j .1  n/kj
j
dn
D ˛.1  n/  ˇn:
dt

Two different subunits. We consider now two subunits of different type, that we
will denote by m and h. The transition rate constants are ˛; ˇ for m and ; ı for h,
as indicated in the following kinetic diagram
2α α
S S10 S
00 20
β 2β
γ δ γ δ γ δ
α α
S S S
01 11 21
β 2β

After a change of variable in the associated system of differential equations, we

obtain

s00 D .1  m/3 .1  h/; s10 D 3m.1  m/2 .1  h/;

s20 D 3m2 .1  m/.1  h/; s30 D m3 .1  h/
s01 D .1  m/3 h; s11 D 3m.1  m/2 h; s21 D 3m2 .1  m/h; s31 D m3 h

dm dh
D ˛.1  m/  ˇm; D .1  h/  ıh: (2.21)
dt dt

2.9 Cardiac Action Potential Models

The first mathematical model that describes accurately the action potential wave-
form and the associated permeability changes was proposed by Hodgkin and Huxley
in [229]. Although the Hodgkin-Huxley (HH) model was developed specifically for
nerve action potentials, the ideas and the mathematical formalism have been used
later in a number of models describing cardiac action potentials, see the survey
papers [356, 443] and also [141, 215, 282].
2.9 Cardiac Action Potential Models 41

2.9.1 Hodgkin-Huxley Model

The celebrated Hodgkin-Huxley (HH) model consists of the following system,

coupling the equation of the equivalent circuit model (2.17) with the equations of
channel gating as in (2.18)–(2.21) for three gating variables m; h; n

8
ˆ dv
ˆ
ˆ Cm C Iion .v; m; h; n/ D Iapp
ˆ
ˆ dt
ˆ
ˆ dm m1 .v/  m
ˆ
ˆ D
<
dt m .v/
ˆ dh h1 .v/  h
ˆ
ˆ D
ˆ
ˆ dt h .v/
ˆ
ˆ
ˆ dn n1 .v/  n
:̂ D
dt n .v/

where the ionic current is the sum of sodium, potassium and leakage currents

Iion D INa C IK C IL :

Each of these currents has a linear current-voltage structure as in (2.16), with three
equal activation and one inactivation independent subunits for the sodium channel

INa D GN Na m3 h .v  vNa /;

four equal activation independent subunits for the potassium channel

IK D GN K n4 .v  vK /;

and no units for the leakage current

IL D GN L .v  vL /:

Here GN Na ; GN K ; GN L are the maximal conductances and vNa ; vK ; vL are the Nernst
potentials of each channel type, respectively.
We recall that the coefficients of the gating variables equations are given by

˛m 1
m1 D ; m D ;
˛m C ˇm ˛m C ˇm
˛h 1
h1 D ; h D ;
˛h C ˇh ˛h C ˇh
˛n 1
n1 D ; n D ;
˛n C ˇn ˛n C ˇn
42 2 Mathematical Models of Cellular Bioelectrical Activity

with

0:1.25  v/  v 
˛m D ; ˇm D 4 exp ;
 vv/  1
expŒ0:1.25 18
1
˛h D 0:07 exp ; ˇh D ;
20 expŒ0:1.30  v/ C 1
0:01.10  v/  v 
˛n D ; ˇn D 0:125 exp :
expŒ0:1.10  v/  1 80

The remaining parameters are

GN Na D 120; GN K D 36; GN L D 0:3; vNa D 115; vK D 12; vL D 10:6:

2.9.2 General Structure of Cardiac Cellular Membrane Models

The ionic current through channels of the membrane is modulated by the transmem-
brane potential v D ui  ue , by gating variables w WD .w1 ; : : : ; wM / and by ionic
intracellular concentration variables c WD .c1 ; : : : ; cS /. In the membrane models
considered in the remainder of this chapter, the ionic current has the following
general structure

X
N Y
M
pj
Iion .v; w; c/ D Gk .v; c/ wj k .v  vk .c// C In .v; w; c/;
kD1 j D1

where N is the number of ionic currents, Gk is the membrane conductance and vk is

the reversal potential for the k-th current, pjk are integers and In accounts for time
independent ionic fluxes.
The dynamics of the gating variables w is described in the Hodgkin-Huxley
formalism by a system of ODEs having the following structure
8
ˆ
ˆ dwj
< D Rj .v; w/ D ˛j .v/.1  wj /  ˇj .v/wj
dt
(2.22)
ˆ wj .0/ D wj;0
:̂ ˛ ; ˇ > 0; 0  w  1; j D 1; : : : ; M:
j j j

The dynamics of the ionic concentration variables c is described by the additional

system of ODEs
8
ˆ dcj Icj .v; w/  Acap
ˆ
< D Sj .v; w; c/ D 
dt Vcj  zcj  F
(2.23)
ˆ cj .0/ D cj;0
:̂
j D 1; : : : ; S;
2.9 Cardiac Action Potential Models 43

where Icj is the sum of ionic currents carrying ion cj , Acap is the capacitive
membrane area, Vcj is the volume of the compartment where cj is updated, zcj
is the valence of ion cj and F is the Faraday constant.
The transmembrane current Im is the sum of the capacitive current, associated
with the membrane lipidic bilayer and of the ionic current Iion . Since Im must
balance the applied current Iapp (see e.g. [250]), then the evolution of the trans-
membrane potential of a single myocyte is given by the following system of ODEs

8
ˆ dv
ˆ
< Cm dt C Iion .v; w; c/ D Im D Iapp ;
ˆ
dw dc (2.24)
ˆ dt  R.v; w/ D 0;
ˆ
dt
 S.v; w; c/ D 0;
:̂
v.0/ D v0 ; w.0/ D w0 ; c.0/ D c0 ;

where Cm , Iion , and Iapp are the surface capacitance, the ionic current of the mem-
brane and the applied current per unit area of the membrane surface, respectively.

2.9.3 Ionic Models of Purkinje Fibers, Sinoatrial Node (SAN),

Atria

In this work, we will focus on cardiac ionic models of ventricular cells. Among the
many other cardiac ionic models, we mention the Punkinje fiber models by Noble
[355], McAllister-Noble-Tsien [323], DiFrancesco-Noble [149], the Sinoatrial node
(SAN) models by Zhang et al. [585], Severi et al. [472], and the references therein,
the atrial models by Nygren et al. [360], Kneller et al. [280] and the references
therein. For many additional models, we refer e.g. to Pullan et al. [406].

2.9.4 Ventricular Models

The first ventricular membrane model was formulated by Beeler and Reuter in
1977 [33]. This model, as the McAllister-Noble-Tsien model [323] developed in
1975 for Purkinje fibers, relied on the Hodgkin-Huxley formalism and, similarly
to the Hodgkin-Huxley model, assumed that sodium and potassium intracellu-
lar concentrations (ŒNaC i , ŒK C i ) remain constant during the action potential.
However, in cardiac myocytes entry of Ca2C through the L-type calcium channel
ICa.L/ produces a significant change in the calcium intracellular concentration
ŒCa2C i , mainly by triggering Ca2C release from the sarcoplasmic reticulum (SR)
via the calcium-induced calcium-release (CICR) process. A first attempt to take into
account the ŒCa2C i dynamics necessary to reproduce action potential morphology
was introduced in the earliest ventricular cell models, such as the Beeler-Reuter
44 2 Mathematical Models of Cellular Bioelectrical Activity

[33] and the Luo-Rudy phase I [308], but only at a phenomenological level. More
biophysically detailed ŒCa2C i models were introduced in later models, see e.g.
[254, 306, 405, 473].
Changes in ŒNaC i and ŒK C i can also influence the action potential morphology
when cells are paced at a fast rate. The first model incorporating detailed information
regarding ŒNaC i and ŒK C i dynamic concentration changes was the DiFrancesco-
Noble model [149] of the Purkinje fibers. Rasmusson et al. [418] developed a
similar model for a bullfrog atrial cell. The Luo-Rudy dynamic (LRd) model
[309, 310] of the guinea pig ventricular action potential formulated these processes
for the ventricular myocytes. These models were founding members in a new
class of second generation models that account for dynamic ion concentration
changes. Several such models for ventricular myocytes in different species have
been published since that time (see Table 2.2) and in particular Rudy’s group has
recently published a detailed ionic model for human ventricular myocytes, see [361].

Table 2.2 Some of the most used ventricular ionic models for different species and dimensions
of the associated dynamical systems of differential equations
Reference Species Equations
Beeler and Reuter [33] Mammalian 8
Ebihara and Johnson [154] Chick embrio
Luo and Rudy [308] Guinea pig 8
Noble et al. [357] Guinea pig 17
Luo and Rudy [309, 310] Guinea pig 19
Fenton and Karma [170] Mammalian 3
Jafri, Rice and Winslow [254] Guinea pig 31
Noble et al. [358] Guinea pig 22
Priebe and Beuckelmann [402] Human 22
Winslow et al. [563] Canine 33
Clancy and Rudy [93, 94] Guinea pig
Pandit et al. [367] Rat 26
Puglisi et al. [405] Rabbit
Bondarenko et al. [53] Mouse 41
ten Tusscher et al. [521] Human 17
Hund and Rudy [243] Canine 29
Shannon et al. [473] Rabbit 46
Cortassa et al. [136] Mammalian 50
Livshitz and Rudy [306] Canine 18
Niederer and Smith [351] Rat
Bueno-Orovio, Cherry and Fenton [65] Human 4
Mahajan et al. [315] Rabbit 26
Li et al. [299] Mouse 36
O’Hara and Rudy [361] Human 41
2.9 Cardiac Action Potential Models 45

More recently, detailed ionic models have been coupled with detailed signaling
pathways models; see for example [224] for a simulation study of beta-adrenergic
signaling and its whole-cell effects in ventricular myocytes by incorporating
receptor isoforms, multiple pathways and phosphorylation.

2.9.4.1 The Beeler-Reuter Model

In 1977, Beeler and Reuter developed the first mathematical model of a mammalian
ventricular muscle cell (see [33]) based on the Hodgkin-Huxley formalism. The
ionic current Iion is given by the sum of N D 4 currents

Iion D INa C Is C IK1 C Ix1 :

The fast inward sodium current INa is primarily responsible for the rapid upstroke
of the action potential, while the other currents determine the configuration of the
plateau and repolarization phases. The slow inward current Is , primarily carried by
calcium ions, influences the duration of the action potential. The time-dependent and
time-independent outward potassium currents Ix1 and IK1 are instead responsible
for the repolarization phase. It was also discovered that the intracellular calcium
concentration changes significantly during the course of an action potential, hence
this was included in the model as an additional dynamic variable.
Fast inward sodium current INa . In addition to the activation gate m and the
inactivation gate h of the Hodgkin-Huxley model, Beeler and Reuter added a slow
inactivation gate j . All the three gating variables follow the dynamics governed by
Eq. (2.22). The expression of the sodium current is given by

INa D .gN Na m3 hj C gN NaC /.v  ENa /;

where ENa is the sodium reversal potential (50 mV), gN Na is the maximal sodium con-
ductance (0:04 mS=mm2) and gN NaC is the constant background sodium conductance
(3  105 mS=mm2).
Slow inward current Is . This current is controlled by an activation gate d and
an inactivation gate f , both following the dynamics of (2.22). The magnitude of the
current is given by

Is D gN s df .v  Es /;

where gN s is the maximal channel conductance 9  104 mS=mm2 . The reversal

potential Es depends on the intracellular calcium concentration ŒCa2C i , precisely

Es D 82:3  13:0287 log.0:001ŒCa2C i /:

46 2 Mathematical Models of Cellular Bioelectrical Activity

The time dependence of the intracellular calcium concentration depends in turn on

the current Is , according to

d ŒCa2C i
D 0:01Is C 0:07.104  ŒCa2C i /:
dt
Time-dependent outward potassium current Ix1 . As in the Hodgkin-Huxley
model, this current is controlled by a single gating variable x1 , following (2.22).
The expression of the current is
 
exp.0:04.v C 77//  1
Ix1 D 8  103 x1 :
exp.0:04.v C 35/
Time-independent outward potassium current IK1 . The experimental evi-
dence suggested the presence of a time-independent background potassium current
with magnitude given by
 
4.exp.0:04.v C 85//  1/
IK1 D 0:0035 C
 exp.0:08.v C 53// C exp.0:04.v
 C 53//
0:2.v C 23/
0:035 :
1  exp.0:04.v C 23//

2.9.4.2 The Luo Rudy I Model (LR1)

In 1991, Luo and Rudy developed a new mammalian ventricular membrane model
[308], based on the Beeler-Reuter model, adjusting some parameters to reproduce
more recent experimental results and including additional potassium currents. In the
LR1 model, the ionic current Iion is given by the sum of N D 6 currents

Iion D INa C Isi C IK C IK1 C IKp C Ib ;

two inwards (INa , Isi ) and four outwards (IK , IK1 , IKp , Ib ). The first three currents
depend on six gating variables and one ion (intracellular calcium) concentration,
while the last three are time-independent. The reversal potentials are set to the
Nernst potentials for the ions involved, allowing the reversal potentials to vary
according to ionic concentrations gradients. Figure 2.3 reports the time evolution
of the transmembrane potential v, the gating variables m, h, j , d , f , X and the
intracellular calcium concentration ŒCa2C i .
Fast inward sodium current INa . The expression of this current is the same as
in the Beeler-Reuter model with the omission of the constant background sodium
conductance, thus

INa D gN Na m3 hj.v  ENa /;

where gNa D 0:23 mS=mm2.

2.9 Cardiac Action Potential Models 47

0 1

m
V

0.5
−90 0
0 100 200 300 400 0 100 200 300 400
0 1
INa

0.5

h
−200
−400 0
0 100 200 300 400 0 100 200 300 400

0 1
Isi

−2 0.5

j
−4 0
0 100 200 300 400 0 100 200 300 400

1 1
IK

0 0.5

d
−1 0
0 100 200 300 400 0 100 200 300 400

4 1
2
IK1

0.5
f

0 0
0 100 200 300 400 0 100 200 300 400
2 1
IKp

0.5
0
−1 0
0 100 200 300 400 0 100 200 300 400
−3
x 10
5
4
Cai
Ib

2
0
0
0 100 200 300 400 0 100 200 300 400
time (msec) time (msec)

Fig. 2.3 LR1 model: Transmembrane potential v, ionic currents INa ; Isi ; IK ; IK1 ; IKp ; Ib , gating
variables m, h, j , d , f , X and intracellular calcium concentration ŒCa2C i as functions of time

Slow inward current Isi . The general form of this current is the same as the one
of the Is current in the Beeler-Reuter model

Isi D gN si df .v  Esi /;

where the reversal potential Esi is given by

Esi D 7:7  13:0287 log.ŒCa2C i /:

The dynamics of the gating variable d , f and the intracellular calcium concentration
ŒCa2C i is the same as in the Beeler-Reuter model.
Time-dependent potassium current IK . Unlike in the Beeler-Reuter model,
this current is controlled not only by an activation gate X , but also by an inactivation
gate Xi . The magnitude of the current is given by

IK D gN K XX i .v  EK /;
48 2 Mathematical Models of Cellular Bioelectrical Activity

where the maximal potassium conductance depends on the extracellular potassium

concentration according to
r
3 ŒK C o
gN K D 2:82  10 mS=mm2 :
5:4

The reversal potential EK is dependent on the concentration of both K C and NaC

ions and is given by
 C 
RT ŒK o C PRNaK ŒNaC o
EK D log ;
F ŒK C i C PRNaK ŒNaC i

where PRNaK is a dimensionless permeability ratio for NaC that is expressed

relative to K C . The activation gate X follows the classical Hodgkin-Huxley
dynamics (2.22), while the new inactivation gate Xi is given by
8
< 2:837 exp.0:04.v C 77//  1
v > 100 mV
Xi D .v C 77/ exp.0:04.v C 35//
:
1 v  100 mV:

Time-independent potassium current IK1 . This current was reformulated using

one gating variable with a time constant small enough that it may be approximated
by a steady-state formulation. The magnitude of the current is

IK1 D gN K1 K11 .v  EK1 /;

where the maximal channel conductance is

r
ŒK C o
gN K D 6:047  103 mS=mm2
5:4
and the reversal potential is the Nernst potential for potassium ions. The steady-state
gate is given by
˛K1
K11 D ;
˛K1 C ˇK1

where
1:02
˛K1 D ;
1 C exp.0:2385.v  EK1  59:215//

0:49124 exp.0:08032.v  EK1 C5:476//C exp.0:06175.vEK1  594:31//

ˇK1 D :
1C exp.0:5143.vEK1 C4:753//
2.9 Cardiac Action Potential Models 49

Plateau potassium current IKp . This current, given by

IKp D gN Kp Kp .v  EKp /;

plays a role during the plateau phase of the action potential, restoring the cell to
its resting state. The maximal current conductance is gN Kp D 1:83  104 mS=mm2 ,
the reversal potential is the same of the time-independent potassium current, thus
EKp D EK1 and

1
Kp D :
1 C exp..7:488  v/=5:98/

Background current Ib . This current is a linear function of the transmembrane

potential with magnitude given by

Ib D gN b .v  Eb /;

where gN b D 0:03921 mS=cm2 and Eb D 59:87 mV.

2.9.4.3 The Luo Rudy Dynamic Model (LRd)

In 1994, Luo and Rudy developed a new mammalian ventricular membrane model
[309, 310], mainly based on data from the guinea pig. The new model includes a
detailed phenomenological description of calcium-induced calcium-release, intra-
cellular calcium cycling and calcium buffering. The ionic current Iion is given by
the sum of N D 11 currents

Iion D INa C ICa.L/ C IK C IK1 C IKp C INaCa C INaK C InsCa C IpCa C IbCa C IbNa :

Figure 2.4 reports the time evolution of the transmembrane potential v, the gating
variables m, h, j , d , f , Xs1 , Xs2 , Xr and the intracellular calcium concentration
ŒCa2C i .
Fast inward sodium current INa . The formulation of this current is the same as
the LR1 model, but the maximum conductance was decreased to gN Na D 16 mS=cm2.
L-type calcium current ICa.L/ . This current, equivalent to the slow inward
current Isi from the LR1 model, is the sum of the currents through the L-type
calcium channel that is permeable to Ca2C , NaC and K C . The name is due to its
activity, which is long lasting in comparison to other calcium currents, such as the
T-type transient calcium current. The magnitude of the current is given by

ICa.L/ D ICa.L/Ca C ICa.L/Na C ICa.L/K ;

50 2 Mathematical Models of Cellular Bioelectrical Activity

100 1.5
1

m
0 0.5
v

0
−100 −0.5
1.5 1.5
1 1
h

0.5 0.5

j
0 0
−0.5 −0.5
1.5 1.5
1 1
d

0.5 0.5

f
0 0
−0.5 −0.5

0.4 0.2
s1

Xs2 0.1
0.2
X

0
0
−0.1 −4
x 10
1.5 4
[Ca2+]i

1
r

0.5 2
X

0
−0.5 0
0 100 200 300 400 0 100 200 300 400

Time (ms) Time (ms)

Fig. 2.4 LRd model: Transmembrane potential v, gating variables m, h, j , d , f , Xs1 , Xs2 , Xr and
the intracellular calcium concentration ŒCa2C i as functions of time

where

ICa.L/Ca D d  f  fCa  INCa.L/Ca

ICa.L/Na D d  f  fCa  INCa.L/Na
ICa.L/K D d  f  fCa  INCa.L/K :

The INCa.L/Ca , INCa.L/Na and INCa.L/K terms represent the maximum currents through
the channel for each ion species. Because there is only one physical channel, each
of these equations is governed by the same gating variables. The activation gate d
and the inactivation gate f follow the Hodgkin-Huxley formalism (2.22), while the
further inactivation gate fCa is governed by the following equation

1
fCa D  2 ;
ŒCa2C i
1C KmCa
2.9 Cardiac Action Potential Models 51

where KmCa is the half activation concentration for Ca2C set to 0:6  103 mM. The
three fully activated current terms follow the short channel limit, thus

vF 2
Si ŒS i exp.zS vF=RT/  So ŒS o
INS D PS z2S ;
RT exp.zS vF=RT/  1

where S D Ca2C ; NaC ; K C . PS is the permeability of the membrane to ion S with

valence zS and Si ; So are the intracellular and extracellular activity coefficients of
ion S .
Time-dependent potassium current IK . The kinetics of the gating variables
was updated from the LR1 model and an additional pair of activation gates was
added, thus

IK D gN K X 2 Xi .v  EK /;

where
1
Xi D  :
1 C exp v56:26
32:1

Time-independent potassium current IK1 . The formulation of this current is

the same as the one in the LR1 model, but the maximum channel conductance was
changed to
p
gK1 D 0:75 ŒK C o =5:4 mS=cm2:

Plateau potassium current IKp . The formulation of this current is the same as
the one in the LR1 model.
Sodium-calcium exchanger INaCa .

1 1 1
INaCa D kNaCa  3 C 3
 
Km;Na C ŒNa o Km;Ca C ŒCa o
2C 1 C k exp..  1/vF=RT/
 C 3
sat
C 3

 exp. vF=RT/ŒNa i ŒCa o  exp..  1/vF=RT/ŒNa o ŒCa2C i ;
2C

where kNaCa D 2 000 A=F, Km;Na D 87:5 mM, Km;Ca D 1:38 mM, ksat D 0:1
and D 0:35.
Sodium-potassium pump INaK .

1 ŒK C o
INaK D INNaK  fNaK  C
 C ;
1 C .Km;Nai =ŒNa i / 1:5 ŒK o C Km;Ko
52 2 Mathematical Models of Cellular Bioelectrical Activity

where INNaK D 1:5 A=F, Km;Nai D 10 mM, Km;Ko D 1:5 mM and

1
fNaK D
1 C 0:1245 exp.0:1vF=RT/ C 0:0365 exp.vF=RT/
 
 D 1=7 exp.ŒNaC o =67:3/  1 :

Non-specific calcium-activated current InsCa . Although activated by calcium,

this current arises from the transport of sodium and potassium ions.
1
InsCa D .INnsNa C INnsK /  3 ;
1C KmnsCa
ŒCa2C i

where KmnsCa D 1:2 M and INnsNa ; INnsK are computed according to the short
channel limit formula.
Sarcolemmal calcium pump IpCa .

ŒCa2C i
IpCa D INpCa ;
Km;p.Ca/ C ŒCa2C i

where INpCa D 1:15 A=F and Km;p.Ca/ D 0:5 mM.

Calcium background current IbCa .

IbCa D gbCa .v  ECa;N /;

where gbCa D 0:003016 mS=F and

ECa;N D .RT=2F / log.ŒCa2C o =ŒCa2C i /:

Sodium background current IbNa .

IbNa D gbNa .v  ENa;N /;

where gbNa D 0:00141 mS=F and ENa;N D ENa .

Intracellular potassium dynamics. The evolution of the intracellular K C
concentration is described by the following ODE

d ŒK C i Acap
D .IK;t  2INaK /; (2.25)
dt Vmyo F
where Acap is the capacitative area of the membrane, Vmyo is the volume of the
myoplasm, F is the Faraday constant and IK;t is the total K C current through all
ion channels, i.e.

IK;t D IK C IK1 C IKp :

2.9 Cardiac Action Potential Models 53

Intracellular sodium dynamics. The evolution of the intracellular NaC concen-

tration is described by the following ODE

d ŒNaC i Acap
D .INa;t C 3INaK C 3INaCa /;
dt Vmyo F

where INa;t is the total NaC current through all ion channels, i.e.

INa;t D INa C IbNa :

Intracellular calcium buffering. The formulation of the intracellular calcium

handling is based on the concepts detailed in the DiFrancesco-Noble model [149],
but includes two additional features. The first is a calcium leakage current from the
Sarcoplasmic Reticulum back into the cytosol and the second is the inclusion of
calcium buffers within the intracellular spaces. Therefore, the evolution of the free
calcium ŒCa2C i , junctional sarcoplasmic reticulum (JSR) free calcium ŒCa2C jsr
and network sarcoplasmic reticulum (NSR) calcium ŒCa2C nsr concentrations are
described by the following ODEs:
 
d ŒCa2C i Acap Vnsr Vjsr
D ˇmyo .ICa;t  2INaCa / C .Iup  Ileak /  Irel ;
dt 2Vmyo F Vmyo Vmyo
(2.26)
d ŒCa2C jsr
D ˇjsr .Itr  Irel /; (2.27)
dt

d ŒCa2C nsr Vjsr

D Iup  Ileak  Itr ; (2.28)
dt Vnsr

where ICa;t is the total Ca2C current through all ion channels, i.e.

ICa;t D ICa.L/ C InsCa C IpCa C IbCa ;

and the buffering of CaC2 is modeled by incorporating instantaneous buffering to

troponin C in the myoplasm and to calmodulin in JSR binding sites (see e.g. [547])
by means of
!1
ŒtrpnKm;trpn ŒcmdnKm;cmdn
ˇmyo D 1C C ;
.ŒCa i C Km;trpn /
2C 2 .ŒCa2C i C Km;cmdn /2
 1
ŒcsqnKm;csqn
ˇjsr D 1 C :
.ŒCa2C jsr C Km;csqn /2

All the other symbols are defined in Table 2.3.

54 2 Mathematical Models of Cellular Bioelectrical Activity

Table 2.3 Abbreviations used in the description of the LRd model

Acap Capacitative area of membrane cm2
Cm Membrane capacitance F=cm2
F Faradays constant 96 485 C=mol
IK;t Total K C current through all ion channels A=cm2
INa;t Total NaC current through all ion channels A=cm2
ICa;t Total Ca2C current through all ion channels A=cm2
INaCa NaC  Ca2C exchanger A=cm2
INaK NaC  K C pump A=cm2
JSR Junctional sarcoplasmic reticulum
NSR Network sarcoplasmic reticulum
Ileak Ca2C leak from NSR mM=ms
Irel Ca2C release from JSR mM=ms
Itr Ca2C transfer from NSR to JSR mM=ms
Iup Ca2C uptake into NSR mM=ms
Vmyo Volume of myoplasm L
Vjsr Volume of JSR L
Vnsr Volume of NSR L
ŒK C i Intracellular concentration of K C mM
ŒNaC i Intracellular concentration of NaC mM
ŒCa2C i Intracellular concentration of free Ca2C mM
ŒCa2C i;t Total intracellular concentration of bound and free Ca2C mM
ŒCa2C jsr JSR concentration of free Ca2C mM
ŒCa2C jsr;t Total JSR concentration of bound and free Ca2C mM
ŒCa2C nsr NSR concentration of free Ca2C mM

2.9.4.4 LRd Model Updates

The development of the LRd model continued after 1994 and the latest version is
described in the 2007 work [306] by Livshitz and Rudy. Among the major updates,
we recall those introduced in 1995 by [583], in particular the original formulation
of the T-type transient calcium current ICa.T / and the introduction of the rapid and
slow potassium currents IKr , IKs , replacing the obsolete time-dependent potassium
current IK .
T-type calcium current ICa.T / . The magnitude of this current is given by

ICa.T / D gCa.T / b 2 g.v  ECa /;

where gCa.T / D 0:05 mS=cm2 and

ECa D .RT=2F / log.ŒCa2C o =ŒCa2C i /:

The gating variables b and g follow the Hodgkin-Huxley formalism (2.22).

2.9 Cardiac Action Potential Models 55

Time-dependent rapid potassium current IKr .

IKr D gKr Xr r.v  EKr /;

with
p
gKr D 0:02614 ŒK C o =5:4; EKr D .RT=F / log.ŒK C o =ŒK C i /;
1
rD
1 C exp..v C 9/=22:4/

and Xr following (2.22).

Time-dependent slow potassium current IKs .

IKs D gKs Xs1 Xs2 .v  EKs /;

where
 
0:6
gKs D 0:433 1 C ;
1 C .0:000038=ŒCa2C i /1:4
 C 
ŒK o C PNa;K ŒNaC o
EKs D .RT=F / log :
ŒK C i C PNa;K ŒNaC i

and Xs1 , Xs2 satisfying (2.22).

With these updates, the total K C and Ca2C currents become

IK;t D IKr C IKs C IK1 C IKp

and

ICa;t D ICa.L/ C ICa.T / C InsCa C IpCa C IbCa ;

respectively.

2.9.5 Charge Conservation in Ionic Models

Membrane models have proven to be a useful and widely accepted tool for studying
the electrophysiology and contractility of excitable cells. The second-generation of
membrane ionic models incorporates the dynamics of the cytoplasmic concentra-
tions of NaC , K C and Ca2C and in the literature concerns regarding their behavior
in response to prolonged periods of rapid pacing have been reported, see e.g.
[163, 213, 244, 307, 417, 573]. In particular, drift of intracellular ion concentrations
and transmembrane potential v and the existence of an infinite number of steady
56 2 Mathematical Models of Cellular Bioelectrical Activity

states, are often cited as problems with such models, see [213, 573]. In order to
address these concerns, models based on the principle of charge conservation have
been formulated. In this formulation, the differential equation computing v from
the transmembrane current, i.e. the first equation in (2.24), is replaced with an
algebraic equation relating v to intracellular ion concentrations, see [163, 539]. This
algebraic approach has proven to be stable with respect to drift in the computed ion
concentrations and v. We present here the algebraic approach applied to the LRd
model, as done in [244].
Introducing the total intracellular calcium concentration ŒCa2C i;t , given by the
sum of troponin (trpn) and calmodulin (cmdn) bound and free Ca2C , i.e.

ŒtrpnŒCa2C i ŒcmdnŒCa2C i
ŒCa2C i;t D ŒCa2C i C C ;
ŒCa2C i C Km;trpn ŒCa2C i C Km;cmdn

and the total JSR calcium concentration ŒCa2C jsr;t , given by the sum of calse-
questrin (csqn) bound and free JSR Ca2C , i.e.

ŒcsqnŒCa2C jsr
ŒCa2C jsr;t D ŒCa2C jsr C ;
ŒCa2C jsr C Km;csqn
we have that
d ŒCa2C i;t 2C
1 d ŒCa i d ŒCa2C jsr;t 1 d ŒCa
2C
jsr
D ˇmyo ; D ˇjsr : (2.29)
dt dt dt dt
Substituting (2.29) into Eqs. (2.26) and (2.27), we obtain

d ŒCa2C i;t Acap Vnsr Vjsr

D .ICa;t  2INaCa /  .Iup  Ileak / C Irel ; (2.30)
dt 2Vmyo F Vmyo Vmyo

d ŒCa2C jsr;t
D Itr  Irel : (2.31)
dt
Combining (2.28), (2.30) and (2.31), we have

d ŒCa2C i;t Acap Vnsr d ŒCa2C nsr Vjsr d ŒCa2C jsr;t

D .ICa;t  2INaCa /   ;
dt 2Vmyo F Vmyo dt Vmyo dt
and we finally derive the system
8
ˆ Vmyo F d ŒK C i
ˆ
ˆ D .IK;t  2INaK /;
ˆ
ˆ Acap dt
ˆ
< V F d ŒNaC 
myo i
D .INa;t C 3INaK C 3INaCa /;
ˆ
ˆ Acap  dt
ˆ
ˆ 
ˆ 2Vmyo F d ŒCa i;t Vnsr d ŒCa2Cnsr Vjsr d ŒCa2Cjsr;t
2C
:̂ C C D.ICa;t 2INaCa /:
Acap dt Vmyo dt Vmyo dt
(2.32)
2.9 Cardiac Action Potential Models 57

The total transmembrane ionic current Iion , given by

Iion D IK;t C INa;t C ICa;t C INaK C INaCa ;

provides the change in the transmembrane potential v according to the current

conservation law applied to the cell membrane modeled as a capacitor in parallel
with the ionic current
dv
Cm C Iion D 0: (2.33)
dt
Combining (2.32) and (2.33), we have

dv
Cm D Iion
dt
D .IK;t C INa;t C ICa;t C INaK C INaCa /

Vmyo F d ŒK C i d ŒNaC i d ŒCa2C i;t (2.34)
D C C2
Acap dt dt dt 
Vnsr d ŒCa2C nsr Vjsr d ŒCa2C jsr;t
C2 C2 ;
Vmyo dt Vmyo dt

hence it holds
 
V F C C Vjsr
Cm v Amyo
cap
ŒK i CŒNa i C2ŒCa 2C
i;t C2 Vnsr
Vmyo
ŒCa 2C
nsr C2 Vmyo
ŒCa 2C
jsr;t DC0 ;

with C0 constant. This equation can be interpreted as a relationship between the

electric potential and the charge across a capacitor; indeed, applying the Gauss law
to the membrane as a capacitor in an electric circuit, we have Cm .ui  ue / D  with
 the net charge difference between the intra- and extracellular charge densities
given by i  e . In the previous model, we have considered only NaC , K C and
Ca2C , which are the most important ions for the generation of the action potential,
disregarding the contribution of other ions and we have implicitly assumed that
the extracellular ionic concentrations are constant. Therefore, the constant C0 is
given by

Vmyo F  C 
C0 D e D  ŒK e C ŒNaC e C 2ŒCa2C e ;
Acap

yielding the latent algebraic conservation relationship

C ŒNaC i  ŒNaC e C 2.ŒCa2C i;t C

Vmyo F C C
vD Acap Cm .ŒK i  ŒK e
Vnsr Vjsr (2.35)
ŒCa2C nsr C jsr;t   ŒCa
ŒCa2C 2C
e //:
Vmyo Vmyo
58 2 Mathematical Models of Cellular Bioelectrical Activity

As a consequence, it is clear from this conservation law that the initial values of v
and of the ion concentrations can not be selected arbitrarily.
Summarizing, we have two equivalent mathematical formulations:
• The common full-differential approach, composed of the differential system for
the state variables transmembrane potential, gating variables and concentrations
of ŒNaC i , ŒK C i , ŒCa2C i , ŒCa2C nsr , ŒCa2C jsr , where due to the latent conserva-
tion relationship, the state variables are not linearly independent, requiring that
the initial condition must satisfy the conservation law;
• The differential-algebraic approach, composed of the differential system for
the concentrations of ŒNaC i , ŒK C i , ŒCa2C i , ŒCa2C nsr , ŒCa2C jsr and gating
variables, that is coupled with the algebraic relationship for the transmembrane
potential.
We remark that in the former approach, as a consequence of the linear dependence of
the state variables, the resulting jacobian matrix is singular. Therefore, the study of
stability of equilibrium points and cycles as functions of relevant parameters should
be performed using the differential-algebraic formulation.
When an external current is applied, Eq. (2.33) becomes

dv
Cm C Iion D Iapp :
dt
To satisfy the algebraic conservation relation between the transmembrane poten-
tial and ionic concentrations, it is required that charges carried by a stimulation
current are accounted for a specific ionic species. Usually, in many simulation
studies of periodical pacing, K C is chosen as a charge ion for the stimulus current.
Therefore Eq. (2.25) is modified as follows:

d ŒK C i Acap
D .IK;t  2INaK  Iapp /:
dt Vmyo F

In this way, we eliminate the drift behavior of ion concentrations. Otherwise,

proceeding as in (2.34) and integrating on Œ0; t, the drift term
Z t
Vdrift D Iapp .s/ ds;
0

would be added at the right hand side of Eq. (2.35), which might alterate the
membrane dynamics for long-term pacing, see [244].
2.9 Cardiac Action Potential Models 59

2.9.6 Action Potential Duration Restitution Curve

The restitution function is a mesoscopic characteristic of cardiac tissue. It describes

the duration of an action potential (APD), which is the time interval when the cardiac
cell is in an excited state, as a function of the diastolic interval (DI), which is the
time interval when the cardiac cell is in an unexcited state. At large DIs, the ionic
currents in the cell have time to go back to the resting state and the corresponding
APD is close to the maximum APD. If the DI is decreased, some of the ionic currents
are still activated and the corresponding APD is smaller. There is a minimum DI for
which the ionic currents are so activated that is not possible to elicit another action
potential. This produces a minimum APD different from zero in the restitution curve
of the APD. Figure 2.5 reports the restitution curve calculated using the LRd model.

2.9.7 Reduced Ionic Models

2.9.7.1 The FitzHugh-Nagumo Model and Other Cubic-Like Models

for the Excitation Phase

In order to investigate phenomena on larger spatial and temporal scales, and also for
theoretical purposes, several reduced membrane models have been developed that
do not seek to model sub-cellular processes, but only to provide an action potential

180

160
(ms)

140
90
APD

120

100

0 500 1000 2000 3000 4000 5000

DI (ms)

Fig. 2.5 Action potential duration restitution curve computed with the LRd model
60 2 Mathematical Models of Cellular Bioelectrical Activity

at a minimal computational cost. The unknowns of these models are a normalized

transmembrane potential v and a gating variable w, following the general kinetics
8
ˆ dv
< D f .v; w/
dt
:̂ dw D g.v; w/:
dt

• FitzHugh-Nagumo model [178]:

f .v; w/ D kv.v  a/.v  1/  w; g.v; w/ D .v  w/I

• Roger-McCulloch model [422]:

f .v; w/ D kv.v  a/.v  1/  vw; g.v; w/ D .v  w/I

• Aliev-Panfilov model [1]:

f .v; w/ D kv.v  a/.v  1/  w; g.v; w/ D . v.v  1  a/ C w/I

• Mitchell-Schaeffer model [333]:

8
ˆ 1w
w v < if v  vgate ;
f .v; w/ D  v2 .v  1/  ; g.v; w/ D open
w
in out :̂ if v > vgate :
close

Here 0 < a < 1; k; ; ; in ; out ; open ; close ; 0 < vgate < 1 are given
constants;
• Morris-Lecar model [337]:

gCa m1 .v  vCa / C gK w.v  vK / C gL .v  vL /

f .v; w/ D
C (2.36)
w1 mw
g.v; w/ D
w

where

m1 .v/ D 0:5Œ1 C tanh..v  v1 /=v2 /

w1 .v/ D 0:5Œ1 C tanh..v  v3 /=v4 /
w .v/ D 1= cosh..v  v3 /=.2v4 //:
2.9 Cardiac Action Potential Models 61

Table 2.4 Three calibrations gCa v3 v4

of the Morris-Lecar
.mS=cm2 / (mV) (mV) .ms1 /
parameters gCa ; v3 ; v4 ;
ML1 4.4 2. 30. 0.04
ML2 4 12 17.4 0.066
ML3 4 12 17.4 0.23

The parameters of the Morris-Lecar model are: Cm D 20 F=cm2, vCa D 120 mV,
vK D 84 mV, gK D 8 mS=cm2, vL D 60 mV, gL D 2 mS=cm2, v1 D 1:2 mV,
v2 D 18 mV. For the remaining parameters, we consider three different calibrations
denoted by ML1, ML2, ML3 in Table 2.4 and used later in the section on
bifurcation diagrams.

2.9.7.2 The Fenton-Karma Model

This reduced model, proposed in [170], consists of three ordinary differential

equations for the transmembrane potential v and two gating variables u and w:
8
ˆ dv
ˆ
ˆ D f .v; u; w/
ˆ
ˆ
< dt
du  C
D H.v  vc /.1  u/= u.v/  H.v  vc /u=
ˆ dt
ˆ
u
ˆ
ˆ dw
:̂ D H.vc  v/.1  w/= w  H.v  vc /w= C
w ;
dt

where H is the Heaviside function centered at zero and the ionic current
f .v; u; w/ D Ifi  Iso  Isi is the sum of the three currents

Ifi .v; u/ D uH.v  vc /.1  v/.v  vc /= d

Iso .v/ D H.v  vc /= o C H.v  vc /= r

Isi .v; w/ D w.1 C tanh.k.v  vsic ///=.2 si /:

The values of the parameters in the previous expressions can be found in Table I
of [170].
62 2 Mathematical Models of Cellular Bioelectrical Activity

2.9.8 Phase-Plane Analysis of the FitzHugh-Nagumo (FHN)

Model

The FitzHugh-Nagumo model can be derived from a simplified electrical circuit

model of the cell membrane. The circuit consists of a capacitor (with current ic )
in parallel with a nonlinear current–voltage device (with current j D F .u/) and
a serial system with a resistor, an inductance, and a battery (with current i ). Let
u D ui ue be the transmembrane potential and vR ; vL ; v0 be the resistor, inductance,
and battery potentials, respectively. Given an applied current Ia , Kirchhoff’s laws
state that

i c C i C j D Ia ;
u D v0 C vL C vR :

From ic D C du
dt , vR D Ri (Ohm’s law), vL D L dt (Faraday’s law), we have
di

8
ˆ du
<C D i  F .u/ C Ia ;
dt
:̂ L di D u  Ri  v0 :
dt
In order to write this system in dimensionless form, we scale the time as D
Rm t=L, where Rm D 1=F 0 .0/ and we define the dimensionless variables v. / D
u.L =Rm /=vm ; w. / D Rm i.L =Rm/=vm , where vm is the maximal value of u.
Then
8
ˆ
ˆ
dv
D
L du
D
L L
.i  F .u/ C Ia /D 2 .
Rm i Rm F .u/ Rm
 C
< Ia /;
d vm Rm dt vm Rm C Rm C vm vm vm
ˆ dw Rm L di L u R Rm i v0
:̂ D D .v  Ri  v0 / D   :
d vm Rm dt vm L vm Rm vm vm

Defining vO 0 D v0 =vm ; FO .v/ D Rvmm F .vm v/; IOa D Rvmm Ia ; D R=Rm ;  D

2
CRm =L, we have
8
ˆ dv
< D w  FO .v/ C IOa ;
d
:̂ dw D v  w  vO 0 :
d
This system is a particular instance of the generalized FitzHugh-Nagumo planar
dynamical system
8
ˆ dv
< D f .v; w/ C IOa ;
d (2.37)
:̂ dw D g.v; w/:
d
2.9 Cardiac Action Potential Models 63

0.5
Ω−, g<0
g
0.4
Ω+g, g>0
Ω−f , f<0
0.3 g(v,w)=0

Ω+, f>0
0.2 f
w

f(v,w)=0
0.1
wmax
h h h+
− 0
0
w
min

−0.1

−0.2
−0.4 −0.2 0 0.2 0.4 0.6 0.8 1 1.2
v

Fig. 2.6 v; w-phase plane, nullclines and notations for the FHN model

We assume a situation as the one in Fig. 2.6, where the nullcline f .v; w/ D 0
is a cubic-like function in v with 3 distinct zeros, local maximum value wmax , local
minimum value wmin and the nullcline g.v; w/ D 0 has only one intersection point
.v ; w / with f D 0 (for an example with more than one intersection, see Fig. 2.9).
More precisely, we assume that it exists an interval .wmin ; wmax / such that:
– For w > wmax 9Šv W f .v; w/ D 0, defining a function v D h .w/;
– For w < wmin 9Šv W f .v; w/ D 0, defining a function v D hC .w/;
– For wmin < w < wmax there are three real roots h .w/ < h0 .w/ < hC .w/ of
f .v; w/ D 0.
The cubic-like nullcline f .v; w/ D 0 partitions the v; w-plane into two regions,
˝f D f.v; w/ W w > w; where f .v; w/ D 0g and ˝fC D f.v; w/ W
w < w; where f .v; w/ D 0g. We assume that f < 0 in ˝f and f > 0 in
˝fC . Analogously, the other nullcline g.v; w/ D 0 partitions the plane into two
regions ˝g D f.v; w/ W w > w; where g.v; w/ D 0g and ˝gC D f.v; w/ W
w < w; where g.v; w/ D 0g, and we assume that g < 0 in ˝g and g > 0
in ˝gC . These assumptions guarantee that the partial derivatives of f and g are
nonpositive or nonnegative but we will make the stronger assumptions @w f .v; w/ <
0; @v f .v; w/ > 0; and @v g.v; w/ > 0; @w g.v; w/ < 0.
64 2 Mathematical Models of Cellular Bioelectrical Activity

According to the chosen initial conditions .v0 ; w0 / and the applied current IOa , the
generalized FitzHugh-Nagumo system (2.37) admits solutions with quite different
dynamics. The intersection point .v ; w / between the nullclines f D 0, g D 0 is
an equilibrium point and its stability is determined by the eigenvalues 1 ; 2 of the
Jacobian matrix evaluated at .v ; w /

@v f @w f
J D W
@v g @w g

– We have stability if 1 ; 2 are both real and negative (stable node or sink) or if
they are a complex conjugate pair with negative real part (stable spiral);
– We have instability if 1 ; 2 are both real positive (unstable node or source)
or they have opposite signs (saddle) or they are a complex conjugate pair with
positive real part (unstable spiral).
Equivalently, the stability of a planar dynamical system is often described in terms
of the trace tr.J / D 1 C 2 and determinant det.J / D 1 2 of the Jacobian matrix.
For our generalized FitzHugh-Nagumo system (2.37), we have stability if
.v ; w / lies on the lower branch h .w/ or upper branch hC .w/ of the nullcline
f D 0, while we have instability if .v ; w / lies on its middle branch h0 .w/. In
Fig. 2.7, we consider first some examples without applied current, IOa D 0. We will
then vary IOa in Figs. 2.8 and 2.9 we will consider a case with three equilibrium
points. In each figure, the left panel shows the nullclines (dashed lines) and the
trajectory (continuous line) in the phase plane v; w starting from the initial point
.v0 ; w0 / (marked with ˘); the equilibrium point .v ; w / is marked with ?.
(a) FHN dynamics with zero applied current IOa (Fig. 2.7).
(a1) Stable equilibrium: subthreshold response. Consider first w0 2 .wmin ; wmax /
and v0 < h0 .w0 /. Then v returns to the equilibrium point v (Fig. 2.7 top,
stable spiral), because in the region where f < 0 if the initial point .v0 ; w0 /
is above the nullcline g D 0, then here g < 0 and .v; w/ returns to the
equilibrium point .v ; w /, and if .v0 ; w0 / is below the nullcline g D 0,
then here g > 0 so w initially increases until the trajectory crosses g D 0
into the region where g < 0 and .v; w/ again returns to .v ; w /. Analogous
considerations apply when w0 2 .wmin ; wmax / and v0 < h0 .w0 / in the region
where f > 0.
(a2) Stable equilibrium: action potential (Fig. 2.7 middle, stable spiral). If instead
v0 > h0 .w0 / (for simplicity we still assume w0 2 .wmin ; wmax /), then we
have an action potential. First, we have an excitation phase during a time
D O./, where v tends to the upper branch hC .w/ of the nullcline f D 0,
2.9 Cardiac Action Potential Models 65

Phase Plane and Nullclines

0.3 1

V
0
0.2
−1
0 1 2 3
t
W

0.1
0.2

W
0
V =0 eig 1 =−5.25+8.79986i
W =0 eig 2 =−5.25−8.79986i −0.2
−0.1 0 1 2 3
−0.5 0 0.5 1 t
V
Phase Plane and Nullclines
0.3 1
V
0
0.2
−1
0 1 2 3
t
W

0.1
0.2

0
W

0
V =0 eig 1 =−5.25+8.79986i
W =0 eig 2 =−5.25−8.79986i −0.2
−0.1 0 1 2 3
−0.5 0 0.5 1 t
V
Phase Plane and Nullclines
0.3 1
V

0
0.2
−1
0 1 2 3 4 5
t
W

0.1
0.5

0
W

0
V =0 eig 1 =4.75+8.511i
W =0 eig 2 =4.75−8.511i −0.5
−0.1 0 1 2 3 4 5
−0.5 0 0.5 1 t
V

Fig. 2.7 FHN model (2.37) with IOa D 0;  D 0:01. In top and middle panels f .v; w/ D v.v 
0:1/.1  v/  w; g.v; w/ D v  w=2, while in bottom panels f .v; w/ D v.v C 0:1/.1  v/ 
w; g.v; w/ D v  w=2. Phase plane trajectories, nullclines, coordinates of the marked equilibrium
point ? and associated eigenvalues (left), solutions v; w as functions of time (right)
66 2 Mathematical Models of Cellular Bioelectrical Activity

Phase Plane and Nullclines

0.5 1

0.4

V
0

0.3 −1
0 1 2 3 4 5
t
W

0.2
0.2
0.1

W
0
0
V =0 eig 1 =−6+9.16515i
W =0 eig 2 =−6−9.16515i −0.2
−0.1 0 1 2 3 4 5
−0.5 0 0.5 1 t
V
Phase Plane and Nullclines
0.5 1

0.4
V
0

0.3 −1
0 1 2 3 4 5
t
W

0.2
0.4
0.1
W

0.2
0 V = 0.5 eig 1 =20.5692
W =0.25 eig 2 =2.43082
0
−0.1 0 1 2 3 4 5
−0.5 0 0.5 1 t
V
Phase Plane and Nullclines
0.5 1
V =0.81538 eig 1 =−25.8826
W =0.40769 eig 2 =−6.18714
0.4
V

0.3 −1
0 1 2 3 4 5
t
W

0.2
0.5
0.1
W

0
0
−0.1 0 1 2 3 4 5
−0.5 0 0.5 1 t
V

Fig. 2.8 FHN model (2.37) with varying IOa , with f .v; w/ D v.v  0:1/.1  v/  w; g.v; w/ D
v  2w;  D 0:01. IOa D 0 (top), 0:15 (middle), 0:3 (bottom). Phase plane trajectories, nullclines,
coordinates of the marked equilibrium point ? and associated eigenvalues (left), solutions v; w as
functions of time (right)
2.9 Cardiac Action Potential Models 67

Phase Plane and Nullclines

0.3 1
V =0.77016 eig 1 =−24.0703

V
W =0.15403 eig 2 =−10.2438 0
0.2
−1
0 1 2 3 4 5
t
W

0.1 V = 0.12984 eig 1 = 24.9783

W =0.025969 eig 2 =−1.66426 0.2

W
0
V =0 eig 1 =2.5+6.6144i
W =0 eig 2 =2.5−6.6144i −0.2
−0.1 0 1 2 3 4 5
−0.5 0 0.5 1
t
V

Fig. 2.9 FHN model (2.37) with nullclines intersecting in three points: f .v; w/ D v.v C 0:1/.1 
v/  w; g.v; w/ D v  5w;  D 0:01; IOa D 0. Phase plane trajectories, nullclines, coordinates
of the marked equilibrium points ? and associated eigenvalues (left), solutions v; w as functions of
time (right)

because in this region f > 0. Indeed, if we scale time as 0 D = and define

vO . 0 / D v. 0 /; w.
O 0 / D w. 0 /; we have @ 0 vO D @ v; @ 0 wO D @ w; and
8
ˆ d vO
< D f .Ov; w/;
O
d 0
:̂ d wO D g.Ov; w/:
O
d 0
Therefore, for small , wO  w0 is approximately constant and vO evolves
toward hC .w0 /. Then for > O./ we have an excited state where .v; w/
tends to stay on the upper branch, where f D 0 and w evolves approximately
dw
according to D g.hC .w/; w/. This “plateau” lasts approximately up to
d
the finite time
Z wmax
dw
;
wmin g.hC .w/; w/

when w reaches wmax at the local maximum of f D 0. After this time, we

have a recovery phase, where w is again approximately constant and v now
evolves toward the lower branch h .w/ since we are again in the region
where f < 0. Once the lower branch is reached, .v; w/ tends to return to
the equilibrium point .v ; w / since the latter lies on the lower branch, i.e.
g.h .w /; w / D 0.
(a3) Unstable equilibrium: limit cycle associated with a periodic solution (Fig. 2.7
bottom, unstable spiral). If otherwise .v ; w / lies on the middle branch
h0 .w/, the equilibrium is unstable and the trajectory of .v; w/ starts another
excitation phase generating another action potential and so on. In this case,
we have a limit cycle associated with a periodic solution.
68 2 Mathematical Models of Cellular Bioelectrical Activity

(b) Effect of increasing the applied current IOa (Fig. 2.8).

(b1) Stable equilibrium: subthreshold response (Fig. 2.8 top, stable spiral). For
IOa D 0, we have a subthreshold response as in case (a1) shown in
Fig. 2.7, top.
(b2) Unstable equilibrium: periodic limit cycle (Fig. 2.8 middle, source). For IOa D
0:15, the nullcline f D 0 is shifted upward and .v ; w / becomes a source
point, yielding a periodic limit cycle corresponding to a periodic train of
action potentials.
(b3) Stable equilibrium (Fig. 2.8 bottom, sink). For IOa D 0:3, the nullcline f D 0
is further shifted upward and .v ; w / becomes a sink point quickly attracting
the trajectory to the equilibrium.
(c) Multiple equilibrium points (Fig. 2.9). A FHN system with f .v; w/ D v.v C
0:1/.1  v/  w; g.v; w/ D v  5w; IOa D 0; having three equilibrium points
is shown in Fig. 2.9: from left to right, they are an unstable spiral, a saddle
point, and a sink. The trajectory quickly tends to the lower branch of the f D 0
nullcline, follows it until it is repelled by the two unstable equilibrium points,
tends to the upper branch of the f D 0 nullcline and terminates in the sink.
(d) Cathodal and anodal current pulses (Figs. 2.10 and 2.11).
In the previous simulations, the action potentials were generated by appro-
priate choices of initial conditions v0 ; w0 for the transmembrane potential
and recovery variables. A more physiological technique to generate an action
potential consists in starting from resting initial conditions v0 ; w0 and applying
an appropriate current pulse IOa for a sufficiently long time interval Tstim . In the
FitzHugh-Nagumo model (2.37), the application of such current pulse shifts the
f D 0 nullcline upward if IOa > 0 (see Fig. 2.10) or downward if IOa < 0 (see
Fig. 2.11). This shift changes temporarily the equilibrium point for the duration
of the stimulation interval Tstim so that v0 ; w0 is no longer an equilibrium point

0.4 1
v

0.3 0

−1
0.2 0 1 2 3
t
w

0.1 0.2
0.1
w

0 0
−0.1
−0.1 0 1 2 3
−1 −0.5 0 0.5 1
t
v

Fig. 2.10 Cathode make due to current pulse Ia D 0:2 lasting 0.02 ms for FHN model (2.37) with
f .v; w/ D v.v  0:1/.1  v/  w; g.v; w/ D v  w=2;  D 0:01. Phase plane trajectories and
nullclines (left), solutions v; w as functions of time (right)
2.9 Cardiac Action Potential Models 69

1
0.2

v
0
0.1
−1
0 1 2 3
0 t
w

0.2
−0.1 0.1

w
0
−0.2 −0.1
0 1 2 3
−1 −0.5 0 0.5 1 t
v

1
0.2

v
0
0.1
−1
0 1 2 3
0 t
w

0.2
−0.1 0.1
w

0
−0.2 −0.1
0 1 2 3
−1 −0.5 0 0.5 1 t
v

Fig. 2.11 Anode break due to current pulse Ia D 0:2 lasting 0.1 ms (top) or 0.6 ms (bottom) for
FHN model (2.37) with f .v; w/ D v.v  0:1/.1  v/  w; g.v; w/ D v  w=2;  D 0:01. Phase
plane trajectories and nullclines (left), solutions v; w as functions of time (right)

and the variables v; w start a trajectory toward the closest nullcline branch.
When the pulse is turned off at time Tstim , the f nullcline returns to the original
curve f D 0. By this time, v and w have reached a point in phase-space that, for
proper choices of pulse amplitude and interval duration, belongs to the excitable
region, so that an action potential is generated. Figure 2.10 shows such an
action potential generated by a pulse with Ia D 0:2, Tstim D 0:02, known
as cathode make mechanism. Instead Fig. 2.11 shows two cases with action
potentials generated by pulses with Ia D 0:2, Tstim D 0:1 (top) or Ia D 0:2,
Tstim D 0:6 (bottom), known as anode break mechanism. Additional activation
mechanisms (known as cathode break and anode make) are present in space-
time cardiac domains, in particular in three-dimensional domains, where more
complex virtual electrode polarization (VEP) activation patterns are present, see
Chap. 9.1.
70 2 Mathematical Models of Cellular Bioelectrical Activity

2.9.9 Bifurcation Diagrams

There are several mechanisms from which limit cycles arise in models of excitable
cells. A periodic behavior of the transmembrane potential can be thought of as a
transition back and forth between hyperpolarized and depolarized states. We review
here four main mechanisms that govern the onset of oscillations in some of the
previous ionic models. For more details we refer to e.g. [288, 419].
A possible motivation for studying the dynamical response of an ionic model
to a sustained injection of a constant current, i.e. a bias current, is for example
the fact that during ischemia there is an accumulation of extracellular potassium in
the ischemic zone, leading to an increase of the membrane potential. This creates
a current flow between the injured and healthy zone. We study here this effect
by modeling the injury current as a time- and voltage-independent bias current,
and considering simple membrane models. We illustrate the first two mechanisms
considering the simplest FHN model.
The dynamics of the FHN system as a function of an applied bias current IOa ,
i.e. a continuous constant current, can be summarized by a bifurcation diagram with
respect to IOa . We present now some bifurcation and frequency diagrams computed
by the XPPAUT software [165].
(a) Choosing  D 0:1, i.e. modifying the time scale of the v variable, we
find in Fig. 2.12 that by increasing IOa , a periodic solution appears through
a supercritical Hopf-Bifurcation (HB) point near IOa D 0:075, leading to a
branch of small-amplitude periodic orbits of nonzero frequencies. By increasing
IOa further, we find that the periodic orbits have decreasing amplitudes and
then disappear at another HB point near IOa D 0:17. This type of dynamical
phenomenon, with oscillations arising with arbitrary small amplitude via a
supercritical HB point and with nonzero oscillation frequency, is also called
soft excitation by some authors.

V Frequency
1
0.0395
0.8
0.039
0.6
0.0385

0.4 0.038

0.0375
0.2
0.037
0
0.0365

-0.2 0.036

0.0355
-0.4
0.035
0 0.05 0.1 0.15 0.2 0.25 0 0.05 0.1 0.15 0.2 0.25
Ia Ia

Fig. 2.12 Left: bifurcation diagram for FHN model with f .v; w/ D v.v  0:1/.1  v/  w C
IOa ; g.v; w/ D v  2w;  D 0:1 with respect to IOa . Right: associated frequency diagram
2.9 Cardiac Action Potential Models 71

(b) Choosing  D 0:01, i.e. shortening the time constant of v, we find in Fig. 2.13
that for all values of IOa  0 there is a unique equilibrium point (EP), which
loses its asymptotical stability for values of IOa between two subcritical Hopf
bifurcation points, HB1 and HB2 and we see also the presence of a stable
limit cycle for values of IOa ranging approximately between 0.03 and 0.22.
Near the subcritical HB1 point there is a large-amplitude stable limit cycles and
bistability between stable oscillations and a stable equilibrium point, separated
by an unstable small-amplitude periodic orbit. The unstable limit cycle turns
back and coalesces at a turning bifurcation point with the stable limit cycle at a
value of IOa that is smaller than the one associated to the subcritical HB point. A
zoom of the left part of this interval is shown in Fig. 2.13, right panel, revealing a
subcritical HB1 near IOa D 0:0319. The branch of unstable periodic orbits comes
out near HB1 and lies on the side of HB1 where there is a stable branch of EP.
Decreasing IOa from the HB1 value, a saddle-node periodic bifurcation (SNPB)
point appears, connecting a stable and an unstable branch of limit cycles. Above
it two stable structures appear, a stable EP and a stable limit cycle, as well as an
unstable limit cycle.
The frequency of the limit cycle as a function of IOa , also computed with
XPPAUT, is shown in Fig. 2.14, left panel, together with a zoom near the Hopf
bifurcation point shown in the right panel.
This type of onset of oscillations arising abruptly with stable large-amplitude
oscillations and with relatively high frequencies it is also called hard excitation
mechanism or also saddle-node bifurcation of limit cycles (SNBLC), since for
increasing IOa there is the creation of a pair of limit cycles, one stable and the
other unstable with non zero frequencies.
An analogous reflected situation happens at the right part of the bifurcation
interval (zoom not shown), where unstable small-amplitude periodic orbits now
appear on the stable equilibrium branch on the right side of the right HB point.

V V

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0

-0.2 -0.2

-0.4 -0.4

0 0.05 0.1 0.15 0.2 0.25 0.3 0.0315 0.0316 0.0317 0.0318 0.0319
Ia Ia

Fig. 2.13 Left: bifurcation diagram for FHN model with f .v; w/ D v.v  0:1/.1  v/  w C
IOa ; g.v; w/ D v  2w;  D 0:01 with respect to IOa . Right: zoom of the left bifurcation point
72 2 Mathematical Models of Cellular Bioelectrical Activity

Frequency Frequency
0.02 0.02

0.015 0.015

0.01 0.01

0.005 0.005

0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.0315 0.0316 0.0317 0.0318 0.0319
Ia Ia

Fig. 2.14 Left: frequency diagram for FHN model with f .v; w/ D v.v0:1/.1v/w; g.v; w/ D
v  2w;  D 0:01. Right: zoom of the left interval

V Frequency
40 0.025

20 0.02

0 0.015

-20 0.01

-40 0.005

-60 0
60 80 100 120 140 160 180 200 220 240 60 80 100 120 140 160 180 200 220 240

Ia Ia

Fig. 2.15 Left: bifurcation diagram for the Morris-Lecar model (2.36) with respect to IOa with the
parameter calibration ML1 in Table 2.4. Right: frequency diagram

The previous mechanisms (a) and (b) of onset of oscillations are associated
with local bifurcations that are local phenomena where limit cycles originate
with small amplitude when a single equilibrium state changes from stable to
unstable or viceversa. A bifurcation diagram of the same local type (b) is also
obtained for the Morris-Lecar model using the parameter calibration (ML1)
reported in Table 2.4; see Fig. 2.15.
Moreover, the bifurcation diagram of the Hodgkin-Huxley model, displayed
in Fig. 2.16, shows the same bistable behavior between a SNPB and a HB1.
Differently from the other models, a second supercritical Hopf bifurcation point
appears, where stable periodic orbits of small amplitude come out lying near the
unstable branch. See Fig. 2.17 for the associated frequency diagram.
2.9 Cardiac Action Potential Models 73

V V
40 40

20 20

0 0

-20 -20

-40 -40

-60 -60

-80 -80
0 20 40 60 80 100 120 140 160 6 8 10 12 14
Ia Ia

Fig. 2.16 Left: bifurcation diagram for HH model with respect to IOa . Right: zoom of the left
bifurcation point

Frequency
0.2

0.15

0.1

0.05

0
0 20 40 60 80 100 120 140 160
Ia

Fig. 2.17 Left: frequency diagram for HH model with respect to IOa

(c) We now look at two other mechanisms of onset of oscillations that are very
different from the previous mechanisms (a) and (b) since they are associated
with global bifurcations and show that it is possible to create or destroy spiking
with arbitrary low frequencies when the amplitude of the sustained injected
current is increased. A type of global bifurcation is related to the appearance
of homoclinic bifurcation. We recall that an homoclinic orbit is a closed curve
that has a single equilibrium point located along the curve and the period of the
orbit is infinite. Figure 2.18 displays the bifurcation diagram with respect to the
injected current IOa for the Morris-Lecar model using the calibration parameter
74 2 Mathematical Models of Cellular Bioelectrical Activity

V Frequency
0.05
40 0.045

0.04
20
0.035

0.03
0
0.025

0.02
-20
0.015

-40 0.01

0.005

-60 0
-20 0 20 40 60 80 100 120 140 20 40 60 80 100 120
Ia Ia

Fig. 2.18 Transmembrane potential (left) and frequency (right) bifurcation diagrams with respect
to IOa for Morris-Lecar model (2.36) with the parameter calibration ML2 in Table 2.4

(ML2) reported in Table 2.4, still computed with XPPAUT. The diagram starts
on the left with a stable node branch but by increasing IOa two unstable branches
of equilibrium points appear, the upper one associated with unstable spirals,
and the intermediate with saddle points. By increasing IOa further, the unstable
saddle and stable nodes coalesce at a HB bifurcation point and subsequently
disappear. A limit cycle then emerges as a homoclinic orbit that has a finite
(large) amplitude with zero frequency as shown in Fig. 2.18, right panel. This
type of dynamical behavior associated with saddle-node on a limit cycle is
sometimes called a saddle-node homoclinic orbit. By increasing IOa further, the
stable limit cycle disappears through a saddle-node bifurcation of limit cycles.
(d) We now increase the constant from 0.066 to 0.23, yielding a decrease of the
time constant of the potassium gating variable w, thus increasing its changing
rate. Figure 2.19 displays the bifurcation diagram for the Morris-Lecar model
using the parameter calibration (ML3) reported in Table 2.4. After an initial
branch of stable node, we have again two saddle node bifurcation points
connecting two unstable branches and a subcritical HB appears on the upper
branch, but now with a value between the two saddle node bifurcation points.
As in the previous case (c), the stable branch of limit cycles terminates again
with an homoclinic orbit which is now on the middle branch before the right
saddle node bifurcation point. For IOa ranging between the subcritical HB values
and the right saddle node bifurcation there are two stable equilibrium points and
a stable limit cycle, yielding a twistable dynamics.
This type of global bifurcation is called saddle-homoclinic bifurcation. The
frequency of the stable limit cycles approaches the zero frequency of the homoclinic
orbit more rapidly than in the previous SNBLC diagram.
2.9 Cardiac Action Potential Models 75

V Frequency
20 0.07

10
0.06
0
0.05
-10
0.04
-20

-30 0.03

-40 0.02

-50
0.01
-60
0
-20 -10 0 10 20 30 40 50 60 30 32 34 36 38 40 42 44

Ia Ia

Fig. 2.19 Transmembrane potential (left) and frequency (right) bifurcation diagrams with respect
to IOa for Morris-Lecar model (2.36) with the parameter calibration ML3 in Table 2.4

V V
20 20

0 0

-20 -20

-40 -40

-60 -60

-80 -80

-100 -100
0 0.5 1 1.5 2 2.5 3 0 1 2 3 4 5
iapp
iapp

Fig. 2.20 Bifurcation diagram for LR1 model with the original calibration parameters (left) and
the with the IK current scaled by a factor 3 (right)

We have also computed the bifurcation diagrams for the LR1 model with both
the original calibration parameters and with the IK current scaled by a factor
3. The results obtained with the XPPAUT software display bifurcation diagrams
of Fig. 2.20, left and right panels, similar to the diagram of Figs. 2.19 and 2.18,
respectively, but with different stability properties.
For bifurcation diagrams related to second generation models, we refer e.g.
to [57].
Chapter 3
Mathematical Models of Cardiac Cells
Arrangements: The Bidomain Model

3.1 Models of Cardiac Fibers

3.1.1 Cable Equation

We model an excitable cell as a long cylindrical cable and assume the so-called core
conduction assumption, stating that the potential along the cable depends only on
the length variable and not on radial or angular variables, so that the cable model is
one-dimensional. Then we assume that the cable is composed of a number of short
sections of length dx with isopotential membrane, and that in each section we have
only the transmembrane and axial currents, as illustrated in the following diagram
Je(x) ue(x) ue(x+Δ x)

i
reΔ x m

r Δx
i

Ji(x) u (x) u (x+Δ x)

i i

The relationship between the intra- and extracellular current fluxes Ji;e (A cm2 )
and the intra- and extracellular potentials ui;e (V ) is given by Ohm’s law

1
Ji D  @x ui D i @x ui ;
ri
1
Je D  @x ue D e @x ue ;
re

© Springer International Publishing Switzerland 2014 77

P. Colli Franzone et al., Mathematical Cardiac Electrophysiology, MS&A 13,
DOI 10.1007/978-3-319-04801-7__3
78 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

where i;e are the intra- and extracellular conductivities (1 cm1 ) and ri;e are the
intra- and extracellular resistances per unit space ( cm).
We denote by im the current across the membrane; it is an outward current, i.e. it
is considered positive when directed toward the extracellular space. By Kirchhoff’s
law of conservation of electric charges, we have
Z xCx
Ji .x/ D Ji .x C x/ C im .s/ds;
x
Z xCx
Je .x C x/ D Je .x/ C im .s/ds;
x

as illustrated in the following diagram

Je(x) Je(x+Δ x)

im Δ x

Ji(x) Ji(x+Δ x)

Δx

Assuming Ji ; Je differentiable, in the limit x ! 0 we obtain the following current

conservation laws

@x Ji D im ;
@x Je D im ;

which imply

@x .Ji C Je / D 0; i.e. Ji C Je D const:

We consider a fiber of length L, cut and healed at the two ends, leading to insulating
boundary conditions for the intracellular medium

Ji D 0 at x D 0; x D L:

Moreover, we inject a cathodal extracellular current and eject an anodal extracellular

current of strength Iapp .t/ at the two ends of the fiber, i.e.

Je D Iapp .t/ at x D 0; x D L;
3.1 Models of Cardiac Fibers 79

thus we finally obtain

Ji .0/ C Je .0/ D Ji .L/ C Je .L/ D Iapp .t/; i.e. Ji .x; t/ C Je .x; t/ D Iapp .t/:

The membrane current per unit area of the membrane surface is given by

Im D Cm @t v C Iion :

This can be expressed as a current im per unit volume by multiplying by the ratio
 between the membrane surface area for a given length of fiber and the volume
enclosed by this surface, i.e.

im D  Im D  .Cm @t v C Iion /:

Collecting together this last equation with Kirchhoff’s and Ohm’s laws, we have the
system
8
ˆ
ˆ v D ui  ue
<
i @x ui  e @x ue D Iapp .t/
ˆ @x .i @x ui / D im
:̂
im D .Cm @t v C Iion /:

By setting ue D ui  v in the second equation, we obtain

1
i @x ui  e @x ui C e @x v D Iapp .t/ ) @x ui D .e @x v  Iapp .t//;
i C e

and from the third and fourth equation, we finally obtain the cable equation
 
i
@x .@x v/  @x Iapp .t/ D im D .Cm @t v C Iion /; (3.1)
i C e

where
i e
D
i C e

is the conductivity of the bulk medium, corresponding to the conductivity of the

intra- and extracellular media connected in series. Due to the presence of gap
junctions between two subsequent myocytes, the conductivity i .x/ is practically
constant in the interior of the cell, with value given by the cytosolic conductivity,
but in the small length of the gap junctions it undergoes a strong reduction due to
the high resistivity of the gap junctions.
80 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

3.1.2 Homogenization

Let us consider an infinite fiber composed by cells connected by end-to-end

junctions and embedded in an extracellular equipotential conductor medium, i.e.
ue is constant. Supposing that the cytoplasmic fluid and the junctions are ohmic
conductors and applying the current conservation law, in the stationary regime the
intracellular current density Ji .x/ satisfies

dui d
Ji .x/ D i ; Ji .x/ D im .x/; (3.2)
dx dx
where im is the membrane current. Since the extracellular medium is assumed
equipotential, the extracellular flux vanishes and, from v D ui  ue , we have
@x D @x . Setting  D i and combining the two equations in (3.2), we get
@v @ui

 
d dv
 D im : (3.3)
dx dx

The junctions are more resistive than the cytoplasmic fluid, therefore we assume
that  varies periodically with period equal to the distance  between two subsequent
junctions. Hence,  D . x / is periodic with period , or, defining  D x ,  D ./
is periodic with period 1.
In general, if the cells present some kinds of heterogeneities, we suppose that

 D .x; /  0 > 0; (3.4)

periodic in  with period 1.

Combining (3.3) and (3.4), we obtain the following equation, describing the
profile of the electric potential along the one-dimensional fiber,
  
d x  dv
 x; .x/ D im .x/: (3.5)
dx  dx

We have two spatial scales:

• x, varying from 0 to L, the length of the fiber, is the macroscopic variable;
•  D x , varying in Œ0; 1, describes the singular cellular element and is the
adimensional microscopic variable.
Assume now that the solution of Eq. (3.5) has the following structure
 x
v .x/ D V x; ;

i.e. it is a function depending on the macro- and microscopic scales, periodic with
respect to  D x with period 1.
3.1 Models of Cardiac Fibers 81

If   1, we can let  ! 0 and expand v in powers of 

 x  x  x  x
v .x/ D V x; D V0 x; C V1 x; C  2 V2 x; C O. 3 /;
   

with the coefficients Vi .x; x / periodic in  D x of period 1.

We suppose that the coefficients Vi have zero mean value on  2 .0; 1/ for i  1,
Z 1
Vi .x; / d  D 0 for i  1;
0

then it holds
Z 1 Z 1
v .x/ D V .x; / d  D V0 .x; / d :
0 0

The derivative of v with respect to x is given by

d d  x @  x @  x  d
v .x/ D V x; D V x; C V x;
dx dx  @x  @  dx
@  x @  x 1
D V x; C V x; ;
@x  @  

thus

d @ 1 @
D C
dx @x  @

and
    
d dv d @V 1 @V
 D  C
dx dx dx   @x  @    
@ @V 1 @V 1 @ @V 1 @V
D  C C  C
@x  @x
  @   @
 @x  @  
@ @V 1 @ @V 1 @ @V 1 @ @V
D  C  C  C 2  :
@x @x  @x @  @ @x  @ @

Imposing that v is solution of (3.5) and expanding in powers of , we get

    
d dv 1 @ @V0
 D 2 
dx dx  @  @     
1 @ @V1 @ @V0 @ @V0
C  C  C 
 @ @ @x @ @ @x  
@ @V2 @ @V1 @ @V1 @ @V0
C  C  C  C 
@ @ @x @ @ @x @x @x
CO. 3 / D im .x/:
82 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

1 1
If we equal the terms of order , ,
2 
1, we get the following differential problems
8  
< @ @
.x; / V0 .x; / D 0 for  2 .0; 1/
.P 0/ @ @ (3.6)
:
V0 .x; 0/ D V0 .x; 1/;
8    
ˆ
ˆ
@ @
D 
@ @
C
ˆ
ˆ .x; / V 1 .x; / .x; / V 0 .x; /
ˆ
ˆ @ @ @x  @ 
ˆ
ˆ @ @
<  .x; / V0 .x; / for  2 .0; 1/
.P1/ @ @x
ˆ
ˆ
ˆ
ˆ V .x; 0/ D V1 .x; 1/
ˆ
ˆ Z1
ˆ 1
:̂ V1 .x; / d  D 0;
0
8     (3.7)
ˆ
ˆ
@ @
.x; / V2 .x; / D im .x; / 
@ @
.x; / V0 .x; / C
ˆ
ˆ
ˆ
ˆ @ @  @x @x
ˆ
ˆ
ˆ
ˆ 
@ @
.x; / V1 .x; / C
ˆ
ˆ
< @x  @ 
.P 2/ @ @
ˆ
ˆ  .x; / V 1 .x; / for  2 .0; 1/
ˆ
ˆ @ @x
ˆ
ˆ
ˆ
ˆ V2 .x; 0/ D V2 .x; 1/
ˆZ 1
ˆ
ˆ
:̂ V2 .x; / d  D 0:
0

All the previous problems have the following general structure

8  
< @ .x; / @ v.x; / D S.x; / for  2 .0; 1/
@ @ (3.8)
:
v.x; 0/ D v.x; 1/:

If v is a solution of (3.8), then integrating both the sides of the equation we get
Z 1   Z 1
@ @ @ @
 v d  D .x; 1/ v.x; 1/  .x; 0/ v.x; 0/ D S.x; / d :
0 @ @ @ @ 0

Since v is periodic, it must hold

@ @
.x; 1/ v.x; 1/  .x; 0/ v.x; 0/ D 0;
@ @

from which it follows

Z 1
S.x; / d  D 0: (3.9)
0
3.1 Models of Cardiac Fibers 83

It can be proved that (3.9) is also a sufficient condition for the solvability of
(3.8),Ri.e.:
1
if 0 S.x; / d  D 0, then 9Š solution v.x; / periodic in  with period 1 of (3.8).
v is unique up to a constant, that can be fixed by imposing zero mean value for
 2 .0; 1/,
Z 1
v.x; / d  D 0:
0

Problem (P0). If we multiply both the sides of (3.6) times V0 and integrate on
.0; 1/, we get
Z   Z 1  
1
@ @V0 @V0 2
0D  V0 d  D   d
0 @ @ 0 @

and
Z 1  2 Z 1  2
@V0 @V0
0D  d   0 d ;
0 @ 0 @

hence

@V0
D 0:
@

This means that V0 D V0 .x/ is constant with respect to .

Problem (P1). Since @V@ D 0, Eq. (3.7) becomes
0

     
@ @V1 @ @V0 @ @V1 @V0
 D  )  C D 0;
@ @ @ @x @ @ @x

implying that
 
@V1 @V0
 C D .x/
@ @x

is constant with respect to . Integrating the resulting equation

@V1 @V0 .x/

C D (3.10)
@ @x .x; /

over .0; 1/, we get

@ 1
V0 .x/ D .x/ ;
@x N
.x/
84 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

where we have introduced the harmonic average of  with respect to 

1
N
.x/ D R1 :
1
0  .x;/ d

Equation (3.10) becomes

 
@ N .x/ @
V1 .x; / D 1 V0 .x/:
@ .x; / @x

Setting

@
V1 .x; / D w.x; / V0 .x/;
@x
it follows
 
@V1 @w @V0 N @V0
D D 1 ;
@ @ @x  @x

and then

@w N
D  1:
@ 

Let us introduce now the cellular problem

8
ˆ @ N
.x/
ˆ
ˆ w.x; / D  1 for  2 .0; 1/
ˆ
< @ .x; /
w.x; 0/ D w.x; 1/
ˆ
ˆ Z
ˆ 1
:̂ w.x; / d  D 0;
0

whose solution is given by

Z  

N
.x/
w.x; / D w.x; 0/ C 1 ds:
0 .x; s/

In order to have
Z 1
w.x; / d  D 0;
0
3.1 Models of Cardiac Fibers 85

w.x; 0/ must satisfy

Z Z !
1 
1
w.x; 0/ D  N .x/ ds   d 
0 0 .x; s/
Z 1Z 
1 1
D N .x/ ds d  C :
0 0 .x; s/ 2

Therefore

@
V1 .x; / D w.x; / V0 .x/
@x
with
Z  Z 1 Z 
1 1 1
w.x; / D .x/
N ds    N .x/ ds d  C :
0 .x; s/ 0 0 .x; s/ 2

Problem (P2). To ensure the solvability, it must hold that

Z 1
0D S.x; / d 
Z0 1   
@ @ @
D im .x/  .x; / V0 .x/ C .x; / V1 .x; /
0  @x @x
 @
@ @
 .x; / V1 .x; / d
@ Z @x  
1
@ @ @
D im .x/  .x; / 1 C w.x; / V0 .x/ d 
Z 1 @x
 0 @
 @x
@ @
 .x; / V1 .x; / d :
0 @ @x

Since

@ N .x/
w.x; / D 1
@ .x; /

and the last term

Z 1  
@ @ @ @
.x; / V1 .x; / d  D .x; 1/ V1 .x; 1/.x; 0/ V1 .x; 0/ D 0
0 @ @x @x @x

because of the periodicity of  and V1 , we finally get the equation

 
d d
N
.x/ V0 .x/ D im .x/:
dx dx
86 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

We recall that
Z 1 Z 1 Z 1
V .x; / d  D V0 .x; / d  D V0 .x/ d  D V0 .x/;
0 0 0

hence V0 is denoted as the average potential.

In order to give a detailed approximation of the microscopic scale, one can
consider the first order correction
 x  x d
v .x/ D V x; D V0 .x/ C w x; V0 .x/ C O. 2 /:
  dx

3.1.3 Traveling Waves

A traveling wave is a solution of a PDE on an infinite domain that travels at constant

velocity with fixed shape. Traveling waves are mainly divided into two types. The
first is a traveling front, where the wave consists of a transition between two different
levels: if v denotes the wave variable, then in front of the wave, v is steady at some
low value, and behind the wave, v is steady at a higher value. The second type of
traveling wave is a traveling pulse, where the wave begins and ends at the same value
of v and resembles a moving bump. Thus, the main difference between traveling
fronts and traveling pulses is that the former do not present recovery, while in the
latter recovery plays an important dynamic role; see e.g. [63, 273].

3.1.3.1 Traveling Fronts

The cable equation (3.1) has been extensively studied in the special version known
as the bistable equation

@t v D @xx v C f .v/; (3.11)

where
• f .v/ has three zeros at v D 0; ˛; 1, with 0 < ˛ < 1;
• f 0 .0/ < 0 and f 0 .1/ < 0.
The values v D 0 and v D 1 are stable steady solutions of the ODE @t v D f .v/. For
existence, uniqueness and stability of traveling wavefront solutions and long time
behavior of the solution of (3.11), see [173–175].
An example of f often used in this context is the cubic polynomial

f .v/ D av.v  1/.˛  v/; 0 < ˛ < 1:

3.1 Models of Cardiac Fibers 87

A traveling wave is a translation-invariant solution of (3.11), providing a

transition between the two stable rest states (zeros of f .v/) and it travels with
a constant speed c, i.e. a solution of the form

v.x; t/ D V .x C ct/ D V ./; (3.12)

for a suitable value of c. When written as a function of the new traveling wave
variable , the wave appears stationary. Because we use  D x C ct, a solution with
a positive c corresponds to a wave moving from right to left, while a solution with
a negative c moves from left to right.
By imposing that v.x; t/ given in (3.12) satisfies the bistable equation (3.11), one
obtains that any traveling wave solution solves the following second order ODE

V   cV C f .V / D 0: (3.13)

If V ./ provides a transition between rest values, it must hold that

lim V ./ D 0 and lim V ./ D 1; implying lim f .V .// D 0:

!1 !C1 !˙1

From (3.13), it follows that a traveling wave solution of the bistable equation is a
solution of the first order ODE system

V D W
W D cW  f .V /;

connecting the steady states .V; W / D .0; 0/ and .V; W / D .1; 0/ in the .V; W /
phase plane. Such a trajectory, connecting two different steady states, is called a
heteroclinic trajectory and it approaches .0; 0/ as  ! 1 and .1; 0/ as  ! C1.
Beyond the steady states .0; 0/ and .1; 0/, which are saddle points, the other steady
state is .˛; 0/, which is a node, if c is positive, or a spiral point, if c is negative.
Because .0; 0/ and .1; 0/ are saddle points, in order to find a traveling wave solution,
we determine whether a parameter c exists such that the trajectory leaving .0; 0/ at
 D 1 reaches .1; 0/ at  D C1.
We first determine the sign of the velocity c. Supposing a monotone increasing
(V > 0) connecting trajectory exists, multiplying (3.13) by V and integrating from
 D 1 to  D C1, one obtains the following relationship
Z C1 Z 1
c W ./2 d  D f .u/du: (3.14)
1 0

Thus, if the integral of f .u/ between u D 0 and u D 1 is positive, the traveling

waves move from right to left and the state variable V moves from V D 0 to V D 1,
while if the integral is negative viceversa.
88 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

R1
Theorem 3.1. Suppose 0 f .u/du > 0. Then there exists a unique velocity c > 0
with a corresponding traveling wave solution of the bistable equation.
Proof. To verify uniqueness, from

dW f .V /
Dc ;
dV W

we observe that the slope dW dV of trajectories in the .V; W / phase plane is a monotone
increasing function of the parameter c. Let c D c0 be a velocity value for which
a connecting trajectory exists. Suppose now that for a value c1 > c0 , another
connecting trajectory exists. The trajectory leaving the saddle point .0; 0/ for c1
must lie above the connecting trajectory for c0 . For the same reason, the trajectory
approaching the saddle point .1; 0/ for c1 must lie below the connecting trajectory
for c0 . A single trajectory cannot lie above and below another one, thus there cannot
be a connecting trajectory for c > c0 . Analogously, there cannot be a connecting
trajectory for c < c0 . Hence the connecting trajectory is unique.
To verify existence, we first suppose c large. Let K be the smallest positive
number for which f .u/ u  K; 8u W 0 < u  1 and let  be any fixed positive
number. On the line W D V the slope of trajectories satisfies

dW f .V / f .V / K
Dc Dc c :
dV W V 

With c large enough, it is assured that c  K > . Therefore once trajectories

are above the line W D V , they stay above it. Because the slope of the unstable
trajectory leaving the saddle point .V; W / D .0; 0/ is the positive root of 2  c C
f 0 .0/ D 0, which is always larger than c, this trajectory starts above the curve
W D V . Thus is stays always above W D V and consequently passes above the
saddle point .V; W / D .1; 0/.
With c D 0, by multiplying (3.13) by V and integrating, one obtains that the
trajectory leaving .V; W / D .0; 0/ satisfies
Z V
W2
C f .u/du D 0;
2 0

and if it reaches V D 1 for some value of W , then

Z 1
W2
C f .u/du D 0;
2 0

R1
which can occur only if 0 f .u/du  0. Since this is in contradiction with the
R1
hypothesis that 0 f .u/du > 0, it follows that the trajectory cannot reach V D 1 and
neither it can remain in the first quadrant, because W > 0 implies that V increases.
3.1 Models of Cardiac Fibers 89

As a consequence, this trajectory must intersect W D 0 for some V < 1, thus it

cannot be the connecting trajectory.
Summarizing, we have found two trajectories, one with c large, which misses the
saddle point .V; W / D .1; 0/ staying above it, and one with c D 0, which intersects
the W D 0 axis for some V < 1. Because trajectories depend continuously on the
parameters of the problem, there is a family of trajectories depending continuously
on c between these two special trajectories, and therefore there must be at least one
trajectory reaching .V; W / D .1; 0/.
Example. Suppose f .v/ is the cubic polynomial

f .v/ D v.v  ˛/.1  v/:

Our goal is to find a heteroclinic connection between 0 and 1, so we guess that

W D BV.1  V /: (3.15)

By substituting (3.15) into (3.13), it follows that it must hold

.2B 2 C 1/V C .B 2  cB  ˛/ D 0;

which is identically zero only if we choose

1 1
BDp c D p .1  2˛/;
2 2

showing that the velocity is a decreasing function of ˛ and the direction of

R1
propagation changes at ˛ D 1=2, according to the change of sign of 0 f .u/du.
From the solution of (3.15), we obtain the profile of the traveling wave
 
1 x
V ./ D 1 C tanh p :
2 2 2

3.1.3.2 Traveling Pulses

A traveling pulse is a traveling wave solution that starts and ends at the same
steady state of the governing equations. In the previous section, we have seen that
a traveling front solution of the bistable equation corresponds to a heteroclinic
trajectory in the .V; W / phase plane, i.e. a trajectory that connects two different
steady states of the system. Analogously, a traveling pulse corresponds to a
trajectory that starts and ends at the same steady state in the .V; W / phase plane.
Such trajectories are called homoclinic orbits.
90 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

In order to study wave propagation in excitable media, we first consider the

following FitzHugh-Nagumo system

@t v D  2 @xx v C f .v; w/

(3.16)
@t w D g.v; w/;

where  is a small positive number and, without any loss of generality, the space
variable x has been scaled so that the diffusion coefficient is  2 , see e.g. [173–
175, 532]. This type of reaction-diffusion singular perturbation problem develops in
a short time a sharp propagating interface, i.e. a steep wavefront. In fact, in [85, 86],
it was shown that, after a time period of order O. log /, a sharp transition layer
of thickness of order O./ appears. Let us recall that the variable v represents the
transmembrane potential, while w represents a slow gating variable.
We now proceed further in understanding the structure of traveling wave
solutions of (3.16), by performing a singular perturbation analysis, see e.g. [79].
Starting from smooth initial data .v0 .x/; w0 .x//, the diffusion term @xx v and the
variation of w from its initial data can be neglected for a short time. In fact, setting
D t , vO .x; / D v.x; t/ and w.x;
O / D w.x; t/, then

@t vO D  2 @xx vO C f .Ov; w/
O
(3.17)
@t wO D g.Ov; w/: O

An approximation of order  is obtained by setting  to zero in (3.17), i.e.

@t vO D f .Ov; w/
O
@t wO D 0:

O
Thus w.x/ D w0 .x/ and, using the notations introduced in Sect. 2.9.8 (see also
Fig. 2.6), in a short time (t D O./) vO ! h˙ .w0 /, if v0 .x/ ? h0 .w0 /. The region of
space where v D hC .w/ is called excited region, while the region where v D h .w/
is called recovered region. In these two regions, it holds

O D0
f .Ov; w/
(3.18)
@t wO D g.Ov; w/;
O

thus
vO D h˙ .w/
O
O D G˙ .w/
@t w O D g.h˙ .w/;
O w/:
O

However, there are regions of space, called interfaces, where diffusion is large and
(3.18) cannot be applied. To understand what happens when diffusion is large, we
rescale space and time by setting D t and  D xy.t 
/
, where y.t/ denotes the
position of the wavefront, i.e. the points

y.t/ W v.y.t/; t/ D h0 .w0 .x//:

3.1 Models of Cardiac Fibers 91

The original system of equations (3.16) becomes

vO   C y 0 . /Ov C f .Ov; w/
O D @ vO
y 0 . /wO  D .g.Ov; w/
O  @ w/;
O

and, after setting  D 0, it reduces to

vO   C y 0 . /Ov C f .Ov; w/
O D0
0 (3.19)
y . /wO  D 0:

It follows from (3.19) that wO is independent of , but not necessarily of , i.e. w.

O /D
O
w.y. /; /. Because (3.19) describes the transition layer between regions where
(3.18) holds, we require the matching condition

lim f .Ov.; /; w. // D 0;
!˙1

i.e.

vO .˙1; / D h .w.
O //:

Since y. / represents the position of the wavefront, y 0 . / is the wave velocity.

The first equation in (3.19) is a bistable equation, in the form (3.13), thus, for
O there exists c D c.w/
fixed w, O for which the equation

vO   C c.w/O
O v C f .Ov; w/
O D0 (3.20)

has a heteroclinic orbit connecting two stable roots of f .Ov; w/ O D 0. This means
O is the unique value for which (3.20) has a solution with vO ! hC .w/
that c.w/ O as
 ! 1 and vO ! h .w/ O as  ! C1.
Summarizing, in most of space, (3.18) holds. At any transition between the two
types of (3.18), a sharp transition in v occurs, with a wavefront traveling at the
speed y 0 .t/ D c.w/ if v D hC .w/ on the left and v D h .w/ on the right, or
y 0 .t/ D c.w/ if v D h .w/ on the left and v D hC .w/ on the right. We now
suppose that far to the right the medium is at rest, i.e. 9 wC W G .wC / D 0, and that
a wavefront of excitation moves from left to right with velocity y 0 .t/ D c.wC / > 0.
Following the same procedure used to derive (3.14), one obtains

R hC .w/
h .w/ f .v; w/dv
c.w/ D R C1 2 ;
1 v  d 
92 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

and thus c.wC / > 0 if and only if

Z hC .w/
f .v; w/dv > 0: (3.21)
h .w/

If (3.21) fails to hold, then the medium is not sufficiently excitable to sustain
a traveling pulse.

3.1.4 Conduction Velocity Restitution Curve

Conduction block is a key arrhythmogenic mechanism. It can result from a reduction

of inward membrane currents, a reduction of gap junctional coupling and from tissue
inhomogeneities. Conduction velocity (CV) of the propagating electric impulse in
the cardiac tissue depends, as the action potential duration (see Sect. 2.9.6), on
one or several previous diastolic or interbeat intervals. This dependence, called
restitution, is an important determinant of the stability of conduction and is
involved in the generation of arrhythmias. The CV restitution function describes
this dependence of the conduction velocity on the diastolic interval (DI). Figure 3.1

0.07

0.065
CV (cm/ms)

0.06

0.055

0.05
20 40 60 80 100 120 140 160 180

DI (ms)

Fig. 3.1 Conduction velocity (CV) restitution curve computed with the cable equation coupled
with the LRd model
3.2 Models of Cardiac Tissue 93

reports the CV restitution curve calculated using the cable equation coupled with
the LRd model, i.e. the following system of PDEs
8
ˆ @v @2 v
ˆ
ˆ cm   2 C iion .v; w; c/ D iapp ;
ˆ
ˆ
< @t @x
dw
ˆ  R.v; w/ D 0;
ˆ
ˆ dt
ˆ dc
:̂  S.v; w; c/ D 0;
dt

with cm D Cm , iion D Iion ,  D 1:2  103 1 cm1 and the functions Iion , S , R
given by the LRd model.

3.2 Models of Cardiac Tissue

The cardiac tissue is composed of a collection of elongated cardiac cells having

roughly a cylindrical form with a diameter dc  10 m and length lc  100 m.
The cells are coupled together mainly in end-to-end and also in side–to–side
apposition by gap junctions, [237, 448]. These specialized membrane regions of
densely packed channels provide direct intercellular communication between the
cytoplasmatic compartments of two adjacent cells; they are large at the longitudinal
cell ends and small along the lateral borders. The end-to-end contacts form the
long fiber structure of the cardiac muscle whereas the presence of lateral junctions
establishes a connection between the elongated fibers.
At a cellular level the structure of the cardiac tissue can be viewed as composed
by two ohmic conducting media: the intracellular space ˝i (inside the cells)
and the extracellular space ˝e (outside) separated by the active membrane m .
Due to the presence of gap junctions connecting the cardiac cells end-to-end and
side-to-side, ˝i and ˝e are regarded as two simply-connected open sets of R3 .
The effects of the microstructure on the current flow are also included in the
conductivity tensors ˙i .x/; ˙e .x/ as inhomogeneous functions of space that reflect
the local variations of conductances because of the presence of structural intra
and extracellular inhomogeneities of resistance associated with e.g. gap junctions,
connective tissue, collagen and blood vessels. Let ui ; ue be the intra and extracellular
potentials and Ji:e D ˙i;e rui;e their current densities. Let ni ; ne denote the unit
exterior normals to the boundary of ˝i and ˝e respectively, satisfying ni D ne
on m . Under the quasi-static assumption (see [394] and [215, Ch. 5.3] for the
validity of this condition for biological fields), due to the current conservation law,
the normal current flux through the membrane is continuous, thus Ji  ni D Je  ni ,
i.e in terms of potentials

nTi ˙i rui C nTe ˙e rue D 0; on m :

94 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

On the other hand, since the only active source elements lie on the membrane
m , each flux equals the membrane current per unit area Im , which consists of a
capacitive and an ionic term (see [250, 317]):

@v
 nTi ˙i rui D nTe ˙e rue D Im D Cm C Iion : (3.22)
@t
Here Cm is the surface capacitance of the membrane per unit area and v WD ui j 
m
ue j is the transmembrane potential evidencing that m is a discontinuity surface
m
for the potential (in the following, we will simply write v D ui  ue ).
Denoting by Iis ; Ies the stimulation currents applied to the intra and extracellular
space, we have

 div.˙i rui / D Iis ; in ˝i ;  div.˙e rue / D Ies ; in ˝e . (3.23)

Assuming ˝H WD ˝i [ ˝e [ m embedded into an insulating medium, then

homogeneous Neumann boundary conditions for ui ; ue on the remaining part of
the boundaries i;e D @˝i;e n m must be assigned

nTi;e ˙i;e rui;e D 0:

Finally, the system (3.22), (3.23) must be supplemented by the initial conditions

v.; 0/ D v0 ; w.; 0/ D w0 ; on m :

For the electric potentials ui ; ue we can consider two characteristic length scales:
a micro scale related to typical cell dimensions fdc ; lc g and a macro scale
determined by a suitable length constant of the tissue. At the latter scale length,
i.e. at a macroscopic level, in spite of the discrete cellular structure, the cardiac
tissue can be represented by a continuous model. To identify this macroscopic
scale, following [348], we consider a suitable non-dimensional form of the cellular
mathematical model.
The cellular conductivity matrices ˙i .x/ and ˙e .x/ are symmetric positive
definite matrices; setting N D N i C N e , with N i , N e the average eigenvalues on
a cell element, we consider the dimensionless conductivity matrices

i;e D ˙i;e =:

N

Let Rm be an estimate of the passive membrane resistance near the equilibrium

point vr , i.e. the resting transmembrane potential. Multiplying by Rm =N both sides
of Eq. (3.22), we obtain

Cm Rm @v Rm
 i Rm nTi rui D C Iion : (3.24)
N @t N
3.2 Models of Cardiac Tissue 95

p introduce the membrane time constant m D Rm Cm , the length scale unit  D

We
N m and we consider the following scaling of the space and time variables
lc R

xO D x=; tO D t= m:

Disregarding the presence of applied current terms, rescaling Eqs. (3.23) and (3.24)
in the intra and extracellular media, we obtain:

divxO .i rxO ui / D 0 in ˝i divxO .e rxO ue / D 0 in ˝e ;

@v
nTi i rxO ui D nTe e rxO ue D " C Iion .v; w; c/ on m ;
@tO

where the dimensionless parameter is the ratio " D lc = between the micro and the
macro length constants.
The two-scale method of homogenization can be applied to the previous current
conservation equations. The microscopic space variable measured in unit cell
is defined by  WD x= lc ; then with respect to the dimensionless macroscopic
coordinates, the micro and macro scales are related to each other by the scaling
parameter "

 WD xO ="

For convenience, we will omit in the following the superscripts O of the dimension-
less variables.
Following the standard approach of homogenization theory, we are assuming
that the cells are distributed according to an ideal periodic organization similar to
a regular lattice of interconnected cylinders. Due to the longitudinal and transverse
intercellular interconnections, in the modeled periodic cellular aggregate the intra
and extracellular media are connected regions. If fe 1 ; e 2 ; e 3 g is an orthogonal basis
of R3 , we denote by

Ei ; Ee WD R3 n E i ; with common boundary m WD @Ei \ @Ee ;

two reference open, connected and periodic subsets of R3 with Lipschitz boundary,
i.e. satisfying

Ei;e C e k D Ei;e ; k D 1; 2; 3:

The elementary periodicity region

nX
3 o
Y WD ˛k e k W 0  ˛k < 1; k D 1; 2; 3
kD1
96 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

is composed of the intra and extracellular volumes Yi;e D Y \ Ei;e and represents
a reference volume box containing a single cell Yi with cell membrane surface Sm D
m \ Yi .
The main geometrical assumption is that the physical intra or extracellular regions
are the "-dilation of the reference lattices Ei;e , defined as
˚ ˚
"Ei;e D " W  2 Ei;e ; with "m WD " W  2 m :

Therefore, the decomposition of the physical region ˝H , occupied by the cardiac

"
tissue, into the intra and extracellular domains ˝i;e (see e.g. Fig. 3.2) can be obtained
simply by intersecting ˝H with "Ei;e , i.e.

˝i" D ˝H \ "Ei ; ˝e" D ˝H \ "Ee ; m" D @˝i" \ @˝e" D ˝H \ "m :

The common boundary m" models the cellular membrane.

Y
Ωeε
Γmε
Yi Ωiε
Γmε
Ye Ωeε Ωeε

Sm Γmε
ξ= x Ωiε Ωiε ε
ε

ΩH ε
Fig. 3.2 Right: The ideal periodic geometry in a bidimensional section of the simplified 3–D
periodic network of interconnected cells. Left: Unit cell in the microscopic variable  D x="
3.2 Models of Cardiac Tissue 97

Since the cardiac tissue exhibits a number of significant inhomogeneities, such as

those related to cell-to-cell communications, the conductivity tensors are considered
dependent on both the slow and fast variables, i.e. i;e .x; x" /. The dependence of
i on x" models the inclusion of gap-junction effects. We then define the rescaled
symmetric conductivity matrices
x
"
i;e .x/ D i;e .x; /;
"
where i;e .x; / W ˝H  Ei;e ! M33 are continuous functions satisfying the usual
uniform ellipticity and periodicity conditions
)
jyj2  i;e .x; /y  y   1 jyj2
8 .x; / 2 ˝H  Ei;e ; y 2 R3 ;
i;e .x;  C e k / D i;e .x; /

for a given constant  > 0.

3.2.1 The Dimensionless Cellular Model P "

Summarizing, we formulate the full reaction-diffusion system associated with the

cellular model P" as follows:

Let ˝H WD ˝i" [ ˝e" [ m" be the cardiac tissue volume;

˝i" WD the intracellular space with dimensionless conductivity WD i" ;
˝e" WD the extracellular space with intracellular conductivity WD e" ;
ni ; ne WD the exterior unit normals to @˝i" ; @˝e" ;
m" WD the surface cellular membrane;
n D ni D ne the normal to m" pointing toward ˝e" :

Then the vector .u"i ; u"e ; w" ; c " /, with v" D u"i  u"e satisfies the problem:
(
 div i" .x/ru"i D 0 in ˝i"
(3.25)
 div e" .x/ru"e D 0 in ˝e"
8
ˆ @v"
ˆ
ˆi" .x/ nT ru"i D " Œ C Iion .v" ; w" ; c " / 
< @t
Im" D (3.26)
ˆ
ˆ "
:̂ " .x/ nT ru" D " Œ @v C I .v" ; w" ; c " / 
e e ion
@t
@w" @c "
 R.v" ; w" / D 0 on m" ;  S.v" ; w" ; c " / D 0;
@t @t
98 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

supplemented by the following boundary conditions of Neumann type (assuming

for instance that the cellular aggregate is embedded in an insulated medium)

nT ru"i D 0 on @˝i" =m" and nT ru"e D 0 on @˝e" =m" ;

and the following degenerate initial conditions on v" ; w" ; c "

v" .x; 0/ D v"0 .x/; w" .x; 0/ D w"0 .x/; c " .x; 0/ D c0" .x/ su m" :

The variables v" ; w" ; c " and Im" are defined only on the surface of the cellular
membrane m" .

3.2.2 Formal Two-Scale Homogenization

We briefly indicate how to use the two–scale method (see [23, 44, 260, 362, 454])
and formal asymptotic expansions to convert the microscopic model of the periodic
cellular aggregate into an averaged continuum representation of the cardiac tissue,
neglecting the presence of stimulation currents. We seek a solution .u"i ; u"e ; w" ; c " /
where each component has an asymptotic form in powers of " of the following type:

u D u0 .x; ; t/ C "u1 .x; ; t/ C "2 u2 .x; ; t/ C   

with coefficients uk Y –periodic functions of . Considering the full derivative

operators

ru D "1 r u C rx u;
div ru D "2 div r u C "1 div rx u C "1 divx r u C divx rx u;

substituting the asymptotic forms into the first equations of (3.25, 3.26) and equating
the coefficients of the powers 1; 0; 1; of " to zero, we obtain the following
equations for the functions uk .x; ; t/; k D 0; 1; 2 associated with u D u"i :
8
ˆ
ˆ
< Find uk Y –periodic in  such that:
 div i .x; /r uk D fk .x; /  div i Fk .x; / in Ei (3.27)
ˆ
:̂nT  .x; /r u D g .x; / C nT  F .x; / on m ;
 i  k k  i k

with
(
f0 D 0; F0 D 0 in Ei
(3.28)
g0 D 0 on m ;
3.2 Models of Cardiac Tissue 99

(
f1 D 2divx i r u0 ; F1 D 0 in Ei
(3.29)
g1 D nT i rx u0 on m ;

8
ˆ
ˆ f D divx i r u1 C divx i rx u0 ; F2 D i rx u1 ; in Ei
< 2
@v0
g D . C Iion .v0 ; w0 ; c0 //; on m
ˆ 2 @t
:̂
v0 D ui0  ue0 ; @t w0  R.v0 ; w0 / D 0; @t c0  S.v0 ; w0 ; c0 / D 0:
(3.30)
In problem (3.27), the variable x appears as a parameter. Let fk .x; /, Fk .x; /,
gk .x; / be Y –periodic functions in . Then the problems for k D 0; 1; 2 admit
a unique solution uk , apart from an additive constant (consequence of an easy
extension of the result in [23] Th. 2 or [362] Th. 6.1) if and only if
Z Z
fk d  C gk ds D 0; k D 0; 1; 2: (3.31)
Yi Sm

From the first cellular problem (3.28) for k D 0, it follows that the Y –periodic
solution u0 is independent of  and u0 .x/, depending only on the macroscopic vari-
R terms uk .x; ; t/ are
able x, represents a potential average over Yi if the subsequent
determined with zero mean value on Yi . Since f1 D 0 and Sm nT i rx u0 ds D 0
the solvability of problem (3.28) related to the data (3.29) R is assured and it is
easy to check that the solution with zero mean on Yi , i.e. Yi u1 d  D 0, can be
represented as

u1 .x; ; t/ D w.x; /T rx u0 ; (3.32)

where w.x; / D .w1 ; w2 ; w3 /T is the unique zero mean value solution on Yi

satisfying
(
div i .x; /r wk D 0; in Yi ;
(3.33)
nT i .x; /r wk D nk on Sm ; k D 1; 2; 3:

Then problem (3.27) related to the data (3.30) becomes

8
ˆ
ˆ  div i r u2 D  div i rx .wT rx u0 /
<
C divx i rx u0  divx i r .wT rx u0 / in Yi
ˆ
:̂nT  r u D . @v0 C I .v ; w // C nT  .wT r u / on Sm :
 i  2 ion 0 0  i x 0
@t
100 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

In order to assure its solvability, the following compatibility relation must be

satisfied
Z Z Z
@v0
 divx i r .w rx u0 / d C divx i rx u0 d  .
T
CIion .v0 ; w0 // ds D0:
Yi Yi Sm @t

Considering that u0 is independent of  and using (3.32) it follows that

Z
˚ @v0
divx i .x; / I  r w.x; /T d  rx u0 D jS j. C Iion .v0 ; w0 //;
Yi @t

where I is the identity matrix, r wT D Œr w1 ; r w2 ; r w3  and jYi j, jSm j denote

the volume and the area of Yi and Sm , respectively.
Let ˇ D jSm j=jY j be the ratio between the surface membrane and the volume
of the reference cell and ˇi;e D jYi;e j=jY j. With reference to medium (i) u0 D ui0 ,
w D wi we set
Z
1 ˚
Di .x/ D i .x; / I  r .wi /T d :
jY j Yi
Hence, we obtain the following “averaged equation” for the intracellular potential
 
@v0
div Di .x/rx u0 D ˇ
i
C Iion .v0 ; w0 / :
@t
Following [44], it is easy to verify that the macroscopic conductivity tensor of the
intracellular Di is symmetric and positive definite.
Proceeding similarly for u D ue and taking into account that the unit normal n
points inside ˝e , we obtain the following averaged equations for the extracellular
potential
 
@v0
div De .x/rx ue0 D ˇ C Iion .v0 ; w0 / :
@t

3.2.3 The Dimensionless Averaged Model P

Summarizing, for a periodic network of interconnected cells, the governing dimen-

sionless equations of the macroscopic intra and extracellular potentials at zero order
in " are given by
8  
ˆ @v
ˆ
ˆ div Di .x/rx ui D ˇ C Iion .v; w/
ˆ
ˆ @t
ˆ
ˆ  
<
@v
div De .x/rx ue D ˇ C Iion .v; w/ (3.34)
ˆ
ˆ @t
ˆ
ˆ
ˆ
ˆ
:̂ v D u  u ; @w  R.v; w/ D 0; @c  S.v; w; c/ D 0:
i e
@t @t
3.2 Models of Cardiac Tissue 101

Here the effective conductivity tensors are given by

Z
1 ˚
Di;e .x/ D i;e .x; / I  r .wi;e /T d 
jY j Yi;e

and wi;e D .wi;e i;e i;e T

1 ; w2 ; w3 / are solutions of

(
div i;e .x; /r wek D 0; in Yi;e
nT i;e .x; /r wi;e
k D nk ; on S; k D 1; 2; 3:

The previous derivation based on the two-scale method is only formal, but the
averaged model can be rigorously justified in the framework of  convergence
theory as a limit problem of the cellular model for " ! 0, see [4].

3.2.4 Theoretical Results for the Cellular and Averaged Models

We introduce the functional space

˚
V" D H 1 .˝i" /  H 1 .˝e" / = f. ; / W 2 Rg  L2 .m" /M  L2 .m" /R ; (3.35)

the vector variables U WD .ui ; ue ; w; c/ 2 V" ; UO WD .Oui ; uO e ; w;

O c/
O 2 V" , the vector
@ui @ue @w @c
time derivative @t U WD . ; ; ; / and we set v D ui  ue , vO D uO i  uO e .
@t @t @t @t
Then we introduce the forms
Z
b " .U ; UO / WD " Πv vO C w wO C c cO  d ;
m"
XZ
a" .U ; UO / WD "
i;e rui;e r uO i;e dx;
"
i;e ˝i;e
Z
F .U ; UO / WD "
"
ŒIion .v; w; c/ vO  R.v; w/ wO  S.v; w; c/ cO  d ;
m"

and we consider the variational formulation of the differential problem P" :

Find U " W Œ0; T  ! V" W b " .@t U " ; UO /Ca" .U " ; UO /CF " .U " ; UO / D 0 8 UO 2 V" ;
(3.36)
where the parabolic b " and elliptic a" forms are degenerate, but their sum is coercive
on V" . The variational formulation is supplemented by the initial conditions:

v" .; 0/ D v"0 ; w" .; 0/ D w"0 ; c " .; 0/ D c0" ; on m " ;
102 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

We consider the functional space

˚
V WD H 1 .˝H /  H 1 .˝H / = f. ; / W 2 Rg  L2 .˝H /M  L2 .˝H /R

and we introduce the following bilinear and nonlinear forms associated with the
averaged model (3.34): for U D .ui ; ue ; w; c/ 2 V; UO D .Oui ; uO e ; w;
O c/
O 2 V, and
v D ui  ue ; vO D uO i  uO e , we set
Z
O
b.U ; U / WD ˇ Œ@t v vO C @t w wO C @t c cO  dx;
˝H
XZ
a.U ; UO / WD Di;e rui;e rui;e dx
i;e ˝i;e
Z
F .U ; UO / WD ˇ ŒIion .v; w; c/ vO  R.v; w/ w
O  S.v; w; c/ cO  dx:
˝H

The homogenized conductivity tensors Di;e can be characterized by solving the

following variational cellular problems for every y 2 R3 :
n 1 Z  T  
y T Di;e .x/ y WD min ru./ C y i;e .x; / ru./ C y d  W
jY j Yi;e
o
u 2 Hloc
1
.Rd /; uY -periodic :

These tensors are symmetric and positive definite matrices, and the bilinear form
a.; / is coercive on V.
The variational formulation of the averaged problem P, related to the macro-
scopic model (3.34), is given by:

Find U W Œ0; T  ! V W b.@t U ; UO / C a.U ; UO / C F .U ; UO / D 0 8 UO 2 V;

(3.37)
supplemented with the initial conditions

v.; 0/ D v0 ; w.; 0/ D w0 ; c.; 0/ D c0 ; in ˝H :

We now focus on the FitzHugh-Nagumo membrane model [178, 179]: the ionic
current is a cubic like function in v and is linear in the recovery variable w. In this
simplified model, the unknown is the vector .u"i ; u"e ; w" / and it was shown in [105]
that both the cellular and the averaged models share the same variational structure
and yield well posed problems. More precisely, introducing the quotient space as
in (3.35)
˚
V" D H 1 .˝i" /  H 1 .˝e" / = f. ; / W 2 Rg  L2 .m" /; (3.38)

we have the following result for problem P" .

3.2 Models of Cardiac Tissue 103

Theorem 3.2. Assuming  " is regular and supposing that the initial data satisfy

.v"0 ; w"0 / 2 L2 .m" /  L2 .m" /;

then there exists a unique solution U " D .u"i ; u"e ; w" / 2 C 0 .0; T I V" / of the
variational formulation (3.36) of Problem P" with

@v" @w"
; 2 L2 .0; T I L2 .m" //I
@t @t

u" WD .u" ; u" / solves the differential equations P " in the standard distributional
¯sense. i e

Introducing the quotient space

˚
V WD H 1 .˝H /  H 1 .˝H / = f. ; / W 2 Rg  L2 .˝H /

we have the following result for problem P.

Theorem 3.3. Assuming that the initial data satisfy

.v0 ; w0 / 2 L2 .˝H /  L2 .˝H /;

then there exists a unique solution U D .ui ; ue ; w/ 2 C 0 .0; T I V/ of the

variational formulation (3.37) of the averaged Problem P with

@v @w
; 2 L2 .0; T I L2 .˝H //I
@t @t

u WD .ui ; ue / solves the differential equations P in the standard distributional sense.

¯
We remark that the above abstract variational framework of the cellular (3.36) and
averaged (3.37) models in terms of the forms a" ; b " ; F " and a; b; F respectively,
share the same structural properties, see [105].
Well-posedness results for the cellular and the averaged models with ionic current
membrane dynamics, described by the classical Hodgkin-Huxley model [229] or by
the Luo-Rudy Phase I model [308], can be found in [540, 541].

3.2.5  -Convergence Result for the Averaged Model

with FHN Dynamics

We now present a convergence result for the homogenization process related to the
Bidomain model with Nagumo membrane model (i.e. FHN without the recovery
variable); see [378] for details of the general FHN case.
104 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

The problem P " is not a standard parabolic homogenization problem and its
main difficulties are associated with the fact that b " depends explicitly on ", it is
degenerate and the boundaries of ˝i;e"
could be quite irregular. For z" 2 L2 . " /,
u D .ui ; ue / 2 H .˝i /  H .˝e /, and z 2 L2 .˝H /, u D .ui ; ue / 2 .H 1 .˝H //2 ,
" " " 1 " 1 "
¯ define the energy-like functionals
we ¯

Z XZ
b " .z" / WD " jz" j2 d ; a" .u" / WD "
i;e ru"i;e ru"i;e dx;
m" ¯ i;e
"
˝i;e
Z XZ
b.z/ WD ˇ jzj2 dx; a.u/ WD Di;e rui;e rui;e dx;
˝H ¯ i;e ˝H

Z Z
G .z / WD "
" " "
G.z /d ; G .z/ WD ˇ G.z/dx; (3.39)
m" ˝H

where G is a positive primitive of g in the FHN model.

Theorem 3.4. Let us assume that v"0 D u"i;0  u"e;0 ; w"0 are converging to v0 D
ui;0  ue;0 ; w0 in the following “distributional” sense, i.e.
Z Z
lim " v"0 .x/.x/ d D ˇ v0 .x/.x/ dx; 8 2 C01 .˝H /
"#0 m" ˝H
Z Z
lim " w"0 .x/.x/ d D ˇ w0 .x/.x/ dx; 8 2 C01 .˝H /
"#0 m" ˝H
Z Z
lim u"i;e .x/.x/ dx D ˇi;e ui;e .x/.x/ dx; 8 2 C01 .˝H /
"#0 "
˝i;e ˝H

and the related energies satisfy

 
lim b " .v"0 / D b.v0 /; lim b " .w"0 / D b.w0 /; lim sup a" .u"0 / C G " .v"0 / < C1:
"#0 "#0 "#0

Let ˝0 ˝H be a reference open subdomain with positive measure, u" D

.ui ; ue /; v D u"i  u"e ; w" and u D .uRi ; ue /; v D ui  ue ; w be
" " "
R the solutions of
¯ with D ˝0 \˝H ue dx D 0,
"
the cellular and averaged models ˝0 \˝ " u e dx 0; and
respectively. Then for every time t 2 Œ0; T 

.u"i;e ; v" ; w" / ! .ui;e ; v; w/ as " # 0; in the distributional sense,

3.2 Models of Cardiac Tissue 105

with

a" .u" / ! a.u/; b " .v" / ! b.v/; b " .w" / ! b.w/:

Moreover, there exist extensions Ti " u"i ; Te" u"e of u"i ; u"e in the whole domain ˝H ,
solution of the cellular problem P" , which converge in L2 .0; T I Hloc 1
.˝H // to the
unique solution .ui ; ue / 2 V of the averaged model P.
The variational approach for the convergence of the evolution problem is based
on the introduction of the time-semidiscrete approximation by the implicit Euler
method, which reduces the evolution system to discrete families of stationary
problems. More precisely, given U "0 , we introduce the sequence of variational
problems:
Find U ";n 2 V; n D 1; : : : ; N such that

U ";n  U ";n1
b". ; UO / C a" .U ";n ; UO / C F " .U ";n ; UO / D 0; 8UO 2 V:

For D T =N sufficiently small, the coercivity of a C b and the one-sided Lipschitz

condition on F " guarantee the recursive solvability of these equations.
The previous Theorem follows by combining  convergence and uniform
error estimates for the Euler discretization; assuming, for simplicity, an instanta-
neous ionic current without recovery, i.e. Iion .v/ D g.v/, the discrete solution
U ";1 ; : : : U ";N of the Euler scheme solves the iterated (convex) minimization
problem

n1 o
U ";n D arg min b " .V  U ";n1 / C ˚ " .V / ; (3.40)
V 2V 2

where ˚ " is the Lyapunov functional

˚ " .U " / WDa" .u" / C G " .v" /

¯
and a" ; b " ; G " have been defined in (3.39). Analogously, we can define the same
quantities for the limit case " D 0, obtaining a discrete solution U ;n that solves
an iterated convex minimization problem (3.40) without " involving the Lyapunov
functional

˚.U / WDa.u/ C G .v/;

¯
with a; b; G again defined in (3.39).
106 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

The general strategy for passing to the limit (see [378]) can be summarized in the
following diagram
Stationary Problem
Evolution Problem → U ετ n+1 :
Pε : U ε minW 2τ b (W − Uτε n) + Φ ε (W )
1 ε

ε ↓0
Limit of the Evolution Limit of the Stationary Problem
0←τ
Problem P : U U τn+1 : minW 21τ b(W − Uτn) + Φ (W )

3.3 The Macroscopic Anisotropic Bidomain Model

The macroscopic Bidomain representation of the cardiac tissue (3.34) has been
derived in the previous chapter using homogenization techniques.
We now present a heuristic derivation of the Bidomain model, based on the notion
of the interpenetrating domains introduced in [463], formalized by [199, 329, 531],
see also [225, 391]. We denote by ˝H the cardiac tissue volume that we assume
represented as the superposition of two anisotropic continuous media, the intra- (i)
and extra- (e) cellular media, coexisting at every point of the tissue and separated by
a distributed continuous cellular membrane.
Given a point x 2 ˝H , we denote by Ji .x; t/; Je .x; t/ the local average
current densities per unit area in the intra- and extracellular media and by im
the transmembrane current per unit volume flowing across the membrane surface.
Consider a volume V surrounding x and denote by jV j its volume, @V its smooth
surface, n the outward normal on @V . Then the current conservation law on the
volume V states that
Z Z Z
1 1 1
Je  nd D  Ji  nd D cm im  dx;
jV j @V jV j @V jV j V

i.e. the flux entering the extracellular volume must equal the flux exiting the
intra-cellular volume, and both must equal the transmembrane current across the
membrane (directed from the intra- to the extracellular media). Taking the limit as
jV j ! 0, it then follows

divJe .x; t/ D divJi .x; t/ D im :

We recall that the cardiac tissue consists of an arrangement of fibers that rotate
counterclockwise from epi- to endocardium, and that have a laminar organization
modeled as a set of muscle sheets running radially from epi- to endocardium.
3.3 The Macroscopic Anisotropic Bidomain Model 107

The anisotropy of the intra- and extracellular media, related to the macroscopic
arrangement of the cardiac myocytes in the fiber structure, is described by the
anisotropic conductivity tensors Di .x/ and De .x/, defined as

Di;e .x/ D li;e al .x/aTl .x/ C ti;e at .x/aTt .x/ C ni;e an .x/aTn .x/ D
D li;e I C .ti;e  li;e /at .x/aTt .x/ C .ni;e  li;e /an .x/aTn .x/:
(3.41)
Here al .x/; at .x/; an .x/, is a triplet of orthonormal principal axes with al .x/
parallel to the local fiber direction, at .x/ and an .x/ tangent and orthogonal to the
radial laminae, respectively, and both being transversal to the fiber axis (see e.g.
LeGrice et al. [292]). Moreover, li;e ; ti;e ; ni;e are the conductivity coefficients
in the intra- and extracellular media measured along the corresponding directions
al ; at ; an . For axisymmetric anisotropic media ni;e D ti;e . Since the anisotropic
intra- and extracellular media are ohmic, currents and potentials are related by

Ji D Di rui ; Je D De rue ;

and we obtain the following parabolic - parabolic (PP) formulation of the Bidomain
system.
PP-formulation. Given the applied intra- and extracellular currents per unit
s
volume Ii;e W ˝H  .0; T / ! R, and initial conditions v0 W ˝H ! R, w0 W
˝H ! R , z0 W ˝H ! .0; C1/m , find the intra- and extracellular potentials ui;e W
k

˝H  .0; T / ! R, the transmembrane potential v D ui  ue W ˝H  .0; T / ! R,

the gating variables w W ˝H  .0; T / ! Rk and the ionic concentrations variables
z W ˝H  .0; T / ! Rm such that
8
ˆ @v
ˆ
ˆ cm  div.Di rui / C Iion .v; w; z/ D Iis in ˝H  .0; T /
ˆ
ˆ @t
ˆ
ˆ
ˆ
ˆ @v
ˆ
ˆ cm  div.De rue /  Iion .v; w; z/ D Ies in ˝H  .0; T /
ˆ
ˆ
ˆ
ˆ @t
ˆ
< @w
 F.v; w/ D 0; in ˝H  .0; T /
ˆ @t
ˆ
ˆ
ˆ
ˆ @z
ˆ
ˆ  G.v; w; z/ D 0; in ˝H  .0; T /
ˆ
ˆ @t
ˆ
ˆ
ˆ nT D ru D 0
ˆ in @˝H  .0; T /
ˆ
ˆ i;e i;e
:̂
v.x; 0/ D v0 .x/; w.x; 0/ D w0 .x/; z.x; 0/ D z0 .x/; in ˝H ;
(3.42)
108 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

where we have assumed an insulated cardiac boundary @˝H . The nonlinear reaction
term Iion and the ODE system for the gating variables w and the ionic concentrations
c are given by the ionic membrane model.
PE-formulation. System (3.42) can be equivalently rewritten in terms of the
transmembrane and extracellular potentials v.x; t/ and ue .x; t/, thus obtaining the
parabolic-elliptic (PE) formulation of the Bidomain model
8
ˆ @v
ˆ
ˆ cm  div.Di rv/  div.Di rue / C Iion .v; w; z/ D Iis in ˝H  .0; T /
ˆ
ˆ @t
ˆ
ˆ
ˆ div.Di rv/  div..Di C De /rue / D Iis C Ies
ˆ in ˝H  .0; T /
ˆ
ˆ
ˆ
ˆ
ˆ
ˆ @w
ˆ
ˆ  F.v; w/ D 0; in ˝H  .0; T /
< @t
@z
ˆ
ˆ  G.v; w; z/ D 0; in ˝H  .0; T /
ˆ
ˆ @t
ˆ
ˆ
ˆ
ˆ nT Di r.v C ue / D 0 in @˝H  .0; T /
ˆ
ˆ
ˆ
ˆ
ˆ
ˆ nT .Di C De /rue C nT Di rv D 0; in @˝H  .0; T /
ˆ
ˆ
:̂
v.x; 0/ D v0 .x/; w.x; 0/ D w0 .x/; z.x; 0/ D z0 .x/; in ˝H :
(3.43)
We note that, in the previous models, we have not included many anatomical
components, such as:
• The specialized ventricular conduction system, composed of the bundle of His,
the left and right bundle branches, and an extensive Purkinje network, which
connects to the myocardium at discrete sites known as Purkinje-ventricular
junctions, see e.g. [22, 55, 425, 520];
• The presence of the coronary vasculature, see [51, 52, 490];
• The deformation of the cardiac tissue induced by the mechanical contraction and
relaxation process [232, 245, 246, 527].

3.4 Well-Posedness Results Based on Semi-discretization

in Time

Following [105], we first consider the well-posedness of the variational formulation

of the macroscopic PP Bidomain model (3.42) coupled with a Fitzhugh-Nagumo-
like membrane dynamics.
3.4 Well-Posedness Results Based on Semi-discretization in Time 109

Let us suppose that

˝H is a Lipschitz domain of R3 ; (3.44)

Di .x/ and De .x/ are measurable and satisfy the uniform ellipticity condition

9˛; m > 0 W ˛jj2  Di;e .x/    mjj2 ; 8 2 R3 ; x 2 ˝H ; (3.45)

and finally f is a continuous function with

f .x/  f .y/
f .0/ D 0I 9 f 0W  f ; 8x; y 2 R; with x ¤ y:
xy
(3.46)
We will also require that f has a cubic growth at infinity, i.e.

f .s/ f .s/
0 < lim inf 3
 lim sup 3 < C1: (3.47)
jsj!C1 s jsj!C1 s

Then consider the following Bidomain problem:

Problem (M). Given
s
Ii;e W ˝h  .0; T / ! R; and v0 ; w0 W ˝H ! R;

we seek

ui ; ue ; w W ˝H  .0; T / ! R; with v WD ui  ue ;

which solve
8
ˆ @v
ˆ
ˆ cm  div.Di rui / C f .v/ C w D Iis in ˝H  .0; T /
ˆ
ˆ @t
ˆ
ˆ
ˆ
ˆ @v
ˆ
ˆ cm  div.De rue /  f .v/  w D Ie
s
in ˝H  .0; T /
< @t
@w (3.48)
ˆ
ˆ C w  v D 0; in ˝H  .0; T /
ˆ
ˆ @t
ˆ
ˆ
ˆ
ˆ nT Di;e rui;e D 0 in @˝H  .0; T /
ˆ
ˆ
:̂
v.x; 0/ D v0 .x/; w.x; 0/ D w0 .x/; in ˝H ;

with  positive parameter.

110 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

Let us now introduce the Hilbert spaces

X WD H 1 .˝H /  H 1 .˝H /  L2 .˝H /; V WD X=f.c; c; 0/ W c 2 Rg; (3.49)

whose generic element .ui ; ue ; w/ we denote by u, the bilinear forms on V  V

Z

O WD
b.u; u/ Œcm .ui  ue /.Oui  uO e / C wwdx;
O with  WD ; (3.50)
˝H

XZ Z
O WD
a.u; u/ Di;e rui;e  r uO i;e dx C  Œw.Oui  uO e /  .ui  ue /wO dx;
i;e ˝H ˝H
(3.51)
the functionals on X
8 XZ
ˆ
ˆ O WD
< hL.t/; ui uO i;e dx;
s
Ii;e
˝
Z i;e H
(3.52)
ˆ 0
:̂ h` ; ui
O WD .cm v0 vO C w0 w/ O dx;
˝H

and the operator F W D.F / V ! V0

Z
O WD
hF u; ui f .ui  ue /.Oui  uO e / dx; (3.53)
˝H

with domain

D.F / WD fu 2 V W f .ui  ue / 2 L1 .˝H / \ .H 1 .˝H //0 gI

this means that for every u 2 D.F / there exists a constant C > 0 such that
Z
f .ui  ue / dx  C jjjjH 1.˝H / ; 8 2 H 1 .˝H / \ L1 .˝H /:
˝H

Again we observe that all these definitions are compatible with the quotient space V,
provided that
Z
.Iis C Ies / dx D 0 for a.e. t 2 .0; T /: (3.54)
˝H

We can give now the precise statement of Problem (M) in a variational

abstract form.
3.4 Well-Posedness Results Based on Semi-discretization in Time 111

Problem (AM). Let us assume that

s
Ii;e 2 L2 .˝H  .0; T // and v0 ; w0 2 L2 .˝H /; (3.55)

and (3.54) is satisfied. Then, if V; b; a; L; ` 0 ; F are defined by (3.49), . . . , (3.53)

respectively, and A; B are defined as

O WD a.u; u/;
hAu; ui O O WD b.u; u/
hBu; ui O 8u; uO 2 V;

we seek u.t/ WD .ui .; t/; ue .; t/; w.; t//, with

u 2 L2 .0; T I V/; 1;1

Bu 2 Wloc .0; T I V0 / \ C 0 .Œ0; T I V0 /;

satisfying

.Bu/0 C Au C F u D L in V0 ; a.e. in .0; T /;

together with the initial condition

.Bu/.0/ D ` 0 :

It holds the following theorem, proved in [105] by applying the theory developed
in [457] for abstract evolution inequalities, based on semi-discretization in time,
a-priori error estimates and compactness properties.
Theorem 3.5. Assume that (3.54) and (3.55) are satisfied, together with

Iis C Ies 2 W 1;1 .0; T I L2 .˝H //:

Then Problem (AM) admits a unique solution u. In particular there exist a couple

ui ; ue 2 L2 .0; T I H 1 .˝H //;

uniquely determined up to a family of additive constants c.t/, and a unique couple

.v; w/ with

v 2 C 0 .Œ0; T I L2 .˝H // \ L2 .0; T I H 1 .˝H //; @t v 2 L2loc .0; T I L2 .˝H //;

w; @t w 2 C 0 .Œ0; T I L2 .˝H //;

which solve Problem (M). Moreover, if

v0 2 H 1 .˝H / and v0 f .v0 / 2 L1 .˝H /;

then

ui;e 2 C 0 .Œ0; T I H 1 .˝H //; @t v 2 L2 .˝H .0; T //; w 2 C 0 .Œ0; T I H 1 .˝H //:
112 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

3.5 Well-Posedness Results Based on Faedo-Galerkin

Techniques

In this section, we will review the well-posedness results for weak solutions of
the PE-formulation of the Bidomain model, established in [60] by using a Faedo-
Galerkin technique. See also [42, 59] for a similar approach, based on a parabolic
regularization technique.
The result in [60] holds for simple membrane models, which neglect the ionic
concentrations z. In the following, assume for sake of simplicity that the membrane
model takes into account only one gating variable, thus k D 1, w D w and F D F .
With the previous assumptions, the PE-formulation of the Bidomain model becomes
8
ˆ
ˆ @v
ˆ
ˆ cm  div.Di rv/  div.Di rue / C Iion .v; w/ D Iis in ˝H  .0; T /
ˆ
ˆ @t
ˆ
ˆ
ˆ
ˆ div.Di rv/  div..Di C De /rue / D Iis C Ies in ˝H  .0; T /
ˆ
ˆ
< @w
 F .v; w/ D 0; in ˝H  .0; T /
ˆ @t
ˆ
ˆ
ˆ
ˆ nT Di r.v C ue / D 0 in @˝H  .0; T /
ˆ
ˆ
ˆ
ˆ
ˆ n .Di C De /rue C n Di rv D 0; in @˝H  .0; T /
T T
ˆ
:̂
v.x; 0/ D v0 .x/; w.x; 0/ D w0 .x/; in ˝H :
(3.56)

R by V D H .˝H /, H D L .˝H /, U D
1 2
Preliminary variational results. Denote
V =R and, for any u 2 V , Œu D u  ˝H u 2 U . Consider now the following
variational problem:
find .u; ue / 2 U  U such that

ai .u; v/ C ai .ue ; v/ C ai .u; ve / C .ai C ae /.ue ; ve / D< si ; v > C < si C se ; ve >

(3.57)
8.v; ve / 2 U  U . The bilinear forms ai;e .; / on U  U are
Z Z
ai .u; v/ D Di rurvdx; ae .u; v/ D De rurvdx; 8.u; v/ 2 U U;
˝H ˝H

si ; se 2 V 0 are given source terms and < ;  > is the V 0 -duality.

Under the hypothesis of uniform ellipticity of the conductivity tensors Di;e , the
bilinear forms ai;e .; / are symmetric, continuous and coercive. Define now the
bilinear form

b..u; ue /; .v; ve // D ai .u; v/ C ai .ue ; v/ C ai .u; ve / C .ai C ae /.ue ; ve /

3.5 Well-Posedness Results Based on Faedo-Galerkin Techniques 113

on .U  U /  .U  U /, which is also symmetric, continuous and coercive. We have

the following
Theorem 3.6. Let si;e 2 V 0 and u 2 U be given. The variational equations

.ai C ae /.Que ; ve / D ai .u; ve / 8ve 2 U; (3.58)

.ai C ae /.ue ; ve / D< si C se ; ve > 8ve 2 U; (3.59)

have unique solutions uQ e ; ue 2 U . For any u; v 2 U , we can define the mappings

a.u; v/ D b..u; uQ e /; .v; 0//; < s; v >D< si ; v > ai .ue ; v/:

The mapping a is bilinear, symmetric, continuous and coercive on U  U , and the

mapping s is bilinear and continuous on U .
Equation (3.57) has a unique solution .u; ue / where u is also the unique
solution of

a.u; v/ D< s; v > 8v 2 U;

and ue D uQ e C ue .
We can now define the bilinear form a.; / on V  V

a.u; v/ D a.Œu; Œv/ 8.u; v/ 2 V  V:

The following theorem holds.

Theorem 3.7. The bilinear form a.; / is symmetric, continuous and coercive on
V . There exists an increasing sequence 0 D 0 < : : :  i  : : : in R and an
orthonormal Hilbert basis of H of eigenvectors f i gi 2N such that for all i 2 N,
i 2 V and

8v 2 V a. i ; v/ D i . i ; v/:

Specific assumptions and notations. The existence of weak solutions of Problem

(3.56) hold under minimal regularity assumptions:
• ˝H has a Lipschitz boundary @˝H ;
• Di;e have L1 .˝H / coefficients;
R
s
• Ii;e W Œ0; C1/ ! V 0 and ˝H .Iis .x; t/ C Ies .x; t//dx D 0 for a:e: t > 0.
In addition assume that there exists p  2 such that:
(H1) The Sobolev embedding V D H 1 .˝H / Lp .˝H / holds: p  2 if d D 2,
2  p  6 if d D 3 (see [62]);
114 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

(H2) The functions Iion and F are affine with respect to w:

Iion .v; w/ D Iion

1
.v/ C Iion
2
.v/w; F .v; w/ D F1 .v/ C F2 w;

1
where Iion 2
; Iion ; F1 W R ! R are continuous functions and F2 2 R;
(H3) There exist constants ci  0 (i D 1; : : : ; 6) such that for any v 2 R

jIion
1
.v/j  c1 C c2 jvjp1
jIion .v/j  c3 C c4 jvjp=21
2

jF1 .v/j  c5 C c6 jvjp=2 I

(H4) There exist constants a; > 0, b; c  0 such that for any .v; w/ 2 R2

vIion .v; w/ C wF .v; w/  ajvjp  b. jvj2 C jwj2 /  c:

Remark 3.1. Three examples of membrane models that satisfy these assumptions
are the FitzHugh-Nagumo model [273], the Aliev-Panfilov model [1] and the Roger-
McCulloch model [422].
Existence for the initial value problem. Under the minimal regularity assumptions
s
on the data ˝H , Di;e , Ii;e and defining D.0; T / as the space of real functions C 1
on R with compact support in .0; T /, it is possible to write the
Definition 3.1. (Weak solutions). Consider > 0 and the three functions v W t 2
Œ0; / ! v.t/ 2 H , ue W t 2 Œ0; / ! ue .t/ 2 H , w W t 2 Œ0; / ! w.t/ 2 H .
Given .u0 ; w0 / 2 H , we say that .v; ue ; w/ is a weak solution of (3.56) if, for any
T 2 .0; /,
1. v W Œ0; T  ! H and w W Œ0; T  ! H are continuous, and u.0/ D u0 , w.0/ D w0
in H ;
2. For a.e. t 2 .0; /, we have v.t/ 2 V , ue .t/ 2 V =R, v 2 L2 .0; T I V / \ Lp .QT /
where QT D ˝H  Œ0; T , and .v; ue ; w/ verify in D0 .0; T /
Z Z
d
.v.t/; vO /C Di r.v.t/Cue .t//  r vO C Iion .v.t/; w.t//Ov D< Iis .t/; vO >
dt ZH
˝ ˝H
d
.w.t/; w/O C F .v.t/; w.t//wO D 0
dt ˝H

respectively for all vO 2 V and for all wO 2 H , and

Z Z
Di rv.t/r uO e C .Di CDe /rue .t/r uO e D< Iis .t/CIes .t/; uO e > 8Oue 2 V =R:
˝H ˝H
(3.60)
3.5 Well-Posedness Results Based on Faedo-Galerkin Techniques 115

Lemma 3.1. The functions v; ue ; w are a weak solution of (3.56) if conditions 1–2
of Definition 3.1 hold and .u; w/ verify in D0 .0; T /
Z
d
.v.t/; vO / C a.v; vO / C Iion .v.t/; w.t//Ov D< s.t/; vO > 8Ov 2 V
dt Z ˝H
d
.w.t/; w/O C F .v.t/; w.t//wO D 0 8wO 2 H
dt ˝H

where a.; / and s 2 V 0 are defined in Theorem 3.6. The function ue is then
recovered from (3.60).
Theorem 3.8 (Global existence of a weak solution). Let ˝H , Di;e satisfy the
minimal regularity specified previously and suppose that the hypotheses (H1)–
(H4) on Iion and F hold forR some p  2. Let then be given u0 ; w0 2 H and
s
Ii;e 2 L2 .RC I V 0 / such that ˝H .Iis .x; t/ C Ies .x; t//dx D 0 for a.e. t > 0. Then
the system (3.56) has a weak solution .v; ue ; w/ in the sense of Definition 3.1 with
D C1.
Sketch of the proof. The proof consists of three steps:
• Construction of an approximate solution using the Faedo-Galerkin technique,
based on the special orthonormal Hilbert basis (in H ) f i gi 2N of eigenvectors
defined in Theorem 3.7;
• A priori estimates on the approximate solution;
• Compactness results and convergence of the approximate solution towards a
weak solution.
Uniqueness. Consider the function G W R2 ! R2 defined by

Iion .v; w/
G.v; w/ D 8.v; w/ 2 R2 ; for some  > 0:
F .v; w/

Let z D .v; w/ 2 R2 , Q.z/ D 12 .rG.z/T C rG.z// be the symmetric part of rG

and denote by 1 .z/  2 .z/ its eigenvalues. Suppose that

(U1) 9C 2 R W 8z 2 R2 2 .z/  1 .z/  C:

Theorem 3.9 (Uniqueness). If the condition (U1) is satisfied, then the solution
obtained in Theorem 3.8 is unique.
A different proof based on a regularization argument has been given in [59] for
the PP-formulation of the Bidomain model. The proof holds true for the Bidomain
model coupled with the Laplace equation for the torso and, in addition to the
116 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

generalized FitzHugh-Nagumo models considered in [60], a regularized version of

the Mitchell-Scheffer membrane model has been also taken into account. In this
ionic model, the functions Iion and F are given by:

w 2 v
Iion .v; w/ D v .v  1/  ;
 in out 
1 close  open
F .v; w/ D C h1 .v/ .v  h1 /;
close close open

where
  
1 v  vgate
h1 .v/ D 1  tanh
2 gate

and in ; out ; open < close ; vgate ; gate are positive constants.
The regularized finite dimensional problems used in the Faedo-Galerkin proce-
dure are constructed by adding the terms

1 @ui;n 1 @ue;n
and with n 2 N
n @t n @t

into the first and second equations of the PP-formulation of the Bidomain system,
respectively.

3.6 Well-Posedness Results Based on Fixed Point Arguments

In this section, we will review the well-posedness result for the PP-formulation of
the Bidomain model, obtained in [541] using a fixed point argument. This result
covers a wide range of complex membrane models.
The ionic current. Assume that the ionic current

Iion W R  Rk  .0; C1/m ! R

.v; w; z/ ! Iion .v; w; z/

has the general structure:

X
m
Iion .v; w; z/ WD Ji .v; w; log zi / C HQ .v; w; z/; (3.61)
i D1
3.6 Well-Posedness Results Based on Fixed Point Arguments 117

where, 8i D 1; : : : ; m,

Ji 2 C 1 .R  Rk  R/;
@
0 < G.w/  Ji .v; w; /  G.w/;
ˇ ˇ
@ (3.62)
ˇ@ ˇ
ˇ Ji .v; w; 0/ˇ  Lv .w/;
ˇ @v ˇ

G, G, Lv belong to C 0 .Rk ; RC / and

HQ 2 C 0 .R  Rk  .0; C1/m / \ Lip.R  Œ0; 1k  .0; C1/m /: (3.63)

Remark 3.2. The functions HQ (and Hi in the following) do not have a precise
physical meaning, but allow some currents which differ from Ji to be included in the
model, e.g. they could represent the Na-K pump and the non-specific Ca-activated
currents in [309].
The dynamics of the gating variables is described by the system of ODEs

@wj
D Fj .v; wj /; j D 1; : : : ; k: (3.64)
@t
Assume that

Fj W R2 ! R is locally Lipschitz continuous;

Fj .v; 0/  0 8v 2 R (3.65)
Fj .v; 1/  0 8v 2 R;

8j D 1; : : : ; k.
In the membrane models considered, Fj has the particular form

Fj .v; wj / WD ˛j .v/.1  wj /  ˇj .v/wj ; j D 1; : : : ; k;

where ˛j and ˇj are positive rational functions of exponentials in v. A general

expression for both ˛j and ˇj is given by

C1 exp. vv
C2
n
/ C C3 .v  vn /
;
1 C C4 exp. vv n
C5 /

where C1 , C3 , C4 , vn are non-negative constants and C2 , C5 are positive constants.

The dynamics of the ionic concentrations is described by the system of ODEs

@zi
D Gi .v; w; z/ WD Ji .v; w; zi / C Hi .v; w; z/ i D 1; : : : ; m; (3.66)
@t
118 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

where Ji is the function described in (3.62) and

Hi 2 C 0 .R  Rk  .0; C1/m / \ Lip.R  Œ0; 1k  .0; C1/m /; i D 1; : : : ; m:

(3.67)
Adding the two equations in (3.42), we have

 div.Di rui /  div.De rue / D Iis C Ies :

Integrating on ˝H and applying the divergence theorem and the Neumann boundary
conditions, we have the following compatibility condition for the system to be
solvable
Z
.Iis C Ies /dx D 0: (3.68)
˝H

We recall that the electric potentials in bounded domains are defined up to an

additive constant. In our case, ui and ue are determined up to the same additive
time-dependent constant, while v is uniquely determined. This common constant is
related to the choice of the reference potential. The usual choice consists in selecting
this constant so that ue has zero average on ˝H , i.e.
Z
ue dx D 0: (3.69)
˝H

Suppose then that Di .x/ and De .x/ are measurable and satisfy the uniform
ellipticity condition (3.45).
The condition on the initial datum. In view of the result of continuity for
the solution v of the Bidomain model, we must ask for the initial datum v0 to
be compatible, in a sense to specify, with the Neumann homogeneous boundary
conditions. Intuitively, if v0 D u0i  u0e , then we should have

Di ru0i  n D 0 D De ru0e  n; on @˝H ;

but fixing both ui .x; 0/ and ue .x; 0/, as initial data, may render the problem
unsolvable, because the time derivative involves only the difference ui  ue . The
correct assumption may seem abstract, see [541, Sec. 3]: let v 2 H 1 .˝H / be given,
then the following minimization problem has a unique solution:
( )
XZ Z
min Di;e rui;e  ui;e dx W ui;e 2 H .˝H /;
1
ue dx D 0; ui  ue D v :
i;e ˝H ˝H
(3.70)
3.6 Well-Posedness Results Based on Fixed Point Arguments 119

Now, if Iis .0/ C Ies .0/ 2 L2 .˝H /, then the following elliptic problem admits
a unique solution u0b 2 H 2 .˝H /:
8
ˆ
ˆ  div..Di C De /ru0b / D Iis .0/ C Ies .0/ in ˝H ;
<
Z i C De /rub /  n D 0
0
..D on @˝H ;
ˆ
:̂ ub dx D 0:
0
˝H

Finally, we say that an initial datum v0 satisfies the admissibility property if


the couple .ui ; ue / solution of (3.70) w.r.t. v0 ; satisfies
(3.71)
Di .rui C u0b /  n D De .rue C u0b /  n D 0 on @˝H :

We can now state the main result of [541], concerning the existence of a
variational solution for (3.42).
Theorem 3.10. Assume that

˝H is of class C 1;1 ; Di;e are Lipschitz in ˝H :

Let us take as given data

v0 2 H 2 .˝H /; satisfying the admissibility property (3.71);

w0 W ˝H ! Œ0; 1k ; measurable;
z0 2 .L2 .˝H //m ; with log.z0 / 2 .L2 .˝H //m ;
s
Ii;e 2 Lp .0; T I L2 .˝H //; for p > 4; satisfying (3.68) and
Ii C Ies 2 H 1 .0; T I L2 .˝H //:
s

Let us also take as given the ionic currents satisfying (3.61)–(3.63), the dynamics of
the gating variables F.v; w/, satisfying (3.64) and (3.65), the dynamics of the ionic
concentrations G.v; w; z/, satisfying (3.66) and (3.67).
Then there exists a unique solution of Problem (3.42), given by k C m C 2
functions w1 ; : : : ; wk ; z1 ; : : : ; zm ; ui ; ue satisfying

ui;e 2 Lp .0; T I H 2 .˝H //;

v WD ui ue 2 W 1;p .0; T I L2 .˝H // \ Lp .0; T I H 2 .˝H // \ C 0 .Œ0; T I C 0 .˝H //;
w W Q ! Œ0; 1k measurable; z W Q ! .0; C1/m measurable;
wj .x; / 2 C .0; T / \ C .Œ0; T / for a.e. x 2 ˝H ; j D 1; : : : ; k;
1 0

zi .x; / 2 C 1 .0; T / \ C 0 .Œ0; T / for a.e. x 2 ˝H ; i D 1; : : : ; m;

z 2 H 1 .0; T I L2 .˝H //m \ L1 .Q/m ; log z 2 L1 .Q/m ;

where we have defined Q D ˝H  .0; T /.

120 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

Sketch of the proof. The proof of Theorem 3.10 is divided into three parts.
In a first step, one fixes v and solves the ODE systems of the gating and
ionic concentrations variables, obtaining suitable a priori estimates and qualitative
properties of the solution.
In the second step, a reduction technique is used in order to split the degenerate
parabolic system into an elliptic equation coupled with a non-degenerate parabolic
equation in L2 .˝H /, governed by the generator of an analytic semigroup. Consider-
ing Iion .v; w; z/ as a known function, we apply a result of maximal regularity in Lp ,
obtaining the existence and uniqueness of, and estimates for, the potentials ui ; ue
(and thus for v D ui  ue ) in Lp .0; T I H 2 .˝H // \ W 1;p .0; T I L2 .˝H //.
These estimates, owing to classical interpolation techniques, provide a crucial
bound for v in L1 .Q/. Then, by choosing the correct functional spaces for w, z
and v, it is possible to establish the existence and uniqueness of a solution .v; w; z/
for Problem (3.42), using Banach’s Fixed Point Theorem.
The main difficulties in the parabolic equation reside in its degenerate structure,
which reflects the differences in the anisotropy of the intra- and extracellular tissues,
and in the lack of a maximum principle. Moreover, the concentration variables zi
appear as arguments of a logarithm, both in the dynamics of the concentrations and
in the ionic currents, and therefore it is necessary to bound z far from zero.

3.7 Semi-discrete Approximation of the Bidomain Model

with FHN Dynamics

We conclude this section mentioning some examples of approximation results for

the averaged model P; for other approximation results in the context of reaction-
diffusion problems, see e.g. [230, 259, 321, 334].
At first, we consider a semi-discrete scheme in space obtained using conforming
linear finite elements, (see e.g. [414] for a general introduction to the finite element
method). Assuming that ˝H is a polygonal convex domain in R3 and Th is a
family of triangulations associated with a reference polyhedron EO by an invertible
affine maps TE , for every E 2 T ; we denote by Vh the finite dimensional space
of continuous functions whose restriction to each element of T are polynomial
of degree 1. A semidiscrete problem is obtained by applying a standard Galerkin
procedure on Vh to the averaged model (3.34). In [455] various stability results are
shown and the following error estimate:
Theorem 3.11. For a regular and quasi-uniform mesh and a regular initial datum
w0 2 H 1 .˝H /, denoting by U h .t/ the finite element approximation of the
semidiscrete approximation of (3.37) with FitzHugh-Nagumo dynamics, then the
following optimal a priori error estimate holds
Z
  T  
e2h WD max b U .t/  U h .t/ C a U .t/  U h .t/ dt  Ch2 :
t 2.0;T / 0
3.7 Semi-discrete Approximation of the Bidomain Model with FHN Dynamics 121

We now consider a semi-discrete approximation in time of the averaged model

applying the implicit Euler scheme. More precisely, we choose a partition of the
time interval Œ0; T  into N subintervals

P D f0 D t0 < t1 < t2 < : : : tN 1 < tN D T g; with variable step n D tn  tn1

and we set D max1nN n.

uτ (t)

Given U 0 , we introduce the sequence of variational problems: Find U n 2 Vh ;

n D 1; : : : ; N such that

U n  U n1 O
b. ; U / C a.U n ; UO / C F .U n ; UO / D 0; 8UO 2 Vh : (3.72)
n

Considering the discrete solution U  .t/ given by the continuous piecewise linear
function interpolating the values fU n gN
nD0 on the grid P, we have the following
error estimate
Theorem 3.12. For sufficiently regular initial data, the following a priori error
estimate between u and U  , measured by the natural variational (semi)norms holds:
Z
  T  
e2 WD max b U .t/  U  .t/ C a U .t/  U  .t/ dt  C 2
;
t 2.0;T / 0

or equivalently
p
k b.U  U  /kL1 .0;T / C kU  U  kL2 .0;T IV/  C ;

with C independent of :
We conclude presenting a result related to a posteriori error estimates. In [32],
resorting to the theory developed in [359] for evolution variational problems,
a posteriori error estimates were derived for general degenerate evolution equations.
122 3 Mathematical Models of Cardiac Cells Arrangements: The Bidomain Model

Theorem 3.13. For sufficiently regular initial data, let e be the error between
U and U  measured by the natural variational (semi)norm
h Z
 T   i
e2 WD max max e 2 gt
b U .t/  U  .t//; e 2 gt
a U .t/  U  .t/ dt
t 2.0;T / 0

 
with g D infv2R g 0 .v/ .
Then applying the theory of [32] to the Bidomain model, the error e can be
estimated a posteriori by

X 2 X q  
g
e2   n Dn
2
C 2
n b ıU n ;
n
2 n
     
where Dn D a U n  U n1 C F U n ; U n  U n1  F U n1 ; U n  U n1 ;
U n  U n1
ıU n D :
n
Chapter 4
Reduced Macroscopic Models:
The Monodomain and Eikonal Models

In the Bidomain model (3.42), the transmembrane potential v during the excitation
phase of the heartbeat exhibits a steep propagating layer spreading throughout the
myocardium with a thickness of about 0.5 mm. At every point of the cardiac domain,
this upstroke phase lasts about 1 ms. Therefore, the simulation of the excitation
process requires the numerical solution of problems with small space and time steps
(of the order of 0.1 mm and 0.01 ms). A further inconvenient of the Bidomain model
is its severe ill-conditioning, mainly due to the pure Neumann boundary conditions,
of the linear systems deriving from its discretization. The solution of such linear
systems with iterative methods such as the Conjugate Gradient, preconditioned
by standard cheap preconditioners (ILU or SSOR), is not effective, thus more
sophisticated computational demanding preconditioners are required (see Chaps. 7
and 8).
All these facts constraint 3–D simulations to limited blocks with dimensions of
a few cm, see [101, 102, 104, 226]. For large scale simulations involving the whole
ventricles, the computer memory and time requirements become excessive and less
demanding approximations have been developed, such as Monodomain and Eikonal
models.

4.1 Linear Anisotropic Monodomain Model

In order to reduce further the computational load many large scale simulations have
been performed using the so-called Monodomain model; it is well known that if
the two media have the same anisotropy ratio then the Bidomain system reduces
to the Monodomain model. We remark that this is not a physiological case, as it
clearly follows from well established experimental evidence. We shall present an
interesting derivation of a reduced Bidomain model, which does not make such a
priori assumption (see also [99,269]) and that we will still call Monodomain model.

© Springer International Publishing Switzerland 2014 123

P. Colli Franzone et al., Mathematical Cardiac Electrophysiology, MS&A 13,
DOI 10.1007/978-3-319-04801-7__4
124 4 Reduced Macroscopic Models: The Monodomain and Eikonal Models

Denoting by Jtot D ji C je the total current flowing in the two media and by
D D Di CDe the conductivity of the bulk medium, since Jtot D Di rui De rue ,
substituting ui D v C ue , we obtain

rue D D 1 Di rv  D 1 Jtot : (4.1)

Therefore, the second equation in the Bidomain system (3.42) can be written as

@v
 cm C div.De D 1 Di rv/ C div.De D 1 Jtot /  iion .v; w; c/ D Iapp
e
: (4.2)
@t
Since the conductivity tensors are given by (3.41), then

De D 1 D el I C .et  el /at .x/aTt .x/ C .en  el /an .x/aTn .x/; (4.3)

with el;t;n D l;t;n

e e
=.l;t;n C l;t;n
i
/. Assuming constant conductivity coefficients and
taking into account that div Jtot D Iapp i
C Iapp
e
, we have

div.De D 1 Jtot / D el div Jtot C .et  el / divŒat .x/aTt .x/Jtot 
C.en  el / divŒan .x/aTn .x/Jtot 
(4.4)
D el .Iapp
i
C Iapp
e
/ C .et  el / divŒat .x/aTt .x/Jtot 
C.en  el / divŒan .x/aTn .x/Jtot :

From (4.1) it follows De D 1 Di rv D De D 1 Jtot C De rue , so we have the flux

relationship

 nT De D 1 Di rv D nT De D 1 Jtot C nT De rue : (4.5)

Using the split form (4.3), the first term on the right hand side can be written as

nT .De D 1 Jtot / Del nT Jtot C .et  el /.nT at /.aTt Jtot /

C .en  el /.nT an /.aTn Jtot /: (4.6)

The insulating conditions nT ji D nT je D 0 imply nT Jtot D 0, i.e. Jtot is tangent

to H , and assuming that fibers are also tangent to H we have nT an D 0 and
aTt Jtot D 0; substituting these conditions in (4.6), it follows

nT De D 1 Di rv D 0: (4.7)

We remark that for media having equal anisotropic ratio, i.e. le =li D te =ti D
ne =ni , we have el D et D en . Thus, the two additional terms in (4.4), related
to the projections of Jtot on the directions across fiber, disappear. Disregarding
these two additional source terms aTt Jtot and aTn Jtot , we have div.De D 1 Jtot / 
4.2 Eikonal Models 125

el .Iapp
i
C Iapp
e
/. Substituting this approximation in (4.2) and considering the
boundary condition (4.7), we obtain an approximate model consisting in a single
parabolic reaction-diffusion equation for v and coupled with the same gating system
8
ˆ @v
ˆ
ˆ cm  div.Dm rv/ C iion .v; w; c/ D Iapp
m
in ˝H  .0; T /
ˆ
ˆ @t
< @w @c
 R.v; w/ D 0;  S.v; w; c/ D 0 in ˝H  .0; T /
ˆ
ˆ @t @t
ˆ
ˆ n Dm rv D 0
T
in H  .0; T /
:̂
v.x; 0/ D v0 .x/; w.x; 0/ D w0 .x/; c.x; 0/ D c0 .x/ in ˝H ;
(4.8)
 i e 
with the conductivity tensor Dm D De D 1 Di , Iapp
m
D Iapp l  Iapp
e
li =.le C li /.
The evolution equation determines the distribution of v.x; t/ and then the
extracellular potential distribution ue is derived by solving the elliptic boundary
value problem

(
div.Drue / D div.Di rv/ C Iapp
i
C Iapp
e
in ˝H ;
(4.9)
n Drue D n Di rv on H :
T T

We refer to the system consisting of Eqs. (4.8) and (4.9) as the anisotropic
Monodomain model. We remark that the Bidomain and the Monodomain model
are described by a system of a parabolic equation coupled with an elliptic equation,
but in the latter the evolution equation is fully uncoupled with the elliptic one in the
case of an insulated domain ˝H .

4.2 Eikonal Models

Another route to avoid the high computational costs of the full Bidomain model is
based on the use of eikonal models for the evolution of the excitation wavefront
surface.
With these models the simulation of the activation sequence in large volumes of
cardiac tissue has become computationally practical but at the price of a loss of fine
details concerning the thin layer where the upstroke of the action potential occurs.
These numerical simulations are based on laws describing the macroscopic kinetic
mechanism of the spreading of the excitation wavefronts, and do not require a fine
spatial and temporal resolution.
We now outline the derivation of this kind of approximated models. The
FitzHugh–Nagumo approximation of the membrane kinetics is very useful for a
qualitative analysis of the non-linear dynamics of the R-D system. As we shall focus
only the excitation phase, we can neglect the recovery variable w hence iion D g.v/.
126 4 Reduced Macroscopic Models: The Monodomain and Eikonal Models

The resulting simplified ionic model is widely used for gaining general insight
into wave propagation in the cardiac excitable medium. Although this model is not
suitable in a quantitative detailed study, at a fine scale, of the upstroke of the action
potential v through the excitation wavefront, it is nevertheless appropriate if we are
interested in the large scale behavior of the front-like solution. We note that during
the excitation phase of the heart beat the main feature, at a macroscopic level, is
the excitation wavefront configuration and its motion. In order to investigate the
propagation of this wavefront we must analyze more deeply the internal layer of v
which affects the spreading.
Denoting by  the membrane surface area per unit volume of the tissue, by Cm
the membrane capacitance per unit area of the membrane surface, and by Iion the
ionic current per unit area of the membrane surface, that is carried by the flow of
ions across the membrane, then the transmembrane current im per unit volume is
given by
 
@v
im D Cm C Iion :
@t

We recall that the so-called fast sodium current INa is the main current responsible
for the depolarization of cardiac cells during the excitation phase of the transmem-
brane action potential. Experimental findings in measuring INa in different cardiac
preparations justify the modeling of the sodium current mechanisms by means
of gating equations of the Hodgkin-Huxley type, see e.g. [193, 308] for guinea
pig, [367] for rat, [473] for rabbit, [209, 521] for human myocytes. Hence INa is
represented by

INa D GN Na m3 h j.v  vNa /;

where the m; h; j variables, modeling the sodium conductance dynamics, referred

to as the sodium activation and inactivation, satisfy the equations

dm m1 .v/  m dh h1 .v/  h dj j1 .v/  j

D ; D ; D : (4.10)
dt 1 .v/ dt 1 .v/ dt 1 .v/

In the depolarization phase the time constant of the activation gate m for ventricular
cells is about 0.2 ms, whereas the inactivation time constants are of order 50
and 300 ms for h and j gates, respectively. Thus, we assume that the membrane
obeys the simplified time-independent ionic current-voltage relation, obtained by
considering an instantaneous change of sodium conductance, i.e. taking m D
m1 .v/, h D h1 .vr / ' 0:98, j D j1 .vr / ' 0:99, where vr is the resting potential.
In this simplified approximation, the ionic current model is given by the non-linear
current-voltage law

Iion .v/ D INa .v/ C IL .v/;

4.2 Eikonal Models 127

ZOOM
200 1

−200

−400 0

Iion ( A/cm2)
(μ A/cm2)

−600

μ
−800
ion
I

−1000 −1

−1200

−1400

−1600 −2
−100 −80 −60 −40 −20 0 20 40 60 −90 −85 −80 −75 −70 −65 −60

v (mV) v (mV)

Fig. 4.1 Cubic-like behavior of the ionic current per unit area of membrane surface Iion (A=cm2 )
as a function of the transmembrane potential v (mV)

with

INa .v/ D GN Na m1 .v/3 .v  vNa /; IL D GN L .v  vL /;

where IL is the leakage current. Using the GN Na , vNa and m1 .v/ expressions in the
sodium current of the LR1 model [308] and choosing GN L D 0:05 mS=cm2, Vl D
80 mV, the behavior of Iion .v/ is of cubic-like type as displayed in Fig. 4.1, i.e. it
has the following properties:
1. Iion 2 C 1 .R/;
2. There exist only three zeros vr < vth < vp ;
0 0
3. RIion
vp
.vr / > 0 and Iion .vp / < 0;
4. vr Iion .v/d v < 0.

We want to investigate the approximate kinematic properties of traveling wave-

front solutions of the Bidomain system (3.42), neglecting some details of the
transmembrane action potential, but retaining its essential features. In order to gain
this information, we carry out a singular perturbation analysis of an appropriately
scaled form of system (3.42). Our interest concerns the description of the stationary
or regular propagation of the excitation wavefronts, that is, the spreading away from
the initial phase related to the starting stimuli or from front-front or front-boundary
collisions.
In the stationary propagation, we can scale the x and t variables, taking spatial
(L) and temporal (T ) scale factors in such a way that the corresponding mean
wavefront velocity becomes of unitary order. In this phase of the excitation process,
the velocity varies between 0:3 and 0:7 mm/ms, according as it is measured
longitudinally of transversally to the fiber direction. Since the upstroke of the
transmembrane potential lasts about 2 ms, for an average propagation velocity of
 D 0:5 mm/ms, we estimate an average thickness of the excitation layer 1 mm. It
follows that a characteristic time T D 20 ms is the average time required by the
128 4 Reduced Macroscopic Models: The Monodomain and Eikonal Models

traveling impulse to cover a characteristic distance of L D 1 cm corresponding to

the transmural thickness of the ventricular wall. Therefore, rescaling the x and t
variables using L and T as

x t
xO D tO D ;
L T
and setting

Iion
gN Na D GN Na g.v/ D  ;
gN Na

the rescaled Bidomain system becomes

8  
ˆ
ˆ cm @Ov 1
< D divxO D i xO i C g.O
r O
u v/ in ˝H
gN Na T @tO gN Na L2
  (4.11)
ˆ cm @Ov 1
:̂  D divxO De rxO uO e  g.Ov/ in ˝H :
gN Na T @tO gN Na L2

Using the following typical parameters values

Cm D 1 F=cm2 ; GN Na D 20 ms=cm2;  D 103 cm1 ;

i;e
and recalling that the conductivity coefficients l;t;n are in the range 0.1–2 mS=cm2 ,
we obtain
i;e
cm l;t;n i;e
D m; D  2 O l;t;n ;
gN Na T gN Na L2

with  of order 102 , m of unitary order and O l;t;n

i;e
in the range 0.1–1.
Hence, the macroscopic dimensionless form of the Bidomain model can be
written as the following singularly perturbed reaction-diffusion (R-D) system
8
ˆ @Ov
< m D  2 divxO .DO i rxO uO i / C g.Ov/ in ˝H
@tO (4.12)
:̂  m @Ov D  2 divxO .DO e rxO uO e /  g.Ov/ in ˝H :
@tO

In the following, we will denote by v" D vO , u"i;e D uO i;e , and for the sake of simplicity
in the notations, we will drop all theOsymbols out elsewhere.
The R-D systems with excitable dynamics are studied by mathematical tools
from singular perturbation theory, see e.g. [174, 273]. Given the previous singular
perturbation structure, u"i , u"e diffuse quite slowly, while the reaction takes place
much faster; hence, the development of a moving layer associated with a traveling
wavefront solution is to be expected. Exploiting the singular perturbation approach,
4.2 Eikonal Models 129

we can derive anisotropic geometric evolution laws capturing the asymptotic

behavior of traveling wavefront solutions of the R-D system (4.12) (see [37, 39,
112, 113, 268, 271–273] and for isotropic media [492]).
Assuming that the excitation propagates in fully recovered tissue, a monotonic
temporal behavior of v" is expected; then the excitation wavefront S" .t/ can be
represented by the level surface of the transmembrane potential of value .vr Cvp /=2,
i.e.:

S" .t/ D fx 2 ˝H ; v" .x; t/ D .vr C vp /=2g: (4.13)

We define the activation time " .x/ as the time instant at which the potential v"
at x reaches the value .vr C vp /=2. Then the excitation wavefront S" .t/ is also
represented by the level surface of the activation time at the time instant t, i.e.:

S" .t/ D fx 2 ˝H ; " .x/ D tg: (4.14)

Let

qi;e .x; / D  T Di;e .x/ (4.15)

be the conductivity coefficient at a point x in the intra- and extracellular media,

respectively, measured along the direction of the unit vector . We define the
harmonic mean of the quadratic forms associated with the conductivity tensors Di;e
by

q.x; / D .qi .x; /1 C qi .x; /1 /1 : (4.16)

The non-linear form q.x; / admits the following representation

qi .x; /2 De .x/ C qe .x; /2 Di .x/

q.x; / D  T Q.x; / with Q.x; / D ;
Œqi .x; / C qe .x; /2
(4.17)

which gives the conductivity measured along the direction  of the bulk medium
composed by coupling in series the media (i) and (e). Then we introduce the
following indicatrix function
p
˚.x; / D q.x; /: (4.18)

Under the assumptions 1, 2, 3, 4 let (c,a) be the unique bounded solution of the
eigenvalue problem (see Sect. 3.1.3):

a00 C c a0 C g.a/ D 0
(4.19)
a.1/ D vp ; a.1/ D vr ; a.0/ D .vp C vr /=2:
130 4 Reduced Macroscopic Models: The Monodomain and Eikonal Models

A formal inner asymptotic expansion in powers of " of .u"i ; u"e / solution of (4.12)
and v" D u"i  u"e can be performed using two different types of variable stretching.
Let  be the Euclidean unit vector normal to the wavefront S" .t/ oriented toward
the resting tissue and for s.t/ 2 S" .t/ we define the vector n˚ .s/ D ˚ .s; /. As
in [39] we use a Lagrangian point of view and we consider the moving reference
.s; y; / defined by

yD ; x D s.t/ C  n˚ .s.t//; D t; 8s.t/ 2 S" .t/; (4.20)
"
i.e. stretching the space coordinate along the n˚ direction.
Using the moving frame (4.20) the asymptotic expansion for the Bidomain model
(4.12) shows that (see [39], Appendix B), at least formally, the front associated with
(4.13) moves along the relative normal vector n˚ with velocity  .n˚ / given at any
s.t/ 2 S" .t/ by:

c  " div n˚
" .n˚ / D C O."2 /; (4.21)
m

where c is related to the traveling wave solution a of (4.19). Since n˚   D

˚.s; /, the velocity  ./ in the Euclidean normal direction  of S" .t/ is given
by ˚.s; / .n˚ / then:

˚.s; /
 ./ D .c  " div ˚ .s; // C O."2 /: (4.22)
m

Therefore dropping O."2 / terms, the front behaves as an hypersurface S.t/,

propagating with the anisotropic geometric law with normal velocity ./ given
by:

˚.s; /
./ D .c  " div ˚ .s; //: (4.23)
m

Equations of this type are also called eikonal-curvature models since k˚ D

div n˚ D div ˚ .s; / is the anisotropic mean curvature with respect to a suitable
Finsler metric; see [15, 37–39] for definitions and properties.
A formal derivation based on an Eulerian point of view was developed in [112,
113]; in this approach the new frame .; / is defined by stretching the time variable
with respect to the activation time, i.e.:

t  .x/
 D x; D : (4.24)
"
4.2 Eikonal Models 131

Using the fixed Eulerian frame (4.24) and developing the asymptotic expansion
of the Bidomain model (4.12), we obtain the following expression, equivalent to
formulae (59) and (61) derived in [113]:

1   
m C " div ˚.x; r " /˚ .x; r "/ D c C O."2 /: (4.25)
˚.x; r "/

Since
r " 1
D and " ./ D ;
jr "j jr " j

we obtain
  
" ./ ˚.x; /˚ .x; /
m C " div D c C O."2 /; (4.26)
˚.x; / " ./

or equivalently
 
" ./ ˚.x; / " ./
m D c  " div ˚ .x; / C " r  ˚ .x; / C O."2 /:
˚.x; / " ./ ˚.x; /
(4.27)

" ./
Since both Eqs. (4.22)–(4.27) imply D c C O."/, then the two eikonal
˚.x; /
equations (4.22), (4.27) are equivalent up to second order terms. Equations (4.23)
and (4.27), disregarding the O."2 / term, are called respectively eikonal–curvature
and eikonal–diffusion equations, [121, 522].
The rigorous justification of the connection between the evolution of a suitable
level set of v and the surface flowing under geometric evolution law, remains
to our knowledge an open problem. A partial rigorous characterization in the
 convergence framework was obtained for the stationary Bidomain model in [4].
We introduce the family of vectorial integral Lyapunov functionals dependent on
the couple u" WD .ui ; ue /
Z  
F " .u/ WD " Di rui rui C De rue rue dx
˝
Z (4.28)
1
C G.ui  ue / dx;
" ˝

where G 0 D g.
The degenerate reaction-diffusion system associated with (4.12) in the couple of
unknowns u WD .u"i ; u"e / can be obtained by taking the gradient flow of the Lyapunov
functional with respect to the positive but degenerate bilinear form in L2 .˝I R2 /
132 4 Reduced Macroscopic Models: The Monodomain and Eikonal Models

Z
b.u; w/ WD .ui  ue / .wi  we / dx; u D .ui ; ue /; w D .wi ; we /:
˝

This yields the following system of variational evolution equations

b.@t u" ; v/ C ıF " .u" ; v/ D 0 8v 2 H 1 .˝I R2 /; (4.29)

 
@ui @ue
where @t u" D ; and ıF " .u; / is the Euler-Lagrange first variation
@t @t
of the functional F " [201], which is the variational formulation of (4.12). Since
the anisotropic curvature k˚ corresponds to the first variation of the anisotropic
surface energy integral functional associated with ˚ given by (4.18) see [38], in
order to justify the form of the anisotropic curvature term, we have studied in
[4] the asymptotic limit, as " # 0, of the stationary problem associated with the
singular reaction–diffusion system (4.12) for a function g D G 0 , where now G is
a potential having wells of equal depth. More precisely, the  limit of Lyapunov
functionals associated with the family (4.28) is a surface integral functional whose
energy density is a continuous family of norms ˚  .x; / characterized by solving a
localized minimization problem, see [4] for details.
Formal asymptotic results
p in the Bidomain case (see [39, 112, 113]) suggest that
˚  .x; / D ˚.x; / D q.x; /: On the other hand, in some pathophysiological
setting, such as regional ischemia and a healed infarction, the corresponding
conductivity tensors Di ; De yield a nonconvex ˚: since ˚  is always convex, in
this case the previous equality does not hold. It would be interesting to check if the
convex envelope of ˚ is a good substitute in this case.

4.3 Relaxed Non-linear Anisotropic Monodomain Model

We can easily see that, rescaling as in (4.12) the reaction-diffusion equation related
to the Monodomain model (4.8) and using formal asymptotic expansions as before,
the anisotropic evolution law of the front does not coincide with that derived from
the Bidomain model. In fact, although the eikonal-curvature equation up to terms of
order O."2 / presents the same structure
˚.s; /
./ D .c  " div ˚ .s; //; (4.30)
m
p
the non-linear function ˚.x; / D q.x; / for the Monodomain model is defined
by

q.x; / D  T Dm .x/ where Dm .x/ D De .x/.Di .x/ C De .x//1 Di .x/;

(4.31)

i.e. Dm is the harmonic mean tensor associated with Di;e .

4.4 Dimensional Form of the Reduced Models 133

We now consider the following Monodomain model with non-linear diffusion

term, which we call the relaxed Monodomain model

@v 1
 g.v/  " div .Q.x; rv/rv/ D Iapp ; (4.32)
@t "

with Q.x; / defined in (4.17) and written explicitly in terms of the conductivity
tensors as
!2  2
 T Di .x/  T De .x/
Q.x; / WD De .x/ C Di .x/; D WD Di C De :
 T D.x/  T D.x/
(4.33)

The diffusion term is non-linear except when De D Di , with a constant 2 R,

i.e. for equal anisotropic ratio of the two media, Q.x; / D 1C Di .x/.
Proceeding by formal asymptotic expansions (see [39] Appendix A), the relaxed
Monodomain model admits the same eikonal-curvature equation associated with
the Bidomain model. The non-linear conductivity tensor of the medium Q.x; rv/,
being homogeneous of degree zero in the second variable, is a function of the local
direction of propagation of the front-like solution given by the unit vector rv=jrvj.
In [39], we show that the eikonal-curvature equation as an approximate model for
describing the evolution of the relaxed transmembrane potential v can be justified
rigorously by estimating the distance between a suitable level set of the relaxed
evolution v solution of (4.32) and the surface flowing under the geometric law (4.23).
We observe here that a suitable convexity property of ˚ is crucial for this rigorous
result (see [39] for details). Such a property is true in a wide range of physiological
choices but is not guaranteed for generic choices of matrices Di , De . Pathological
anisotropies, e.g., modeling ischemic tissue can lead to a nonconvex ˚, hence to
a not well-posed relaxed model. This issue requires further study since it could be
related to mechanisms of reentry phenomena associated with cardiac arrhythmias in
presence of ischemic substrates (see [16]).

4.4 Dimensional Form of the Reduced Models

Recalling the definition of the function ˚ in (4.18), the motion of the excitation
wavefront is described by the following kinetic equation
 
1
./ D ˚.x; /   div ˚ .x; / ; (4.34)
cm

which is the dimensional form of Eq. (4.23). In terms of the activation time .x/,
r
since  D jr j
and thanks to the homogeneity properties of ˚ and ˚ , we obtain
the
134 4 Reduced Macroscopic Models: The Monodomain and Eikonal Models

Eikonal-Curvature equation (see (4.23) for the dimensionless form)

1  
 ˚.x; r .x// div ˚ .x; r .x// C ˚.x; r .x// D 1 in ˝H :
cm
(4.35)

An equivalent formulation of (4.34) can be derived using the level set approach,
in terms of the zero level set of a function w.x; t/. Setting the depolarized cardiac
region at time t by

Hd .t/ D fx 2 ˝H W t < .x/g

and considering a function w.x; t/ positive inside Hd .t/ and negative outside, the
excitation wavefront S.t/ at time t can be represented as

S.t/ D @Hd .t/ D fx 2 ˝H W w.x; t/ D 0g:

Denoting as previously by  the normal to the excitation wavefront pointing toward

the resting region, we have that

rw
D :
jrwj

Since the normal velocity can be now written as

@t w
./ D  ;
 T rw
Eq. (4.34) becomes
 
@w 1
D ˚.x; rw/  C div ˚ .x; rw/ : (4.36)
@t cm

Eikonal-Diffusion equation (see (4.25) for the dimensionless form)

1  
 div ˚.x; r .x//˚ .x; r .x// C ˚.x; r .x// D 1 in ˝H :
cm
(4.37)

We remark that, in isotropic homogeneous media (Di D i I and De D e I ),

the propagation velocity ./ along the normal  of a plane activation wavefront is
derived from (4.35) dropping the curvature term, i.e.

˚.r / D jr j˚./ D 1;

4.4 Dimensional Form of the Reduced Models 135

thus
1 p i e
./ D D ˚./ D  ; with D :
jr j i C e

The membrane constant  can be computed analytically in the case of Fitzhugh

equation. In the case of complex membrane models, in order to estimate , one
considers the propagation of the action potential along a one dimensional fiber. For
given membrane area per unit volume  and conductivity coefficient , one solves
numerically the cable equation
 
@v @2 v
 Cm C Iion D  2 ;
@t @x

coupled with the membrane model considered. Then, estimating the steady velocity
of the action potential  along the cable and the conductivity coefficient , we derive
 D p .
While we have focused so far on the use of reduced models for the excitation
phase, we note that the linear anisotropic Monodomain model and the relaxed
Monodomain model could also be used as reduced models in all phases of the
heartbeat. These two models in dimensional form are
Linear Anisotropic Monodomain model
8
ˆ @v
ˆ
ˆ cm  div.Dm rv/ C iion .v; w; c/ D Iapp m
in ˝H  .0; T /
ˆ
ˆ @t
< @w @c
 R.v; w/ D 0;  S.v; w; c/ D 0 in ˝H  .0; T /
ˆ T
ˆ @t @t
ˆ
ˆ n Dm rv D 0 in H  .0; T /
:̂
v.x; 0/ D v0 .x/; w.x; 0/ D w0 .x/; c.x; 0/ D c0 .x/ in ˝H ;
(4.38)

with Dm D De .Di C De /1 Di .

Relaxed Monodomain model (see (4.32) for the dimensionless form)
8
ˆ @v
ˆ
ˆ cm  div.˚.x; rv/˚ .x; rv// C iion .v; w; c/ D Iapp in ˝H  .0; T /
ˆ
ˆ
< @w@t @c
 R.v; w/ D 0;  S.v; w; c/ D 0 in ˝H  .0; T /
ˆ
ˆ @t @t
ˆ
ˆ n ˚.x; rv/˚ .x; rv/rv D 0
T
in H  .0; T /
:̂
v.x; 0/ D v0 .x/; w.x; 0/ D w0 .x/; c.x; 0/ D c0 .x/ in ˝H :
(4.39)
136 4 Reduced Macroscopic Models: The Monodomain and Eikonal Models

4.4.1 Well-Posedness Results for Reduced Models

From (3.45), it follows that there exist c1m ; c2m > 0 such that

c1m jzj2  zT Dm .x/z  c2m jzj2 8z 2 R3 ; 8x 2 ˝H :

This property, coupled with the assumptions of Theorem 3.10, guarantees the well-
posedness of the Linear Anisotropic Monodomain model (4.38).
Assuming that 8x 2 ˝H the map  ! ˚ 2 .x; / is uniformly strongly convex,
i.e. there exists c > 0 independent of x such that ˚ 2 .x; /  c2 jj2 is convex, then
the mapping  ! ˚.x; /˚ .x; / is uniformly strongly monotone, i.e.

.˚.x; /˚ .x; /  ˚.x; /˚ .x; //  .  /  cj  j2 :

This monotonicity property, coupled with the assumptions of Theorem 3.10 on the
non-linear reaction terms of FitzHugh-Nagumo type, ensures the well-posedness of
the Relaxed Monodomain model (4.39).
Considering the Eikonal-Diffusion equation (4.37) with mixed Dirichlet-
Neumann boundary conditions, the same monotonicity property should guarantee
the well-posedness.
Regarding the level set formulation (4.36) of the Eikonal-Curvature model, we
remark that the mapping associated to the right hand side of (4.36) is continuous,
degenerate elliptic and geometric. Moreover, it is easy to verify that the map ˚.x; /
satisfies the properties (6.1) and (6.2) of [37]. Thus, proceeding as in [37], one
can prove that Eq. (4.36) admits a unique continuous viscosity solution by applying
Theorem (4.9) of [204], see also [27, 202, 203].
We recall that differently from to the Relaxed Monodomain and Eikonal models,
the convexity assumption on the map ˚ 2 is not needed for the well-posedness of
the Bidomain model, and the only assumption required on the tensors Di;e is the
uniform ellipticity (3.45).
We refer to [3,40] for works investigating a generalized multidomain system and
eikonal models with non-convex indicatrix function ˚.

4.4.2 Frank and Wulff Diagrams

We have seen before that the solvability of both the eikonal curvature and diffusion
equations, as well as of the relaxed Monodomain model, is guaranteed if the function
 ! ˚ 2 .x; / is strictly convex. Such property depends on the conductivity tensors
Di .x/, De .x/. After some relatively easy computations, this strict convexity is
equivalent to the following property:
4.4 Dimensional Form of the Reduced Models 137

!.x; p/T ŒQ.x; p/1 !.x; p/

14 > 0; 8p 2 Rn jjpjj2 D 1 8x 2 ˝H ; (4.40)
pT D.x/p

where

pT Di .x/p pT De .x/p
!.x; p/ WD T
De .x/ p  T Di .x/ p; D WD Di C De ;
p D.x/p p D.x/p
 T 2  T  2
p Di .x/p p De .x/p
Q.x; p/ WD De .x/ C Di .x/:
pT D.x/p pT D.x/p

The conductivity tensors are defined as in (3.41) by


Di;e .x/ D A.x/Di;e A.x/T ;

where

Di;e .x/ D diag.li;e .x/; ti;e .x/; ni;e .x//:

Since A.x/ is an orthogonal matrix, then it easy to verify that the previous property
(4.40) is equivalent to

min  T D  .x/  4!  .x; /T ŒQ .x; /1 !  .x; / > 0 8x 2 ˝H ; (4.41)
jjjj2 D1

where D  WD Di C De , the diagonal matrix Q is defined by

Q .x; / WD . T Di .x// De .x/ C . T De .x// Di .x/;

and

!.x; / WD . T Di .x// De .x/  . T De .x// Di .x/:

After some computations, the property (4.41) reduces to the condition

i
0 < F .l;t;n e
.x/; l;t;n .x// D

min  T Di .x/  4 . T Di .x/   T De .x//2  T De .x/Q .x; /1 Di .x/ :
jjjj2 D1

(4.42)

Therefore the property (4.42) depends on the six parameters li;e .x/; ti;e .x/; ni;e .x/.
Assuming axially symmetric anisotropy, F depends on the anisotropy ratios
defined by
li le ti
i D ; e D ; rt D :
ti te te
138 4 Reduced Macroscopic Models: The Monodomain and Eikonal Models

In order to visualize the convexity property, we introduce the unit sphere of the
indicatrix function ˚.x; / at x

B˚ .x/ D f 2 Rn W ˚.x; /  1g;

called Frank diagram, and the unit sphere associated with the dual function ˚ 

B˚  .x/ D f  2 Rn W ˚  .x;   /  1g; with ˚  .x;   / D f    W  2 B˚ .x/g;

usually called the Wulff form of the anisotropy in cristal growth and multiphase
problems.
Neglecting the dependence on x, the boundary of B˚ is given by the set


@B˚ D  2 Rn W  D ; jjjj2 D 1 :
˚./

Thanks to [37, Lemma 2.1], the following identity holds

˚  .˚./˚ .// D ˚./;

implying that

  2 @B˚  ”   D ˚ ./ with  2 @B˚ :

Thus
  
 Q./ 
@B˚  D   2 Rn W   D ˚ D ; D ; jjjj2 D 1 :
˚./ ˚./ ˚./

In Figs. 4.2–4.7, we display the Frank diagram and Wulff form using a nom-
inal anisotropy and various estimates of the conductivity coefficients taken from
[100, 123, 131, 420, 421] and related to a normal cardiac tissue. From the graphics,
we see that all these estimates guarantee the convexity of ˚ and ˚  . In Fig. 4.8,
we display the Frank and Wulff diagrams for the extreme case called reciprocal
anisotropy (i D 1=e ), where it is evident that convexity is lost. We remark that
at the concave portion of the Frank diagram corresponds a caustic in the Wulff
diagram. We recall that the conductivity coefficients are average values over a unit
volume of tissue and in particular ti is related to the density of the gap-junctions
in the across fiber direction. Particularly in a pathological tissue as in presence
of myocardial ischemia, the decreasing number of across gap-junctions yields a
reduction of ti . Considering the conductivity values in Fig. 4.6, but with ti reduced
in order to have i D 20, the diagrams displayed in Fig. 4.9 show that the two sets
are not convex, as clearly evidenced by the presence of caustics in the Wulff form,
see Fig. 4.10.
4.4 Dimensional Form of the Reduced Models 139

15

10

150
5

100
0
50

0 −5

−50
−10
−100

−150 −15
−250 −200 −150 −100 −50 0 50 100 150 200 250 −10 −5 0 5 10

Fig. 4.2 Frank and Wulff diagrams for the conductivity values li D 0:2, ti D 0:02, le D 0:8,
te D 0:2 (all in mS cm1 )

40

50 20

−20

−50 −40
−80 −40 0 40 80 −20 −10 0 10 20

Fig. 4.3 Frank and Wulff diagrams for the conductivity values li D 1:74, ti D 0:193, le D
6:52, te D 2:36 (all in mS cm1 ) (From [100])
140 4 Reduced Macroscopic Models: The Monodomain and Eikonal Models

40

50 20

−20

−50 −40
−80 −40 0 40 80 −20 −10 0 10 20

Fig. 4.4 Frank and Wulff diagrams for the conductivity values li D 2:78, ti D 0:263, le D
2:94, te D 1:33 (all in mS cm1 ) (From [421])

40

40 20

20
0

−20
−20

−40 −40
−60 −40 −20 0 20 40 60 −30 −15 0 15 30

Fig. 4.5 Frank and Wulff diagrams for the conductivity values li D 3:43, ti D 0:596, le D
1:17, te D 0:802 (all in mS cm1 ) (From [420])
4.4 Dimensional Form of the Reduced Models 141

40

20
40

20 0

−20
−20

−40 −40
−60 −40 −20 0 20 40 60 −20 −10 0 10 20

Fig. 4.6 Frank and Wulff diagrams for the conductivity values li D 3, ti D 0:315, le D 2,
te D 1:35 (all in mS cm1 ) (From [123])

60

40

30
20

15 0

0 −20

−15 −40

−30 −60
−50 −25 0 25 50 −20 0 20

Fig. 4.7 Frank and Wulff diagrams for the conductivity values li D 1:55, ti D 0:243, le D
2:32, te D 1:04 (all in mS cm1 ) (From [131])
142 4 Reduced Macroscopic Models: The Monodomain and Eikonal Models

10
300

200
5

100

0 0

−100
−5

−200

−300 −10
−300 −200 −100 0 100 200 300 −10 −5 0 5 10

Fig. 4.8 Frank and Wulff diagrams for the conductivity values li D 0:2, ti D 0:02, le D 0:02,
te D 0:2 (all in mS cm1 )

40

20

40
0

20

0 −20

−20

−40 −40
−80 −60 −40 −20 0 20 40 60 80 −20 −10 0 10 20

Fig. 4.9 Frank and Wulff diagrams for the conductivity values li D 3, ti D 0:15, le D 2,
te D 1:35 (all in mS cm1 )
4.5 Numerical Comparison 143

Fig. 4.10 Zoom of the Wulff

diagram displayed in Fig. 4.9 32.5
for the conductivity values
li D 3, ti D 0:15, le D 2,
te D 1:35 (all in mS cm1 )

32

31.5

31
−9.5 −9 −8.5 −8

4.5 Numerical Comparison

The complex interactions between epi-to-endocardial counterclockwise fiber rota-

tion and orthotropic anisotropy described in Sect. 1.3 determine a complex propa-
gation of cardiac excitation. Many experimental works have studied the complex
pattern of epicardial excitation generated by epicardial and intramural ventricular
pacing in animals, see e.g. [12, 184, 506, 507, 516]. The first computer simulations
of the spread of cardiac excitation were based on eikonal equations in order
to understand possible mechanisms of the complex excitation patterns, see e.g.
[102, 112–115, 226, 251, 268, 271, 272, 275, 470, 575].
In this section, we compare the activation time (AT) maps computed by solving
the Eikonal-Diffusion (4.37), the linear Monodomain (4.38) and the Relaxed
Monodomain (4.39) models with those computed by solving the reference Bidomain
model (3.42).
The domain considered is a half truncated ellipsoid described by the parametric
equations
8
< x D a.r/ cos  cos ' 'mi n  '  'max ;
y D b.r/ cos  sin ' mi n    max ;
:
z D c.r/ sin  0  r  1;

where a1 D b1 D 0:2; a2 D b2 D 0:5; c1 D 0:85; c2 D 1, all in cm, 'mi n D  2 ,

'max D 2 , mi n D  38
and max D 8 . The fibers rotate counterclockwise from
epicardium .r D 1/ and endocardium .r D 0/ for a total amount of 120ı , see the
schematic drawing of Figs. 1.4 and 9.28. The orthotropic conductivity coefficients
of the 3D conductivity tensors (3.41) are
144 4 Reduced Macroscopic Models: The Monodomain and Eikonal Models

li D 2:687  103 ; ti D 3:515  104 ; ni D 1:137  104 ;

le D 1:792  103 ; te D 1:141  103 ; ne D 3:839  104 :

All values are given in 1 cm1 . The membrane model considered is the Luo-
Rudy I model [308]. The external stimulus of 200 mA=cm3 lasting 1 ms is
applied in a small volume of about 0:04  0:04  0:02 cm3 at the center of
the epicardial surface. The computed AT is defined at each point in space x
as the unique instant ta during the upstroke phase of the action potential when
v.x; ta / D 50 mV.
The space discretization of all the four models considered is performed by Q1
finite elements in space and semi-implicit finite differences in time; for details see
Chap. 7. The AT considered as a reference solution is computed by solving the
Bidomain model on a mesh of 512  512  128 finite elements, yielding a total
amount of 67 897 602 degree of freedom (dof).
In Table 4.1, we report the l 2 relative errors of the AT computed by solving
the Bidomain, Linear Monodomain and Relaxed Monodomain models on four
increasing Q1 finite element meshes with respect to the reference AT. At the finest
level, the errors of the Linear Monodomain and Relaxed Monodomain models are
about 4:5 and 1:5 %, respectively.
The Eikonal-Diffusion equation is solved on a very coarse mesh of 64  64  16
elements, with a total amount of only 71 825 dof. The relative error in l 2 norm
between the AT computed by the Eikonal-Diffusion equation and the reference AT
is 0.0772. The AT maps reported in Figs. 4.11–4.14 confirm that, even on a such
coarse mesh, the Eikonal-Diffusion equation is able to approximate very accurately
the AT.

Table 4.1 Relative errors in l 2 norm of the activation time computed by solving the Bidomain,
Linear Monodomain and Relaxed Monodomain models on four increasing Q1 finite element
meshes with respect to the reference activation time computed by solving the Bidomain model
on a mesh of 512  512  128 finite elements. The degrees of freedom (dof) are also reported. Note
that in the case of the Bidomain model the dof are doubled because the unknowns are both the
transmembrane potential v and extracellular potential ue . The Bidomain dof for the computation of
the reference solution are 67 897 602
Mesh dof Bidomain Linear monodomain Relaxed monodomain
128  128  32 549 153 0.0301 0.0649 0.0387
192  192  48 1 825 201 0.0198 0.0493 0.0239
256  256  64 4 293 185 0.0129 0.0463 0.0185
384  384  96 14 377 825 0.0047 0.0471 0.0158
4.5 Numerical Comparison 145

a EPI
b EPI
22

20
0.2 0.2 20
18

0 16 0
15
14

−0.2 12 −0.2

10 10
−0.4 −0.4
8

6
−0.6 −0.6
5
4

−0.8 2 −0.8

0
−0.5 0 0.5 −0.5 0 0.5

0.11 22.50 1.00 0.00 23.29 1.00

c EPI
d EPI
22
22
20
0.2 20 0.2
18
18
0 0 16
16
14
14
−0.2 −0.2 12
12
10
10
−0.4 −0.4
8 8

6 6
−0.6 −0.6

4 4

−0.8 2 −0.8 2

−0.5 0 0.5 −0.5 0 0.5

0.10 23.87 1.00 0.11 22.92 1.00

Fig. 4.11 Epicardial activation time isochrones computed by solving the Bidomain model (a), the
Linear Monodomain model (c) and the Relaxed Monodomain model (d) with a mesh of 384 
384  96 finite elements, and the Eikonal-Diffusion equation (b) with a mesh of 64  64  16
finite elements (b). Below each panel are reported the minimum, maximum and step in mV of the
displayed map and the colorbar denotes the range of values of the displayed equipotential lines
146 4 Reduced Macroscopic Models: The Monodomain and Eikonal Models

a MID
b
MID

18 20
0.2 0.2

16

0 0

14
15

−0.2 −0.2
12

−0.4 10 −0.4
10

8
−0.6 −0.6

−0.8 −0.8
5

−0.3 −0.2 −0.1 0 0.1 0.2 0.3 −0.3 −0.2 −0.1 0 0.1 0.2 0.3

4.42 20.29 1.00 4.51 21.96 1.00

c MID
d MID
20
20
0.2 0.2

0 0

15
15

−0.2 −0.2

−0.4 −0.4
10
10

−0.6 −0.6

−0.8 −0.8
5 5
−0.3 −0.2 −0.1 0 0.1 0.2 0.3 −0.3 −0.2 −0.1 0 0.1 0.2 0.3

4.42 21.60 1.00 4.46 20.77 1.00

Fig. 4.12 Mid-myocardial activation time isochrones computed by solving the Bidomain model
(a), the Linear Monodomain model (c) and the Relaxed Monodomain model (d) with a mesh of
384  384  96 finite elements, and the Eikonal-Diffusion equation (b) with a mesh of 64  64  16
finite elements (b). Below each panel are reported the minimum, maximum and step in mV of the
displayed map and the colorbar denotes the range of values of the displayed equipotential lines
4.5 Numerical Comparison 147

a ENDO
b
ENDO
0.3 0.3

0.2 20 0.2 22

0.1 0.1
20
18
0 0

−0.1 −0.1 18
16
−0.2 −0.2
16
−0.3 14 −0.3

−0.4 −0.4 14

12 −0.5
−0.5
12
−0.6 −0.6
10
−0.7 −0.7 10

−0.2 0 0.2 −0.2 −0.1 0 0.1 0.2

8.46 21.75 1.00 8.63 23.72 1.00

c ENDO d ENDO
0.3 0.3
22

0.2 0.2 20

20
0.1 0.1

18
0 0
18

−0.1 −0.1
16
−0.2 16 −0.2

−0.3 −0.3
14
14
−0.4 −0.4

−0.5 −0.5 12
12

−0.6 −0.6

10 10
−0.7 −0.7

−0.2 0 0.2 −0.2 0 0.2

8.57 22.82 1.00 8.59 22.29 1.00

Fig. 4.13 Endocardial activation time isochrones computed by solving the Bidomain model (a),
the Linear Monodomain model (c) and the Relaxed Monodomain model (d) with a mesh of 384 
384  96 finite elements, and the Eikonal-Diffusion equation (b) with a mesh of 64  64  16
finite elements (b). Below each panel are reported the minimum, maximum and step in mV of the
displayed map and the colorbar denotes the range of values of the displayed equipotential lines
148 4 Reduced Macroscopic Models: The Monodomain and Eikonal Models

a 22
b

20
0.2 0.2 20
18

0 16 0
15
14
−0.2 12 −0.2

10 10
−0.4 −0.4
8

6
−0.6 −0.6
5
4

−0.8 2 −0.8

0
0.2 0.4 0.2 0.4
0.11 22.49 1.00 0.00 23.20 1.00

c d
22
22
20
0.2 20 0.2
18
18
16
0 16 0
14
14
−0.2 −0.2 12
12

10 10
−0.4 −0.4
8 8

6 6
−0.6 −0.6
4 4

−0.8 2 −0.8 2

0.2 0.4 0.2 0.4

0.10 23.79 1.00 0.11 22.86 1.00

Fig. 4.14 Transmural activation time isochrones computed by solving the Bidomain model (a),
the Linear Monodomain model (c) and the Relaxed Monodomain model (d) with a mesh of 384 
384  96 finite elements, and the Eikonal-Diffusion equation (b) with a mesh of 64  64  16
finite elements (b). Below each panel are reported the minimum, maximum and step in mV of the
displayed map and the colorbar denotes the range of values of the displayed equipotential lines
Chapter 5
Anisotropic Cardiac Sources

The electrical activity of the heart is revealed and usually detected by measuring the
time–dependent extracellular or extracardiac potential u.x; t/.
Definition 5.1. The electrogram (EG) at a point x inside or outside the myocardium
is defined as the time course of the potential u.x; t/, i.e. with x fixed and t variable.
In electrocardiography what is really measured is a difference between the
potential at an observation point and a reference potential. In the usual practice
body surface, epicardial, intramural and intracavitary EGs are recorded against
the potential average on a given set of points or against the potential at a distant
electrode. The classical reference potential is the average of the potentials at three
different points on the torso (Wilson Central Terminal), see Sect. 1.5. More generally
we can consider as reference potential the average potential on a given surface.
In this study we choose the reference potential which yields zero average on the
insulated boundary of the volume conductor where the heart is embedded (see
[510]).
As described in Sect. 3.3, the representation at macroscopic level of the cardiac
tissue is given by the anisotropic Bidomain [225], constituted by two interpene-
trating anisotropic continua, intra- and extracellular, connected by the distributed
cellular membrane. In the Bidomain model the following potentials are consid-
ered:
• ui .x; t/; ue .x; t/ intra- and extracellular potential;
• v.x; t/ D ui .x; t/  ue .x; t/ transmembrane potential;
• u0 .x; t/ the extracardiac potential;
then the potential u.x; t/ is given by ue .x; t/ inside the myocardium and by u0 .x; t/
outside. The bioelectric sources generating the electric potential field in the heart
(extracellular potential ue ) and outside (extracardiac potential u0 ) are related to
the gradient of the transmembrane potential distribution v.x; t/, as detailed in the
following of this chapter.

© Springer International Publishing Switzerland 2014 149

P. Colli Franzone et al., Mathematical Cardiac Electrophysiology, MS&A 13,
DOI 10.1007/978-3-319-04801-7__5
150 5 Anisotropic Cardiac Sources

We shall present and compare two different equivalent formulations of the

mathematical model related to the generation of electrograms; this model assumes
that the transmembrane potential is known, which is equivalent to know the time
dependent distribution of the electric sources.
In the following we shall denote by ˝H the heart tissue, by ˝0 an extracardiac
medium in contact with ˝H (cavitary blood and/or fluid adjacent to the epicardium),
by Di ; De ; D0 the conductivity tensors in the intra- extracellular and extracardiac
media.

5.1 Differential Formulation of the Potential Field

We recall that the current vector densities associated to the intra, extracellular
and extracardiac potentials are given by Ji D Di rui , Je D De rue and
J0 D D0 ru0 . Assuming that there are no externally applied sources, then the
total field Ji C Je in the cardiac tissue ˝H must be solenoidal, see Sect. 3.3; the
same holds for J0 since there are no sources in ˝0 . Hence we have:

div .Ji C Je / D 0 in ˝H (5.1)

div J0 D 0 in ˝0 : (5.2)

If ˙ D @˝H \ @˝0 denotes the parts of the epi and/or endocardium in contact with
˝0 , then the flux continuity across ˙ implies the transmission conditions:

ue D u0 .Ji C Je /  n D J0  n on ˙:

Setting ˝ D ˝ 0 [ ˝H , H D @˝H n ˙, 0 D @˝0 n ˙,  D @˝ D H [ 0 ,

and assuming that the total volume ˝ is insulated, we have:

.Ji C Je /  n D 0 on H J0  n D 0 on 0 :

The unit normal n is outward to ˝H or ˝ according as we consider the boundary

condition on ˙ or on  ; hence on ˙ the normal n points outside ˝H (see Fig. 5.1).
In the following, we set D D Di C De (“bulk” conductivity tensor),
 
D in ˝H ue .x; t/ in x 2 ˝H
DO D and u.x; t/ D
D0 in ˝0 u0 .x; t/ in x 2 ˝0

and we define the equivalent cardiac source for the extracardiac and extracellular
potentials by

Jv .x; t/ D Di rv.x; t/; (5.3)

5.1 Differential Formulation of the Potential Field 151

Fig. 5.1 Schematic

representation of the n
domain ˝

n
n Η
ΓH
Ω0
Γ
0
Σ

where v.x; t/ is the transmembrane potential. Then problem (5.1)–(5.2) can be

written as:
8 
ˆ
ˆ O div Jv .x; t/ in ˝H
ˆ
ˆ div Dru.x; t/ D
ˆ
< 0 in ˝0
ŒŒ u.x; t/ ˙ D 0 O
ŒŒ .Dru.x; t//  n ˙ D Jv .x; t/  n (5.4)
ˆ
ˆ
ˆ
ˆ 0.D ru.x; t//  n D 0 on  0
:̂
.Dru.x; t//  n D Jv .x; t/  n on H

where ŒŒ ' ˙ denotes the jump of ' through ˙, i.e. ŒŒ ' ˙ D '˙ C  'j˙  with
'j ˙ the traces taken on the positive and negative side of ˙ with respect to the
˙
oriented normal. Therefore problem (5.4) provides the extracellular or extracardiac
potential u.x; t/ from the knowledge of v.x; t/.
In the remaind of this Chapter we will assume:
• ˝ and ˝H bounded Lipschitz open set of R3 ;
• The conductivity tensors Di .x/; De .x/ 2 C 1 .˝H /;
• 8t 2 Œ0; T  the transmembrane potential v.x; t/ 2 W 2;p .˝/ with p > 3.
These regularity properties imply that Jv 2 W 1;p .˝H / and div Jv 2 Lp .˝H /,
hence n  Jvj@˝ 2 W 1=p;p .@˝H / and the first Green formula holds, see e.g. [305].
H
The conormal derivative, in the boundary value problem (5.4), exhibits a jump
discontinuity on ˙. Defining the following bilinear form
Z
a.'; / D O
.D.x/r'/  r d x; 8'; 2 W 1;2 .˝/ (5.5)
˝

the variational formulation of (5.4) is:

8t 2 Œ0; T  find u.x; t/ 2 W 1;2 .˝/ such that
Z
a.u; / D Jv .x; t/  r .x/ d x 8 2 W 1;2 .˝/: (5.6)
˝H
152 5 Anisotropic Cardiac Sources

We remark that 8t 2 Œ0; T  problem (5.6) admits a solution u.x; t/ 2 W 1;2 .˝/
unique up to an additive constant. This constant depends on t and it is related to the
reference potential chosen.
It is easy to verify, applying the Green formula, that the solution of (5.6) satisfies
problem (5.4) in the sense of distributions and of the trace theorems (see e.g. [305]).

5.2 Integral Formulation of the Potential Field

Usually EGs are recorded at a limited number of points x, from 12 to 24 derivations,

see Sect. 1.5. Consequently the differential formulation (5.4) is not computationally
convenient, since for any t 2 Œ0; T  the solution of the elliptic problem (5.4) is
needed and yields u.x; t/ for all nodes of the mesh (at the given time instant t),
while we are interested only in the value of u.x; t/ in few points x on Œ0; T . An
integral formulation [197,198] of the potential u.x; t/ turns out to be a more efficient
tool, because this formulation requires the solution of only a number of problems
given by the chosen set of points x, i.e. x appears as a parameter. This integral
representation is given by:
Z Z
u.x; t/  uref D Jv .x; t/  r d D  .Di rv/  r d : (5.7)
˝H ˝H

This formulation gives a potential difference with respect to a reference potential.

In this study, we consider as reference potential the one given by the potential
at a reference point x0 or by the average on the insulated boundary  . The
representation (5.7) uses the function .I x/ defined as the potential field in the
medium ˝ with conductivity tensor D. O This function is generated by a unit current
source at lead x and a unit current sink at lead x0 or a current sink uniformly
distributed on the surface  with density 1=j j according to the first or the second
choice of the reference potential, see [197, 214, 570]. More precisely, the potential
field , usually called lead field in electrocardiographic literature, is the Green
function associated to the following boundary value problem of Neumann type:
8
ˆ
< div  Dr
O  D ˝ in ˝
(5.8)
:̂ Dr
O   n D  on ;

where, for the first choice of reference potential with x0 2 ˝, we define:

˝ D ı.  x/  ı.  x0 /;  D 0; when x 2 ˝

(5.9)
˝ D ı.  x0 /;  D ı.  x/; when x 2 
5.2 Integral Formulation of the Potential Field 153

and, for the second choice of reference potential, we have:

1
˝ D ı.  x/;  D  when x 2 ˝
j j (5.10)
1
˝ D 0;  D ı.  x/  when x 2 ;
j j

with ı.  x/ denoting, as usual, the Dirac ı-function at point x.

Formal derivation of (5.7). Proceeding formally we apply the second Green
formula to the couple .u; / in ˝0 and in ˝H . Adding these two relations we obtain:
Z h i
O
u div Dr  O
div Dru d
˝0 [˝H
Z Z
D O
u .Dr /  n d  O
.Dru/  n d
 
Z Z
C O
ŒŒ .Dr /  n ˙ u d  O
ŒŒ .Dru/  n ˙ d : (5.11)
˙ ˙

O
Taking into account that, if x; x0 … ˙, then ŒŒ .Dr /  n ˙ D 0 otherwise, if
O
x 2 ˙, ŒŒ .Dr /  n ˙ D ı.  x/, from (5.11) we have:
Z Z
u.x; t/ C O
div Dru d  O
.Dru/  n d
˝0 [˝H 
Z
 O
ŒŒ .Dru/  n ˙ d  uref D 0;
˙
Z
1
where uref D u.x0 ; t/ or uref D u d according as .I x/ satisfies (5.9)
j j 
or (5.10). Hence taking into account (5.4) and denoting by w.x; t/ the potential
difference with respect to the chosen reference potential, i.e. w.x; t/ D u.x; t/uref ,
it follows:
Z Z
w.x; t/ D  div Jv d  C Jv  n d
˝H @˝H

and applying the first Green formula we obtain the integral representation:
Z Z
w.x; t/ D Jv  r d D  .Di rv/  r d : (5.12)
˝H ˝H

The previous derivation can be easily justified, in the framework of Sobolev spaces,
when DO is assumed regular in ˝. When DO exhibits, as in real cases, a jump
discontinuity on the interface ˙ then the Green function is solution of a boundary
154 5 Anisotropic Cardiac Sources

value problem with discontinuous coefficients and cannot be uniquely defined in the
sense of distributions.
Rigorous mathematical derivation of (5.7). To rigorously introduce Green
functions associated to the previous Neumann problems we consider a notion of
weak solution similar to the one proposed by Stampacchia for the Dirichlet problem
(see [343, 495]).
We denote by A D div Dr O  and by @A D DrO   n with D.˝/ and D. / the
set of measures with bounded variation and support on ˝ and on  respectively.
Problem (5.8) is a non homogeneous Neumann problem of the type:

A ! D ˝ in ˝
(5.13)
@A ! D  on 

with data as measures given by (5.9) and (5.10). Extending the classical approach
of Stampacchia [495], we introduce the following notion of weak formulation
of (5.13):
Definition 5.2. Given Z˝ ;  measures
Z of bounded variation on ˝ and 
respectively, such that d˝ C d D 0, a function ! 2 L1 .˝/=R is a
˝ 
weak solution of problem (5.13) if
Z Z Z
!A ' D ' d˝ C ' d (5.14)
˝ ˝ 

N \ W 1;2 .˝/; A ' 2 C 0 .˝/;

8' 2 C 0 .˝/ N @A 'j D 0:
@˝H

We remark that the set of the test functions is not empty since, due to Corollary 7.8
in [343], the following regularity result holds:
Theorem 5.1. LetR ˝ be a domain with Lipschitz continuous boundary, f0 2
Lp .˝/ such that ˝ f0 dx D 0 and F 2 .Lp .˝//3 with p > 3. If z 2 W 1;2 .˝/=R
is the unique solution of
Z Z
a.z; '/ D f0 ' dx C r'  F dx 8' 2 W 1;2 .˝/;
˝ ˝

then
 
max jzj  C jjf0 jjLp C jjFjj.Lp /3 :
˝N

Taking into account that the operator A is self-adjoint, by duality of this

regularity result we obtain the existence and uniqueness of the solution ! of
0
problem (5.13) with ! 2 W 1;p .˝/ and p 0 < 3=2 (p 0 conjugate number of p).
Therefore, defining the Green function .I x/ of problem (5.8) as the weak
N
solution of (5.13) with ˝ and  given by (5.9) and (5.10), we have 8x 2 ˝:
5.2 Integral Formulation of the Potential Field 155

Z
.I x/ A ' d  D '.x/  'ref (5.15)
˝

N \ W 1;2 .˝/; A ' 2 C 0 .˝/;

8' 2 C 0 .˝/ N @A 'j D 0;
@˝H

R
where 'ref D '.x0 / or 'ref D j1 j  ' d ; moreover, since A is self-adjoint,
.I x/ D .xI /. R
N with
For g 2 C 0 .˝/ 0 N
˝ g dx D 0, we consider ' 2 C .˝/ \ W
1;2
.˝/ solution
of

A ' D g in ˝
(5.16)
@A ' D 0 on :

From (5.15) we have the following integral representation for the solution of
N
problem (5.16) 8x 2 ˝:
Z
'.x/  'ref D .I x/ g./ d : (5.17)
˝

Using Fubini’s theorem and formula (5.17) with 'ref D 0, it follows that:
Z Z Z Z Z 
' d˝ C ' d D g.x/ .xI / d˝ C .xI / d dx
˝  ˝ ˝ 

and, comparing with (5.14), we obtain that the solution !.x/ of problem (5.13),
apart from an additive constant !ref , is given by:
Z Z
!.x/ D .xI / d˝ C .xI / d C !ref : (5.18)
˝ 

Going back to the variational problem (5.6) and defining the potential u.x; t/, it is
easy to verify that u.x; t/ is also a weak solution of problem (5.13) for the data
N
˝ ;  defined by the functional on C 0 .˝/
Z Z Z
' 7!  ' div Jv C ' Jv  n C ' Jv  n:
˝H ˙ H

0
From the representation formula (5.18), since 2 W 1;p .˝/ with p 0 < 3=2,
applying the Green formula, u.x; t/, apart from an additive constant, is given by
Z
u.x; t/  uref D Jv  r d : (5.19)
˝H
156 5 Anisotropic Cardiac Sources

This formula coincides with (5.12), which represents the solution having zero
reference potential and holds 8x 2 ˝N since, from the regularity Theorem 5.1, we
N In conclusion the following Proposition holds:
have that u.x; t/ 2 C 0 .˝/.
Proposition 5.1. The solution of (5.6) with zero reference potential is given by:
Z
w.x; t/ D Jv  r .I x/ d  N 8t 2 Œ0; T ;
8x 2 ˝;
˝H

N
where, 8x 2 ˝, .I x/ is solution of problem (5.15).
In the following, we denote by

w0 .x; t/ D w.x; t/ for x 2 ˝ n ˝H

we .x; t/ D w.x; t/ for x 2 ˝H :

Finally, we observe that in both the differential and integral formulations of the
Bidomain representation of the multicellular cardiac tissue, the macroscopic cardiac
bioelectric sources generating the extracellular and extracardiac potentials consist of
the volume dipole density Jv .

5.3 Approximate Representation of Cardiac Sources

During the excitation phase of the heartbeat, the transmembrane potential v exhibits
a steep transition layer propagating throughout the myocardium with a thickness
of about 1 mm. At every point of the cardiac domain, this upstroke phase lasts
about 1 ms. Therefore, the accurate simulation of the excitation process requires
discretization methods based on small space and time steps (of the order of 0.1 mm
and 0.01 ms). This fact constrained 3–D simulations, see e.g. [102, 104, 226, 544]
and the recent survey [545], and in order to reduce the computational costs for the
simulations of potential maps and electrocardiograms, approximate models of the
cardiac sources have been developed and used.
The evolving thin layer of cells undergoing the depolarization upstroke and
sweeping the whole ventricle is called the excitation wavefront and the excitation
wavefront surface is defined by means of the so called excitation or activation time
of the cell. During the excitation phase of the heart beat, each cell undergoes a
fast upstroke of the transmembrane potential, starting from the resting state vr ,
increasing monotonically during the depolarization upstroke and reaching finally
a depolarized state vd .
Therefore, at any point x 2 ˝H there exists a unique instant '.x/, called the
activation time, such that

vr C vd
v.x; '.x// D :
2
5.3 Approximate Representation of Cardiac Sources 157

Hence, at time t, the set

Hd .t/ D fx 2 ˝H W '.x/  tg

denotes the depolarized volume and we call its boundary

@Hd .t/ D St D fx 2 ˝H W '.x/ D tg

the excitation wavefront.

5.3.1 Heart Surface Source Model

In this subsection, we assume that the observation point x 2 ˝ n ˝H and the

e
two conducting media have equal anisotropy ratios, i.e. l;t;n
i D constant, which
l;t;n
implies De D Di . By Green’s formula
Z Z
.r v/ Di r .I x/ d  D
T
v nT Di r .I x/ d
˝H @H
Z
 v div.Di r .I x// d ;
˝H

where @H is the surface (endocardial and epicardial) bounding the ventricular

myocardium ˝H . Due to (5.8), because x 2 ˝ n ˝H , the second term above is
zero, in fact in ˝H we have

0 D div..Di CDe /r / D div..1C /Di r / D .1C / div.Di r /: (5.20)

Therefore, from Proposition 5.1, we obtain for x 2 ˝ n ˝H

Z
w0 .x; t/ D  v.; t/ nT Di r .I x/ d : (5.21)
@H

Thus, under the assumption of equal anisotropy ratio, the cardiac sources are
represented as a dipole layer distributed on the epi-endocardial heart surface. For
this reason, this model is called heart surface source model, see [197–199,329,570].
The dipole distribution has dipolar moment proportional to the transmembrane
potential v.; t/, with oblique direction parallel to Di n and oriented outward the
cardiac tissue since nT Di n > 0.
Remark 5.1. If the transmembrane potential v is constant on the heart surface,
then (5.21) predicts a zero potential w0 at any point x outside the cardiac tissue.
Indeed, for v D c constant, due to (5.20) we have
158 5 Anisotropic Cardiac Sources

Z Z
v nT Di r .I x/ d D c nT Di r .I x/ d
@H @H
Z
Dc div.Di r .I x// d x D 0: (5.22)
˝H

It is reasonable to assume that the cardiac fibers are tangential to the cardiac surface,
i.e. al ; an ?n on @H . Then we have Di n D ti n, hence
Z
w0 .x; t/ D  ti v.; t/nT r .I x/ d :
@H

This model describes the evolution of w0 during the entire heartbeat. Given the time
interval Œ0 T , the average of the extracardiac potential w0 can be written as
Z Z Z
1 T
1 T 
w0 .x; t/dt D  v.; t/dt ti nT r .I x/ d :
T 0 @H T 0

This formula can be used to obtain the following two interesting expressions for
time averages of the extracardiac potential.
(a) QRS – T average. During the QRS – T interval, the average transmembrane
potential, shifted with respect to the resting value vr ,
Z T
1
v./ D .v.; t/  vr /dt;
T 0

where T is the time of the T-wave ending, is the normalized area below the
action potential time course. Hence we can express the QRS – T average of the
extracardiac potential
Z T
1
w0 .x/ D w0 .x; t/dt
T 0

as a heart surface integral involving the normalized area v./ below the
transmembrane potential v,
Z
w0 .x/ D  v./ti nT r .I x/ d :
@H

(b) QRS average. During the QRS interval, the action potential is in the depolar-
ization phase and it can be approximated with an activation profile v.; t/ D
A.t  './/, where './ is the activation time at point . Then the QRS average
Z Td
1
w0 .x/ D w0 .x; t/dt;
Td 0
5.3 Approximate Representation of Cardiac Sources 159

where Td is the ending time of the QRS interval, has the representation
Z Z Td 
1
w0 .x/ D  A.t  './/dt ti nT r .I x/ d : (5.23)
Td @H 0

If we further approximate the activation profile with a Heaviside step function

H from the resting value of vr to the depolarized value vd of v

A.t  '.// D H .t  './/.vd  vr / C vr ;

Z Td
then A.t  './/dt D vd Td  .vd  vr /'./ and the QRS average (5.23)
0
becomes
Z
vd  vr
w0 ./ D './ti nT r .I x/ d ;
Td @H
Z
since by (5.22) it holds cti nT r .I x/ d D 0 for any constant c. Thus,
@H
we can express the QRS average of the extracardiac potential as the heart surface
integral involving the activation time.

5.3.2 Oblique Dipole Source Model

In this subsection, we assume general anisotropy of the intra- and extracellular

media, but we focus only on the depolarization phase. Then, if the heart tissue is
initially at rest and all cells undergo the same action potential profile A./ during
the upstroke, the transmembrane potential can be approximated as

v.x; t/ D A.t  '.x//:

From the representation formula in Proposition 5.1, we then have

Z
w.x; t/ D A0 .t  '.x//.r './/T Di r .I x/ d :
˝H

Applying the co-area formula, see [5, 167],

Z Z Z 
g./jr './j d  D g./d d
˝H R ' 1 . /DS
160 5 Anisotropic Cardiac Sources

for a non-negative measurable function g and a Lipschitz function ', and setting
r './
n ./ D , we have
jr './j
Z Z
w.x; t/ D A0 .t  '.x//n ./T Di r .I x/d d
R ' 1 . /DS
Z Z 
D A0 .t  / .Di n /T r .I x/d d : (5.24)
R S

By approximating the action potential profile during the depolarization phase with
the jump discontinuity described by

A.t  / D H .  t/.vd  vr / C vr ;

where H is the Heaviside step function, then A0 .t  / D .vd  vr /ı.  t/. In

the limit when tends to t, we then obtain from (5.24) the following representation
formula for x 62 St
Z Z 
w.x; t/ D .vd  vr / ı.  t/ .Di n /T r .I x/d d
R S
Z
D .vd  vr / .Di n /T r .I x/d d : (5.25)
St

In this formula, the cardiac excitation sources are represented as a dipole distribution
supported on the excitation wavefront St , having oblique directions parallel to Di n
and with dipolar moment per unit surface area given by .vd  vr /kDi n k2 .
Assuming an infinite extracardiac medium and the same constant conductivity
for the extracardiac and bulk tissue, i.e. D0 D Di C De D 0 I , then the Green
function becomes
1 1
.I x/ D ;
40 j  xj

so the extracellular potential w in (5.25) can be written as

Z
.vd  vr / 1
w.x; t/ D .Di n /T r d :
40 St j  xj

If we assume further that Di D i I , then

Z
1 @
w.x; t/ D i .vd  vr / j  xj1 d ; (5.26)
40 St @n
5.3 Approximate Representation of Cardiac Sources 161

which coincides with the representation of the cardiac excitation source in terms of
normal dipoles uniformly distributed over the excitation surface St , also known as
uniform double layer, see [535, Ch. 6]. If we define the moment per unit surface

p D i .vd  vr /;

then (5.26) becomes

Z
1 @
w.x; t/ D p j  xj1 d :
40 St @n

Setting

@
d!.x; / D j  xj1 d D nT r j  xj1 d ;
@n

and defining the so-called solid angle

Z Z nT .x  /
!.x/ D d!.x; / D  d ; (5.27)
St St jx  j3

we obtain the representation

1
w.x; t/ D p!.x/: (5.28)
40

This formula shows that the magnitude of the potential at a point x is proportional
to the solid angle !.x/ under which the activation wavefront is seen from x. The
solid angle formula (5.27) and the associated potential representation (5.28) yield
immediately some important consequences:
(a) The potential field external to a closed surface St is zero;
(b) The potential field due to an open surface St is completely defined by its
boundary (rim);
(c) The potential is positive at a point x located “in front” of the surface St , i.e.
where nT .  x/ < 0, since there the solid angle is negative; viceversa, the
potential is negative at a point x located “behind” the surface St , i.e. where
nT .  x/ > 0, since there the solid angle is positive.
These properties can be applied to the simple case of a single depolarizing wave-
front spreading through the myocardium to obtain an approximate explanation of the
ECG inflections. Indeed, by (c) a depolarization front propagating toward a positive
electrode located at point x produces a positive signal, while a depolarization front
propagating away from a positive electrode located at point x produces a negative
signal. Moreover, if an excitation wavefront St1 at time t1 propagates to a larger
wavefront St2 at time t2 > t1 toward a positive electrode, then the potential increases,
162 5 Anisotropic Cardiac Sources

w.x; t1 / < w.x; t2 /, while if the propagation is away from the electrode then the
potential decreases, w.x; t1 / > w.x; t2 /. The previous prediction takes not into
account the role of the anisotropic structure of the myocardium, which is instead
incorporated in the following Sects. 5.4 and 5.5.

5.4 Cardiac Source Splitting

We present now a useful split form of the potential field, previously introduced in
[116–118, 121, 510] to investigate some features of the potential u.x; t/. It is used
also as a tool for a more accurate computation of u.x; t/. We focus first on the case,
frequently considered, of a cardiac tissue which is characterized by conductivity
tensors that are axially symmetric with respect to the local fiber direction; the
extension of the split form to tensors with no axial symmetry can be found at the
end of this paragraph.

5.4.1 Axially Symmetric Media

If a D a.x/ is the unit vector defining locally the fiber direction and li;e , ti;e are
the conductivity coefficients along and across fiber in the intra- and extracellular
medium, then the assumption of axial symmetry implies that the conductivity
tensors Di ; De can be defined as

Di;e D ti;e I C .li;e  ti;e /a ˝ a:

We recall that the bulk conductivity tensor D is given by

D D Di C De D t I C .l  t /a ˝ a;

where t D ti C te and l D li C le . The conductivity coefficients may depend
on the position x according to the local state, normal or diseased, of the various part
of the myocardial tissue. It is easy to verify the following identity:

Di D ˛ D C ˇ a ˝ a (5.29)
l
with ˛ D ti =t , ˇ D li  ti , l D li C le , t D ti C te . In fact, from the
t
definition of the bulk conductivity tensor D, we have that

t I D D  .l  t /a ˝ a:
5.4 Cardiac Source Splitting 163

Substituting it into the definition of the intracellular conductivity tensor Di , we get

ti
Di D .D  .l  t /a ˝ a/ C .li  ti /a ˝ a
t  
i i
D t D C li  ti  t .l  t / a ˝ a
t  t
ti i l
D D C l  t
i
a ˝ a;
t t

thus identity (5.29) holds true.

The split (5.29) of Di leads to the corresponding split of the “impressed” or
“source” current density Jv D Di rv yielding

Jv D Di rv D Ja C Jc (5.30)

where

Ja D ˇ .a ˝ a/ rv and Jc D ˛Drv: (5.31)

We call Ja ; Jc axial and conormal current densities since they define two distribu-
tions of dipolar current sources in the heart with dipole axes parallel respectively to
the fiber direction a and to the vector Drv.
Setting
Z
1
w.x; t/ D u.x; t/  u.x; t/ d
j j 

and using the split form (5.30) of Jv , relation (5.12) can be written as:

w.x; t/ D wa .x; t/ C wc .x; t/ (5.32)

with
Z Z
wa .x; t/ D Ja  r d and wc .x; t/ D Jc  r d (5.33)
˝H ˝H

axial and conormal potential components. We remark that wc and wa , similarly to

w, have zero average on  .
We can perform a further split of the field component wc ; in fact since ˛rv D
r.˛v/  vr˛ it follows that:

Jc D ˛Drv D Dr.˛v/ C vDr˛

164 5 Anisotropic Cardiac Sources

hence, if the observation point x … ˙:

Z Z
wc .x; t/ D Jc  r d D  ˛ .Drv/  r d
˝H ˝H
Z Z
D .Dr.˛v//  r d C v .Dr˛/  r d
˝H ˝H
Z Z
D ˛v .Dr /  n d C ˛v div Dr d
@H ˝H
Z
C v .Dr˛/  r d :
˝H

Noting that is solution of problem (5.8) with (5.10) if x 2  , we obtain:

wc .x; t/ D uhs .x; t/ C ud .t/ C uj .x; t/ C unh .x; t/; (5.34)

where
Z
uhs .x; t/ D  ˛v .Dr /  n d heart surface component (5.35)
˙
Z
1
ud .t/ D ˛v d drift component (5.36)
j j H

˛v.x; t/ x 2 ˝H
uj .x; t/ D jump component (5.37)
0 x 2 ˝0
Z
unh .x; t/ D v .Dr˛/  r d  non-homogeneous component. (5.38)
˝H

Note that ud .t/ is independent of the observation point x hence it acts as a drift
component; moreover the component unh .x; t/ gives a contribution only for tissue
with non homogeneous anisotropy, i.e. when ˛ is not constant.
The above decomposition of the conormal component holds if the observation
point x does not lie on the non insulated boundary ˙ (e.g. the endocardium in
contact with intracavitary blood). If x 2 ˙ then the splitting of the conormal
component must account for the discontinuity of the two components uhs and uj
on ˙. More precisely applying classical limit formulae for the generalized double-
layer potentials on ˙ (see formula 15.3 of Theorem 15.II in [332]) we obtain the
following trace and jump relationships:
Z
1
u˙
hs D ˛vj  ˛v .Dr /  n d ; ŒŒ uhs ˙ D ˛ vj˙
2 ˙ ˙
5.4 Cardiac Source Splitting 165

where u˙ hs denotes the trace of the function on the positive and negative side of ˙
with respect to the oriented normal n. We remark that wc .x; t/ D uhs .x; t/ C ud .t/ C
uj .x; t/ is continuous across ˙ and has the following expression:
Z
1
x 2 ˙; wc .x; t/ D  ˛vj˙  ˛v .Dr /  n d C ud .t/ C uj .x; t/ C unh .x; t/:
2 ˙

In the special case of ˝H fully insulated (i.e. H D @H and ˙ D ;) and ˛ constant

(i.e. r˛ D 0) the conormal component reduces to:
Z
˛
wc .x; t/ D ˛v.x; t/ C v d (5.39)
j j 

5.4.2 Orthotropic Media

Recent findings suggest that the cardiac tissue anisotropy is not symmetric across
fibers, but, due to the presence of myocardial sheets (or laminae) (see [352]), three
material principal axes can be identified, thus it presents an orthotropic structure.
Actually, the split form of the potential w.x; t/ can be extended to the case of no
axial symmetry for the conductivity tensors Di ; De . In fact we can assume that Di;e
have three distinct eigenvalues li;e ; ti;e and ni;e with the same principal axes given
by the set of orthonormal eigenvectors ak ; k D l; t; n with ak  aj D ıjk and al D a,
i;e
i.e. tangent locally to the fiber. Setting A D .al ; at ; an /, Ddiag D diag.li;e ; ti;e ; ni;e /
and using the property that AAT D AT A D I , where AT and I denote respectively
the transpose of A and the identity matrix, we obtain:

Di;e D ADi;e T
diag A :

The equivalent expression of Di;e is

Di;e D ti;e I C .li;e  ti;e /al ˝ al C .ni;e  ti;e /an ˝ an ;

and as a by-product the bulk conductivity tensor is given by

D D t I C .l  t /al ˝ al C .n  t /an ˝ an ;

with l;t;n D l;t;n

i
C l;t;n
e
. It is easy to verify that, if the axial symmetry is not
assumed (i.e. distinct eigenvalues) then

Di D ˛D C ˇal ˝ al C .I  al ˝ al /; (5.40)

with ˛ D .ni  ti /=.n  t /; ˇ D li  ˛l ; D ti  ˛t . In fact, given ˛ 2 R,
we have
166 5 Anisotropic Cardiac Sources

Di  ˛D D .ti  ˛t /I
C .li  ti  ˛.l  t //al ˝ al
C .ni  ti  ˛.n  t //an ˝ an :

Choosing ˛ D .ni  ti /=.n  t /, the last term vanishes and we get the split
form (5.40). We remark that I  al ˝ al is the orthogonal projection matrix on a
plane perpendicular to the fiber direction al .
If ni;e D i;e ti;e , then the media (i) and (e) tend to become axially symmetric
for i;e ! 1 with the limit values ni;e D ti;e . Setting i;e D 1  i;e , if i = e ! 1
l
as i;e ! 1 then it is easy to verify that ˛; ˇ; tend to ti =t , li  ti ,0
t
respectively, which are the values obtained directly under the assumption of axial
symmetry.
In the orthotropic case, the split of the conductivity tensor Di leads to a
corresponding split of the impressed current Jv yielding:

Jv D Jc C Ja C Jo

with Ja and Jc defined as before with different ˛ and ˇ whereas Jo D  .I  al ˝

al /rv is a new current density of dipoles transverse to the fibers.

5.5 Interpretation of the Field Components

From the relation between the field components and their sources we establish some
correlations between pattern features displayed by the full potential and some struc-
ture of the underlying cardiac sources. The potential components deduced from the
potential split depend on v.x; t/, which is a quantity of direct electrophysiological
significance since it characterizes the electric state of the myocardial cells. The
analysis of these components can shed light on the properties of the full potential
field. In the following, we focus only on the axisymmetric components of the
splitting.
Axial component (wa ): the potential wa can be related to the source Ja by
Z Z
wa .x; t/ D JTa r d  D  ˇ.aTl rv/aTl r d :
˝H ˝H

This expression shows that the potential wa is generated by volume sources repre-
sented by a current dipole density with dipole moment per unit volume proportional
to the vector .aTl rv/al . Therefore the dipole current sources are parallel to the local
fiber direction al and we call wa axial potential. Since the fiber architecture defines
the direction of the current source flow, we expect that structural properties like epi–
endocardial fiber rotation will be the major determinants affecting the axial potential
pattern.
5.5 Interpretation of the Field Components 167

When the excitation wavefront propagates through the heart muscle, it occupies
a thin layer of myocardium, 1 or 2 mm thick, that is passing from the resting to the
excited state. Within this layer, the distribution of axial sources has greater dipole
strength per unit volume in those regions in which the wavefront is spreading mainly
along fibers, i.e. where rv is nearly parallel to al .
Jump component (uj ): we recall that the resting value vr has been set to zero
so that in our model v measures the deviation of the transmembrane potential from
the resting state. We remark that the potential component uj .x; t/ D ˛v.x; t/
is zero in ˝0 , whereas in ˝H it is proportional to the spatial distribution of the
transmembrane potential. Since v, in our model, displays an upstroke in a thin layer
of tissue containing the wavefront and it is practically constant everywhere else, this
field component does not generate an appreciable current flow at distance from the
front. Moreover in the fully activated region the effect of uj on the full potential u
reduces only to a shift proportional to the jump of the transmembrane potential from
the resting to the plateau value. For this reason we call uj jump component.
Heart surface component (uhs ): the potential component uhs is given by
Z
uhs .x; t/ D .˛vDn/T r d : (5.41)
˙

We remark that uhs contributes to the full potential u only when the heart surface is
in contact with an extracardiac conducting medium (i.e. the non-insulated cardiac
interface ˙ is not empty) and depends on the trace on ˙ of the transmembrane
potential v. Since the sources lie on the non-insulated heart surface ˙, we refer to
this field component, as usual in the literature (see e.g. [197, 198, 483]), as heart
surface component.
The differential formulation shows that uhs exhibits a jump through ˙ equal to
˛v and the moment of the dipole layer source on ˙ is proportional to this jump; on
the unexcited parts of ˙, i.e. where v is practically zero, this jump vanishes.
In order to describe in a different and more meaningful way the relationship
between the field component uhs and its heart surface source, we make the following
simplifying assumptions: uniform anisotropy on @H , homogeneity on @˝0 (i.e. li;e ,
ti;e , 0 constant on ˙) and the fibers tangent to the epi- and endocardial surfaces
(i.e. aTl n D 0 on ˙). Hence in particular on ˙ we have Dn D t I , D0 n D 0 I
implying Dn D D0 n with D t =0 ; consequently uhs can be viewed as the
potential generated by a source term Jhs satisfying in the sense of distribution (see
[110, 305]) the equation

div Druhs D div Jhs in ˝; with Jhs D ti vı˙ n˙ ; (5.42)

where D D D in ˝H and D D D0 in ˝0 , n˙ is the unit normal to ˙ and ı˙

the Dirac measure on ˙. Therefore uhs can be viewed as the field generated by a
normal dipole layer on ˙ with moment ti v. If the excitation wavefront is far from
˙ then v and consequently uhs are practically zero. Otherwise, denoting by ˙d the
intersection on ˙ of the activated region, since v, apart a thin strip near @˙d , is
168 5 Anisotropic Cardiac Sources

very close to vJ on ˙d and ˛ is constant, it follows that the normal dipole layer is
practically uniform on ˙d . Thus, the most remarkable property of uhs is that it can
be well approximated by means of the classical model, i.e. the potential generated
by a normal and uniform dipole layer on ˙d in the conducting medium described
O
by the tensor D.
Non-homogeneous component (unh ): the field component unh is given by
Z
unh .x; t/ D v.Dr˛/T r d :
˝H

Component unh makes no contribution to the potential distribution when ˛ is

constant, i.e. when the anisotropy ratio te =ti is constant characterizing a cardiac
tissue with a transverse ‘uniform’ anisotropy. Hence in normal cardiac tissue, this
field component should not affect the potential distribution. This is due to the fact
that the normal tissue can be modeled assuming a uniform anisotropy. This field
component should significantly affect the full potential pattern only in presence
of pathological myocardial tissue with inhomogeneous transverse conductivity,
as occurs for instance in the surviving tissue near a chronic infarct. Due to this
interpretation we call unh the non-homogeneous tissue component.
In the previous derivation of the split form we have implicitly assumed that ˛.x/
is a smooth function; when ˛.x/ presents a discontinuity interface , a new surface
integral term appears given by
Z
v ŒŒ ˛  nT Dr d ; (5.43)


where ŒŒ ˛  measures the jump through the interface , i.e. ŒŒ ˛  D ˛ C  ˛  ,

where ˛ ˙ are the traces taken on the positive and negative side of  with respect
to the oriented normal. A sudden variation of ˛ could model the transition from a
region with normal conduction to a pathological one with depressed conduction.
Moreover if ˛ is piecewise constant in two contiguous tissue volumes, the unh
component reduces to the surface integral (5.43), which is a potential generated
by a secondary source distributed on the discontinuity interface  having a current
dipole moment density v ŒŒ ˛  Dn, hence with dipoles parallel to Dn.
This component should affect the electrocardiogram (ECG) waveform, recorded
from the transition regions, in a manner similar to the one described in [45, 297].

5.6 The Limit Case: Oblique Dipole Layer Model

Let us denote by Hd .t/ the depolarized region, by Hr .t/ the resting region and
by ˙d and ˙r their intersections with ˙ (one of the two can be void). The two
regions Hd .t/ and Hr .t/, are characterized by an activation time '.x/ whose value
is respectively smaller or greater than the fixed time instant t. Remembering that
5.6 The Limit Case: Oblique Dipole Layer Model 169

the transmembrane potential v is approximated by v.x; t/ D A.t  '.x//, the

upstroke phase is characterized by the steepest part of the profile of the action
potential A and by a corresponding thin layer of tissue, containing the excitation
wavefront, across which the fast variation of the transmembrane potential occurs.
Focusing on the potential pattern at a distance from the excitation wavefront,
the details of the upstroke phase can be neglected and it is natural to consider
the limit case obtained when A . / approaches the step function vJ H . /, where
H is the Heaviside function. Remembering that v measures the deviation of the
transmembrane potential from the resting state, then the jump component uj D ˛v
in ˝H approaches ˛vJ H .t  '.x//, i.e. it is constant in the activated and
resting regions and undergoes a jump through the wavefront surface St D fx 2
˝H ; '.x/ D tg. Hence the jump component does not contribute to the extracellular
current. At the same time rv.x; t/ tends to vJ nSt ıSt in the distributional sense,
where ıSt is the Dirac measure on St and nSt is the unit normal to St directed toward
the resting tissue. Thus using Eq. (5.19), in the limit case, we obtain
Z 
ua .x; t/ C uhs .x; t/ t < '.x/
u.x; t/ D vJ nTSt Di r d D
St ua .x; t/ C uhs .x; t/  ˛vJ t > '.x/;
(5.44)

where the axial and the heart surface components, in the limit case, are given by the
following surface integrals:
Z Z
ua .x; t/ D ˇ vJ .aTl nSt /aTl r d and uhs .x; t/ D ˛ vJ nTSt Dr d :
St ˙d
(5.45)

In Eq. (5.44), the current sources are concentrated on the wavefront surface St and
are defined by a dipole layer with local dipole direction Di n. This model, also called
oblique dipole layer model, was proposed in [108] and subsequently investigated in
[110, 113].
We remark that the uhs component presents a jump equal to ˛vJ through ˙d
whereas it is continuous across ˙r . This means that different depolarized regions
Hd having the same ˙d on ˙ give rise to the same field component uhs . This is
reminiscent of a property of the potential field generated by a uniform normal dipole
layer. Indeed, the potential generated by these sources, when they are imbedded in
an infinite homogeneous isotropic medium (i.e. D .jxj1 jx0 j1 /=.40 /)
is given by uhs D ˛vJ !.x/, where ! is the solid angle subtended in x 2 Hr by the
surface ˙d . Thus for a given ˙ the field component uhs depends only on the rim of
the wave front intersection on ˙.
In summary, we have shown that when the anisotropic myocardial tissue is in
contact with conducting media, the potential presents a component uhs , having, in
the anisotropic context, the same features as the classical solid angle formula.
170 5 Anisotropic Cardiac Sources

In particular, this component has the peculiarity that it acts only when either
the epicardial or endocardial surface or both are not completely insulated and is
revealed, becoming measurable, only when activation reaches these non-insulated
boundaries; moreover the magnitude of its contribution is practically independent of
the wave front shape but depends only on the activated areas enclosed by the wave
front rim lying on the noninsulated boundaries. The jump component uj , related to
activated region (see Eq. (5.44)), introduces a shift of the potential. The sum of these
two components uj and uhs is the conormal component uc , usually referred to as the
heart surface source model, see [197, 198, 483].
From the split form it follows that for tissue having uniform anisotropy the full
potential is derived by superimposing the axial potential to the previously mentioned
potential components (due to the heart surface model and activating volume).
For tissue having uniform anisotropy, i.e. li;e , ti;e constant in ˝H , as assumed
in the numerical simulations, the full potential is derived by superimposing the
axial potential to the conormal component which in turn is the sum of the heart
surface component uhs and the potential jump uj across the wave front. The ability
of expressing the field as the sum of the axial and conormal components enables
us to underline the difference between the field produced by the classical conormal
model and the actual cardiac field.
It is worth mentioning here that the estimates of the intra- and extracellular
conductivities do not verify the condition of equal anisotropy ratio. Nevertheless
many computational studies in electrocardiology have often used this assumption.
Under this hypothesis ˇ D 0 hence ua drops out and consequently the cardiac fiber
arrangement plays no role in determining the structure of the current source density.
Cardiac anisotropy influences only the current conduction flow in the bulk medium,
described by the tensor D but frequently handled as an isotropic conducting medium
[108, 113, 198, 295, 483]. Therefore under the equal anisotropy ratio condition the
model reduces to u D uc C uhs C uj and coincides with the heart surface source
model used in [197,198,483]. As a consequence this model reduces to the conormal
one.
The more realistic physiological situation of a Bidomain model with unequal
anisotropy ratio, involves the heart surface model with the addition of the axial
potential component; the latter introduces a strong dependence of the cardiac
sources on the fiber anisotropy and architecture.

5.7 Experimental and Simulation Results

Figure 5.2 illustrates the connections between the epicardial potential patterns and
the intramural currents produced by the various source components described in the
previous section. Potential distributions are shown on the epicardial surface and
on three intramural sections: vertical, horizontal and oblique. On the intramural
5.7 Experimental and Simulation Results 171

Full Axial

Conormal Heart surface

Fig. 5.2 Source splitting of the full extracellular potential into axial, conormal and heart surface
components at 30 ms after endocardial pacing. In each panel, the potentials are shown on epicardial
(a), meridian (b), horizontal (c), oblique (d) sections of a truncated ellipsoid in contact with
intracavitary blood, with i D 9:5; b D 6 mS cm1

sections (panels b, c, d) the axial and conormal components produce substantially

different current fields. In fact, the inspection of the potential gradients shows that
the currents generated by the conormal component flow from all portions of the
wavefront toward the epicardial surface (outflowing currents), whereas the axial
component produces currents that flow toward and into extensive portions of the
advancing wavefront, propagating mainly across fibers (inflowing currents). The
densely packed lines characterizing the intersection of the wavefront with the three
sections b, c, d, indicate a strong gradient associated with the conormal component.
Conversely, the potential gradients produced by the axial component where the
currents flow toward the wavefront are comparatively weaker. The conormal and
heart surface panels of Fig. 5.2 show that, at 30 ms, the two components coincide on
the epicardium. This indicates an extremely weak effect of the jump component (the
difference between the conormal and heart surface components) on the epicardial
surface, in accordance with the lack of epicardial breakthrough. In addition, in the
heart surface panel of Fig. 5.2, the narrow strip of level lines, located at the basis
172 5 Anisotropic Cardiac Sources

of the wavefront in contact with the intracavitary blood, indicates that the heart
surface component presents a jump through this basis. This jump characterizes a
source, lying on this interface, generating a current that spreads through the wall
thickness and reaches the insulated epicardial surface tangentially. If we disregard
the medium anisotropy, the heart surface features are similar to those exhibited by
the potential field generated in an isotropic medium by a classical source model, i.e.
a normal dipole layer lying on the same wavefront basis and having dipole moment
proportional to the jump component.
Figure 5.3 shows other simulation results on an oblique section in a test with
intracavitary blood and extracardiac conducting medium (outside the red contour
delimiting the cardiac tissue). The full (left panel) and conormal (right panel)
components are shown on an oblique section of the domain. The full component
shows that at 30 ms the epicardial breakthrough has not yet occurred, so in the
potential map on the epicardium we observe a typical pattern characterized by
the presence of two positive maxima separated by a negative area surrounding a
central minimum. Most of the epicardial current lines, issuing from the area of
maximum positivity, point toward the minimum site, either directly or after bending.
Consequently this minimum acts as a sink for the current flow, whereas the conormal
component predicts that any portion of the wavefront surface acts as a source for the
current flow.
These simulated results are confirmed by the experimental results of Fig. 5.4,
where the current lines are shown (in red) both inside the cardiac domain and in
the torso, on the indicated diagonal section. As in the simulated results of Fig. 5.3,

28
2.80 25
2.60 23
2.40 20
2.20 17
2.00 14
1.80 11
1.60 8
1.40 6
1.20 3
1.00 0
0.80 -7
0.60 -15
0.40 -22
0.20 -30
0.00 -37
-0.20 -45
-0.40 -52
-0.60 -60
-0.80 -67
-1.00 -75

Fig. 5.3 Source splitting; full (left) and conormal (right) components of extracellular potentials at
30 ms on truncated ellipsoid, with i D 9:5; b D 6 mS cm1
5.7 Experimental and Simulation Results 173

Fig. 5.4 Experimental results of extracellular potentials related to an isolated heart embedded in
a torso tank (Courtesy of prof. B. Taccardi)

the current lines issue from two positive maxima and point toward the sink (marked
in green in the right panel of Fig. 5.4). For more details, analysis and physiological
interpretation of the potential and current pathways in the extracardiac medium, see
[350, 513, 514, 516].
Chapter 6
The Inverse Problem of Electrocardiology

Since the end of the 1970s, automated instruments have been employed to record
potential Body Surface Maps (BSM), see e.g. [144, 504, 505, 511]. The spreading
of these techniques of potential recording is motivated by the fact that BSM
provide diagnostic information which could not be obtained by conventional
electrocardiograms in many heart diseases, such as Wolff-Parkinson-White and
myocardial ischemia. In fact some features of the potential surface field, such as
the number, location and time course of potential maxima and minima on the
chest surface, are correlated to the shape of the depolarization wavefronts and to
the spatial distribution of the repolarization processes in the heart, which are the
phenomena generating bioelectrical currents.
BSM exhibit a highly changeable surface pattern as characterized by equipo-
tential lines; moreover there is a high variability of the surface signal magnitude
over the entire heartbeat. The diffusion of BSM technique has raised the problem
of the interpretation of potential distribution on the thorax. The problem of the
best use of the information content of the large amount of data provided by
BSM may be attempted by solving the so-called Forward and Inverse problems of
Electrocardiology. Two approaches are possible:
• (P) in terms of potential alone;
• (S) in terms of cardiac sources.
In the first approach (P), the Forward problem consists of:
simulating the BSM from the epicardial extracellular potential distribution,
while the Inverse problem consists of:
estimating the epicardial extracellular potential distribution from the BSM.
We remark that this approach is justified by experimental studies on animals both
in vitro and in vivo, which show that epicardial maps contain a great amount of
information more directly readable in terms of the underlying cardiac events than
BSM; moreover in this approach it is possible to compare quantitatively computed
epicardial maps with the measured ones collected in experiments on animals.

© Springer International Publishing Switzerland 2014 175

P. Colli Franzone et al., Mathematical Cardiac Electrophysiology, MS&A 13,
DOI 10.1007/978-3-319-04801-7__6
176 6 The Inverse Problem of Electrocardiology

The second approach (S) consists of modeling the cardiac sources by means of the
so-called equivalent cardiac generators, hence the Forward problem consists of:
simulating the BSM by means of an adequate Equivalent Generator Model,
while the Inverse problem consists of:
identifying an adequate Equivalent Generator of the heart from BSM.
Many groups have investigated methods for reconstructing the cardiac electric
activity with both the approaches, see e.g. [29, 64, 89, 106, 107, 111, 143, 210, 215,
222, 325, 353, 407, 440–442, 534, 550].

6.1 Inverse Problem in Terms of Potential Alone

In relation to the feasibility of the inverse problem in terms of potential alone, the
following question arises: to what extent can the potential distribution on the heart
surface be computed from BSM? We discuss here the relevant features concerning
this question.
The human body ˝ can be considered as an isotropic linear resistive conducting
medium excluding the cardiac region ˝H . At any time instant of the heartbeat the
electric field can be considered quasi-static (see [394]) and the volume conductor
˝ is embedded into an insulating medium (the air), i.e. the normal derivative of
the potential is zero on  D @˝. We set ˝0 D ˝ n ˝H , i.e. the extracardiac
medium, k.x/ the electrical conductivity and H D @˝H represents a fixed surface
surrounding the heart and lying in proximity of the heart surface; in the following,
H is referred to as epicardial surface. At any time t let V .x/ be the electric potential
distribution in ˝0 . If u.x/ D V .x/jH is known, then V .x/ in ˝0 is characterized by
the following mixed boundary problem:
8
ˆ
ˆ div k.x/rV .x/ D 0 in ˝0
<
V .x/ D u.x/ on H
ˆ @V .x/
:̂ D0 on 
@n

and the forward problem consists of evaluating V .x/ on  . If no information is

available about V .x/ on H , but it is possible to measure V .x/ D z.x/ on ˙ ,
then we have:
8
ˆ
ˆ div k.x/rV .x/ D 0 in ˝0
<
V .x/ D z.x/ on ˙ (6.1)
ˆ @V .x/
:̂ D0 on 
@n

and the inverse problem consists of estimating V .x/ on H . If the observed potential
z on ˙ were measured error-free, the Cauchy problem for the elliptic operator (6.1)
would define a unique solution V .x/, but in a highly unstable fashion, since, as it
6.1 Inverse Problem in Terms of Potential Alone 177

is well known, the Cauchy problem for elliptic operators is an ill-posed problem
in usual Sobolev spaces, see [236, 249, 374, 518]. Let v 2 L2 .H / define the state
1
y.xI v/ D y.v/ as the unique solution in H 2 .˝0 / of
8
ˆ
ˆ div k.x/ry.v/ D 0 in ˝0
<
y.v/ D v on H
ˆ @y.v/
:̂ D0 on ;
@n
then we introduce the following operator

Av D y.v/j˙

and for z 2 L2 .˙/ the cost function

Z
J.v/ D jy.v/  zj2 d D jjAv  zjj20;˙ :
˙

We consider the minimization problem:

find u 2 L2 .H / W J.u/ D inf J.v/: (6.2)

v2L2 .H /

If z 2 A.L2 .H //, i.e. there exists a unique solution of the Cauchy problem,
then u D V .x/jH 2 L2 .H / satisfies Au D z, hence u is the unique minimizer
of J . But the problem is still unstable, i.e. the operator A admits an unbounded
inverse operator in the spaces H s , 8s 2 R. In order to solve the problem one must
approximate it with a family of stable problems. Many methods are available to
perform this stabilization. We first present Tikhonov regularization techniques, by
imposing smoothing constraints justified by the physical problem. We consider the
following Tikhonov zero-order, first-order and second-order regularizations:

Space of admissible Regularization

Controls U Operator B
U D L2 .H / with B D IjH
U D H .H / D fv 2 L2 .H / W Bv 2 L2 .H /g
1
with B D rjH
U D H 2 .H / D fv 2 L2 .H / W Bv 2 L2 .H /g with B D jH :

Defining the regularization cost function

Z
J .v/ D J.v/ C  jBvj2 d; v 2 U;  > 0 “small”; (6.3)
H
178 6 The Inverse Problem of Electrocardiology

then the inverse problem (6.2) is approximated by the following family of stable
problems dependent on :

find u 2 U W J .u / D min J .v/: (6.4)

v2U

We refer to [304] for the existence, the uniqueness of the minimizer u of (6.4) and
for the convergence of u to u.
Some Authors have also recently considered L1-regularizations considering the
total variation of v, i.e.
Z
J .v/ D J.v/ C  jrvjd; v 2 U;  > 0 “small”;
H

yielding two non-linear minimization problems, see e.g. [200, 477].

For the numerical approximation of the regularized problem (6.4), it is necessary
to introduce a finite dimension approximation Uh of the control space U and an
approximate observation operator Ah of A. If we consider a parametrization of the
control u (having n degrees of freedom):

X
n X
n
u.x/ D ui w0i .x/ then y.xI u/ D ui yi .x/;
i D1 i D1

where yi .x/ is the solution of:

8
ˆ
ˆ div K.x/ryi .x/ D 0 in ˝
ˆ
ˆ
ˆ
< yi .x/j D w0 .x/; @yi .x/ D 0
0 i i D 1; : : : ; n
@n j1 (6.5)
ˆ
ˆ X n
ˆ
ˆ Au D ui yi .x/j˙ :
:̂
i D1

Hence a way of building an approximate operator of A consists in the numerical

solution of the previous n mixed boundary elliptic problems. The approximate
solution yih .x/ of yi .x/, solution of (6.5), can be obtained by means of the
finite element method applied to the three-dimensional variational formulation of
problem (6.5). The numerical computation can be performed using finite elements
of first order, see [111]. Denoting by ˙ h ; 0h the discretized surfaces respectively of
0
˙ and 0 and by fx˙ j gj D1;m ; fxk gkD1;n the mesh points of ˙ and 0 , we define:
h h

X
n
Ah uh D y h .xI uh /j˙ h D uk ykh .x/j˙ h ; (6.6)
kD1

P
where uh D nkD1 uk w0k .x/, fw0k gkD1;n is a basis of the finite element space used
on 0h and u D .u1 ; : : : ; un /T is the vector of the nodal values of uh on 0h . Setting
6.1 Inverse Problem in Terms of Potential Alone 179

z D f.Ah uh /.x˙ h
j /gj D1;m the vector of nodal values of Ah uh on ˙ then in terms of
z and u the relation (6.6) implies

z D T u;

where

T D fTji g with Tji D yih .x˙

j / i D 1; n j D 1; m

is the so called transfer matrix between 0h and ˙ h . Moreover the matrix T
depends only on the geometric data ˝. The first step of the numerical procedure
consists in the computation of T ; T is obtained solving the n mixed boundary value
problems (6.5), which differ only for the Dirichlet data on 0 , by implementing a
suitable version of the frontal method which performs a triangular factorization only
once for all the n problems. The matrix T is a very ill-conditioned matrix reflecting
the ill-posedness of the inverse problem (6.2). We remark that the transfer matrix can
also be computed by means of methods based on boundary integral representations
of the potential. Since the accuracy of the transfer matrix is not a relevant factor
in the experimental inverse problem, all the following results will be related to the
transfer matrix computed by the three dimensional method outlined above, which is
convenient in order to reduce computational time and storage requirement. It is easy
to verify that the discretized form of the regularized problem (6.4) can be restated
in the following least square form:

find u 2 Rn W minn jjT v  bjj2Rm C jjRvjj2Rn ; (6.7)

v2R

where T is a suitable matrix related to the transfer matrix T , b is related to the

observation z.x/ on ˙ and R is obtained by discretizing the operator B (see [111]
for details). For results about convergence and error estimates of the solution of
the approximate regularized problem to the solution of the continuous one (6.4),
we refer to [111]. We will limit the discussionR about the numerical results obtained
using T D T , i.e. instead of the residual ˙ h jAh uh  zj2 d we consider the
discrete residual jjT u  zm jj, where zm is the vector of the measured potential
values at the nodes of ˙ h . In the application no improvement in accuracy was
achieved using a distributed observation z.x/ obtained by interpolating suitably zm .
We remark that the problem (6.7) must be solved for a sequence of zm data, i.e.
the measured surface potential at different time instants of the heartbeat and several
values of the smoothing parameter  must be considered. Hence in order to reduce
the computational load it is important to apply a very efficient procedure. This can
be done by computing, only once, the generalized singular value decomposition of
the matrix T and R, after that the solution u can be written explicitly as a function
of . Further regularization techniques used in inverse electrocardiography are the
truncated singular value decomposition and truncated iterative Krylov methods. For
an improved spatio-temporal approach for estimating the epicardial potential on a
whole heartbeat we refer to [211].
180 6 The Inverse Problem of Electrocardiology

6.2 Macroscopic Equivalent Excitation Cardiac Sources

Let ˝ be the body volume and ˝H the heart muscle. We suppose that ˝ and ˝H are
open connected sets of class C 2 and ˝H ˝. We recall that at a macroscopic level
the excitable cardiac tissue ˝H is conceived as the superposition of two continuous
media: the intra-(i) and extra-(e) cellular media. We denote by Di and De the intra-
and extracellular conductivity tensors of these media. The anisotropic conduction
properties characterized by the tensors Di;e are related to the fiber structure of the
heart muscle. We make the following assumptions regarding the anatomical fiber
architecture:
(i) The fibers are mathematically described by regular curves;
(ii) Through any point of ˝H passes a unique fiber;
(iii) On those parts of @˝H , which lie on the heart surface (epi- or endocardium),
the fibers are tangent to @˝H .
During the excitation phase of the myocardium, a layer of cardiac cells, changing
in time, undergoes the so-called depolarization process, i.e. a rapid change occurs
in the transmembrane electric potential v.x; t/, with an approximately monotone
variation from a resting value vr to a plateau value va . The following assumption is
usually made:
(iv) The depolarization of a cell is instantaneous and is the same for all cells (vr <
va constants).
For x 2 ˝H , let li;e .x/ be the (i)-(e) conductivity coefficients along the fiber
direction at point x. Assuming axial symmetry around the fiber direction, let ti;e .x/
be the (i)-(e) conductivity coefficients in any direction perpendicular to the fiber
direction at point x. If fal ; at ; an g denotes a local basis of R3 with al parallel to the
fiber direction, then the (i)-(e) conductivity tensors are defined by:

Di;e D ti;e I C .li;e  ti;e /al aTl :

The axis at ; an are defined up to a rotation around al . Moreover Di;e do not depend
on the orientation of al parallel to the longitudinal fiber direction. If n is a unit vector
at point x, then:

Di n D ti n C .li  ti /.n  al /al I

moreover Di n is independent of the orientation of al . In the following, we will make

the simplifying assumption that:
(v) The bulk cardiac medium and the body volume are isotropic homo–geneous
with constant conductivity 0 (Di C De D 0 I ).
6.2 Macroscopic Equivalent Excitation Cardiac Sources 181

Assuming an instantaneous depolarization process at any time instant t, we

denote by

H C .t/ D fx 2 ˝H W v.x; t/ D va g;
H  .t/ D fx 2 ˝H W v.x; t/ D vr g;

C 
with H [ H D ˝H , H C \ H  D ; and S D @H C \ @H  , the excitation
wavefront surface. We also assume that vr ; va are constant and S is an orientable
regular surface. Denoting by u.x/ the potential in ˝ and by J D 0 ru, Ji D Di rv
the current density in ˝ and in the intracellular medium, the current conservation
implies (see Sect. 3.3):
8
< div.J C Ji / D 0 in ˝H
div J D 0 in ˝ n ˝H (no sources)
: T
n JD0 on  D @˝ (insulating boundary):

The boundary condition on  D @˝ describes the fact that the thorax is surrounded
by air, which is an insulating medium. In terms of potential, we have:
8 
ˆ
ˆ  div Di rv in ˝H
< 0 u D
0 in ˝ n ˝H
ˆ @u
:̂ D0 on :
@n

Denoting by ıS the Dirac measure on S :

Z
< ıS ; ' >D 'dS ' 2 D.˝/
S

and observing that

rv D .va  vr /ıS nS

where nS is the normal to S directed toward H  .t/, we obtain the following

boundary value problem:
8 va  vr
ˆ
< u D div Di .ıS nS / in ˝
0 (6.8)
@u
:̂ D0 on :
@n
182 6 The Inverse Problem of Electrocardiology

Setting
va  vr va  vr i va  vr i
M D Di D Adiag.mt ; mt ; ml /AT ; mt D t ; ml D l ;
0 0 0
(6.9)
where A D .al ; at ; an /, and using the results of [305] (for the Sobolev space H s see
e.g. [305]), the following result holds:
Theorem 6.1. Assuming that:
• The elements of M can be extended in the whole ˝ as functions of C 1 .˝/;
• S is a surface of class C 2 and if S is open there exists an open subset ˝ 0 of ˝
of class C 2 such that ˝ 0 ˝,
1
then there exists a solution u 2 H 2  .˝/, 8 > 0 of problem (6.8) defined up to an
additive constant.
Using the Green’s formula it is also possible to prove the following theorem, see
[110]:
Theorem 6.2. Under the assumptions of Theorem 6.1, the solution of (6.8) in
1
H 2  .˝/, 8 > 0, satisfies the following boundary value problem:
8
ˆ
ˆ u D 0 in ˝ n S
ˆ
ˆ
ˆ
ˆ @u
< D0 on  .in the sense of H 1 . //
@n
ˆ ŒuS D ujS C  ujS  D n M n
ˆ
T
.in the sense of H  .S //
ˆ
ˆ
ˆ @u @u @u
:̂ D  D divS ˇ .in the sense of H 1 .S //;
@n S @n jS C @n jS 
(6.10)

where ˇ D M n  .nT M n/n is a tangential field on S and divS is the divergence

operator on S .
In order to investigate the structure of the source model on the wavefront S
generating the potential u and also for the numerical approximation of u we state
the following useful theorem, see [109, 110]:
Theorem 6.3. Under the assumptions of Theorem 6.1, the solution of (6.8) admits
the following boundary integral representation:
Z Z
@s.x; /
u.x/ D  v./ d C nT M r s.x; /dS ; 8x 2 ˝ n S ; (6.11)
 @n S

where v is the unique solution in L2 . /=R of the following Integral Equation on 

Z Z
1 @s.x; /
v.x/ C v./ d D nT M r s.x; /dS a.e. on ; (6.12)
2  @n S
6.2 Macroscopic Equivalent Excitation Cardiac Sources 183

and s.x; / D 4jxj

1
is the fundamental solution of the Laplace operator  in R3 .
Moreover uj D v.
In the integral representation given by Theorem 6.3, there appear two terms: the
first of them takes into account the presence of the insulating boundary  , while the
second is related to the source model on the wavefront S , see also [238]. Since we
have:
Z Z
@s Mn
n M rsdS D
T
jjM njj dS;  D ;
S S @ jjM njj

we derive that, under the assumption (iv), the mathematical model of the excitation
wavefront S consists of:
an oblique dipole (double) layer on S having the same orientation of the vector
M n and moment density on S given by jjM njj.
Hence M can be viewed as a dipolar moment tensor. Introducing the unit vectors
al , at such that al is parallel to the longitudinal fiber direction, at is perpendicular
to al and coplanar with n, al and orienting at , al toward the resting tissue so that
at  n  0 and al  n  0, it is easy to verify:
Z Z Z
@s @s
nT M rsdS D mt .n  at / dS C ml .n  al / dS:
S S @at S @al

The source can be viewed as the superposition of a transverse dipole layer with
density mt .n  at / and an axial dipole layer with density ml .n  al /, where, in this
context, transverse and axial means respectively perpendicular and parallel to the
local fiber direction.
Moreover the following decomposition holds:
Z Z Z
@s @s
n M rsdS D
T
mt dS C .ml  mt /.n  al / dS;
S S @n S @al

which shows that the oblique dipole layer potential can be thought of as the
superposition of a normal dipole layer with density mt and of an axial dipole layer
with density .ml  mt /.n  al /. If mt D ml constant, we recover the classical model,
i.e. the source is represented by a normal dipole layer. We remark that mt D ml
implies ti D li , i.e. the intracellular medium is isotropic; however experimental
measurements have shown that ti ¤ li and the experimental data of [30, 134] did
not agree with the prediction of the normal dipole layer model, thus questioning
the general validity of the classical model. The model presented here generalizes
the classical model and also the axial model introduced in [134] for explaining the
discrepancy with the prediction of the classical model in dog heart experiments. Our
source model is an anisotropic oblique dipole model where the anisotropy derives
from the properties of the tensor M .
184 6 The Inverse Problem of Electrocardiology

6.3 Boundedness of the Potential

If the wavefront S is a closed surface, the solution u of problem (6.8) belongs to

L1 .˝/. We now discuss the condition for boundedness of the potential field u in
the case of an open surface S .
From the integral representation (6.11), we have:
Z Z
@s
u.x/ D  v d C P .x/; with P .x/ D nT M rsdS:
 @n S

If S is an open surface, the following decomposition holds:

Z Z Z
@s
P .x/ D ˛ dS  divS ˇsdS C ˇ  nb sd@S;
S @n S @S

where ˛ D nT M n, ˇ D M n  ˛n and nb is the normal to @S contained in

the tangent plane to S . This decomposition shows that the presence of the simple
layer on @S introduces a logarithmic singularity in the potential. Since the physical
potential is bounded in ˝, we must impose that the line integral must be zero
8x 2 ˝ n S , hence

ˇ  nb D 0 on @S: (6.13)

On electrophysiological grounds, we assume that:

(vi) The fiber direction on @˝H (the epi-endo-cardial heart surface) is tangent to
@˝H , i.e. al  nH D 0 on @˝H (where nH is the unit normal to @˝H ) and, if S
is open, @S @˝H .
It is not difficult to verify that (see [109]):
Proposition 6.1. Under the assumption (vi) and if mt .x/ ¤ ml .x/ 8x 2 @S , then
8
< ˇ  nb D 0 on @S if nS is parallel to nH or
nS is perpendicular to nH or (6.14)
:
al is tangent to @S:

That is the wavefront surface S and the heart surface @˝H are, respectively, tangent
or perpendicular or else intersect at an arbitrary angle in the point of @S in which
al is tangent to @S .
6.4 Numerical Approximation of the Integral Representation of the Potential 185

6.4 Numerical Approximation of the Integral Representation

of the Potential

The numerical approximation of the potential u in ˝ n S was obtained considering

the boundary integral representation of u given by Theorem 6.3, i.e.
Z Z
@s
u.x/ D  v d C nT M rsdS 8x 2 ˝ n S ;
 @n S

where v is the solution of

Z Z
1 @s
v.x/ C v d D nT M rsdS on :
2  @n S

The numerical approximation of the integral equation on  can be performed

using:
(i) The Galerkin boundary element method;
(ii) The collocation boundary element method.
We remark that for (i) an optimal error estimate was proved in [103] in the case of
linear finite element approximation of the surface and function:

jjv  vh jjL2 . /=R D O.h2 /:

For (ii) we do not know error estimate results. However in the following we will
present this second strategy, that is the most used in practice due to the high
computational costs required by the approximation of a double surface integral in
the (i) procedure.
The surfaces  , S are approximated by means of two polyhedral surfaces  h ,
h
S with triangular elements. The potential v is approximated by a piecewise linear
continuous function vh .x/ on  h , i.e.

X
N
vh .x/ D vj pj .x/; with pj .xk / D ıjk ; vk D vh .xk /;
j D1

where xk denote the nodes on  h . Applying the collocation method at the nodes of
 h , the integral equation is approximated by:
Z Z
@s.xk ; /
!.xk /vk C vh ./ dh D nTh M r s.xk ; /dSh for k D 1; : : : ; N;
 h @nh; Sh
(6.15)
186 6 The Inverse Problem of Electrocardiology

where
Z
˛.xk / @s.x; /
!.xk / D with ˛.x/ D  dh ;
4 h @nh;

i.e. ˛.x/ is the solid angle under which the surface  h is seen from the point x, nh
is the normal to the triangular elements belonging to  h or S h . Setting
 Z
@s.xk ; /
K D akj D !.xk /ıkj C pj ./ dh
h @nh
 Z
b D bk D nTh M r s.xk ; /dSh ; v D .v1 ; : : : ; vN /T ;
Sh

the solution of (6.15) is equivalent to the solution of the linear system

Kv D b: (6.16)

With !.x/ defined as above, matrix K admits the zero eigenvalue associated to the
right eigenvector

e D .1; : : : ; 1/T :

We remark that the integral operator associated to the first side of Eq. (6.12) admits
zero as a simple eigenvalue; in the following we assume that zero is a simple
eigenvalue also for the approximate operator K. Hence the solution of the linear
system is defined up to an additive constant; moreover system (6.16) admits a
solution if ` T b D 0, where ` T K D 0T , i.e. ` is the left eigenvector of K, associated
to the zero eigenvalue. While in the continuous
R case the corresponding compatibility
condition is exactly satisfied by the term S nT M rsdS, on the other hand due to
approximation errors, the term b does not satisfy exactly the condition ` T b D 0.
For solving the singular linear system by direct methods, the following deflation
method is used. Let q be a vector such that qT b D 1. It is easy to verify that:
if ` T b D 0 then the solution w of the system Bw D .K C eqT /w D b,
where B is non-singular, satisfies the equations Kw D b; qT w D 0.
For ` T b ' 0, we solve the linear system

.K C eqT /v D b

and we consider v as an approximation of vh .x/.

In practice we choose pi D ıij . We remark that K is a full matrix. The
determination of the coefficients requires the computation of the integrals
Z
@s.x; /
p./ dT  ;
T @n
6.5 Inverse Problem in Terms of Wavefront 187

where T is a triangle of  h . These integrals are computed by means of Gaussian

quadrature rules if x is not very near to T , otherwise the computation is performed
analytically.

6.5 Inverse Problem in Terms of Wavefront

In this section, we present the inverse problem in terms of wavefront, using the
oblique dipole layer model as a representation of the depolarization wavefront.
The problem can be roughly formulated a s follows: is it possible to determine
the wavefront S from the knowledge of the potential on the thorax  ?
We recall that the knowledge of:
• The myocardial fibers direction (al );
• The jump of the intracellular action potential (ua  ur );
• The coefficients of the longitudinal and transverse intracellular conductivity
(li ; ti );
• The extracellular and extracardiac conductivity (0 )
characterizes the matrix M given in (6.9).
In the following, we assume that li ; ti are positive constants and li ¤ ti , i.e.
we have homogeneous intracellular anisotropy, or equivalently we assume:

the eigenvalues of M are independent of x in ˝H and

(6.17)
mt ¤ ml with mt ; ml > 0; .mt double eigenvalue/:

We introduce the following family W of admissible wavefront S :

8 9
ˆ
ˆ set of the open orientable connected surfaces S of >
>
< =
class C 2 contained in ˝H with boundary @S @˝H and S can be
W D
ˆ considered as a part of a C 2 closed surface contained in ˝ >
>
:̂ ;
moreover @S satisfies contition (6.14)

Then for S 2 W we consider the following operator, defined up to an additive

constant:

S ! US D uj ;

where u is the solution of problem (6.8) of Theorem 6.1 or equivalently

S ! US D v;

where v is the solution of the integral equation (6.12) of Theorem 6.3. That is, given
S we consider the potential US on the insulating boundary  generated by the
188 6 The Inverse Problem of Electrocardiology

oblique dipole layer on S . We remark that in the special case in which al is always
perpendicular or tangent to a given surface S , i.e. al D nS or al  nS D 0 then the
oblique dipole layer on S is a normal double layer; if S1 and S2 are two surfaces of
W satisfying the preceding constraint, then it is well known that US1 D US 2 implies
@S1 D @S2 , i.e. the potential characterizes only the boundary @S of S . Assuming
that the wavefront is generally oblique w.r.t the fiber direction, i.e. al is not always
perpendicular or tangent to the wavefront, the following uniqueness result holds (see
[110]):
Theorem 6.4. If S1 ; S2 2 W and on at least one of them, e.g. S1 , al is generally
oblique to S1 , i.e. al  nS ¤ 0 or al  nS ¤ 1, but there exists at least one point where
al  nS D 1 or al  nS D 0, then

US1 D US2 ) S 1 D S 2 :

Under the same hypotheses of Theorem 6.4, the uniqueness result holds also for
closed wavefronts, i.e. S1 and S2 are C 2 closed surfaces one inside the other. These
first results and others suggest the conjecture that the uniqueness result should hold
under the same hypotheses of Theorem 6.4 in a more general admissible wavefront
class W and for ml ¤ mt variable in ˝H . Finally we outline the following open
problems:
• Extension of the results to an extracellular anisotropic medium and to a body
volume with piecewise constant conductivity;
• Numerical approaches for solving the inverse problem.
We now describe the inverse problem presented in [142,143,241,534,536] for the
reconstruction of the activation time on the endo- and epicardial surfaces, assuming
homogeneous isotropic properties of the cardiac and extracardiac media. As we have
seen in Chap. 5, in the general anisotropic case, the spatio-temporal transmembrane
potential distribution over the whole heart muscle must be taken into account as
primary cardiac electric source. Indeed, the integral representation of the electric
potential u at a point x of the body surface is
Z
u.x; t/ D  Di rv.y; t/  r .y; x/d y;
˝H

where Di is the intracellular conductivity tensor and the Bidomain Green func-
tion, see Eq. (5.7). However, if we assume isotropy for the intra- and extracellular
domains and accept the model error involved, only the transmembrane potential
distribution over the heart’s surface (epi- and endocardium) takes effect as primary
source, because the volume integral is converted to a surface integral by means
of Gauss divergence theorem. In fact, assuming all the media isotropic and the
extracardiac medium unbounded, we obtain
Z
i 1
u.x; t/ D  rv.y; t/  r d y;
40 ˝H jy  xj
6.5 Inverse Problem in Terms of Wavefront 189

which, applying the Green formula, becomes

Z
i 1
u.x; t/ D  v.y; t/r  n.y/d y: (6.18)
40 @˝H jy  xj

This simplification is necessary because the fiber orientation of the individual patient
is unknown and cannot be assessed by current imaging methods.
The onset of the transmembrane potential v is described by the activation time
. The non-linear relationship between the activation time and the transmembrane
potential v makes the problem non-linear. This non-linear problem can be reduced
to a linear problem in the unknown activation time by approximating the time
evolution of the transmembrane potential by a Heaviside step function H , i.e.

v.y; t/ D vr C .va  vr /H .t  .y//:

For depolarization, this assumption is reasonable, because the excitation wavefront

has a thin spatial layer. Thus, Eq. (6.18) becomes
Z
i .va  vr / 1
u.x; t/ D  H .t  .y//r  n.y/d y;
40 @˝H jy  xj

and, integrating over the QRS interval .0; T /, we obtain finally the following linear
problem in
Z Z
T
i .va  vr / 1
u.x; t/dt D .y/r  n.y/d y;
0 40 @˝H jy  xj

which relates the measured body surface potential u with the unknown activation
time on the heart surface. A minimum-norm least-square solution to this linear
problem, however, contains unphysiological characteristics and additional regu-
larization is inevitable. The additional regularization yields a non-linear inverse
problem. For a comparison between the potential- and the activation-based inverse
problems see [89].
An alternative strategy followed e.g. by [335] consists of representing the
transmembrane potential v as an analytical function with a sigmoidal shape,
  
va 2 t  .y/
v.y; t/ D 1 C arctan  C vr ;
2  w

where the unknowns are

• The activation time .y/;
• The resting and depolarized potential values vr and va ;
• The time constant of rise w.
190 6 The Inverse Problem of Electrocardiology

The same type of approach can be extended to the entire heartbeat, representing
the transmembrane potential waveform by an analytical function involving some
parameters as vr and va and markers of local depolarization and repolarization time,
see [534, 537].
The inverse method described here attempt to reconstruct the electrical activity
on the epi- and/or endocardial surfaces of the heart. In recent years, there have also
been several attempts to reconstruct cardiac electrical activity inside the myocardial
wall, see e.g. [217, 222, 325, 326, 468, 478, 549–551]; see also [353, 354] for inverse
problems to locate transmural ischemia. This problem is even more difficult than
that of reconstructing surface electrical activity, as it is less constrained and hence
more likely to loose uniqueness of solutions.
Chapter 7
Numerical Methods for the Bidomain
and Reduced Models

In this chapter, we review the main numerical techniques used in the literature for the
space and time discretizations of the Monodomain and Bidomain cardiac models.
Then we present a Bidomain discretization based on the finite element method
in space and finite difference methods in time. The latter can be fully implicit
or semi-implicit, and can employ decoupling techniques and operator splitting
methods between the ordinary and partial differential equation components of the
cardiac models. The chapter is concluded by a review of numerical methods for the
eikonal-diffusion equation.

7.1 Space Discretization of Monodomain and Bidomain

Models

Many different approaches have been developed for the space discretization of
the Bidomain and Monodomain equations. Finite difference methods have been
studied in [66, 223, 368, 399, 413, 449, 529, 568, 569, 591]. Finite element methods
have been widely used, see e.g. [104, 124, 177, 302, 303, 318, 388, 455, 543, 544].
Finite volume methods, which have the advantage of conserving local flux, have
been developed in [43, 139, 218, 220, 252, 375, 528]. Some researchers have also
investigated spectral element methods [54] and high-order finite elements [13, 14].
Non-conforming finite element discretizations have been studied in [376, 377]. In
order to reduce the computational load of the Bidomain system and of general
parabolic reaction-diffusion systems, also adaptive remeshing techniques have been
developed, see e.g. [35, 36, 90, 91, 124, 147, 290, 336, 524, 556, 577, 578]; these
techniques have been proven successful for problems of moderate size and are
currently under investigation. Numerical methods and simulations for different
eikonal approaches can be found in [268, 271, 272, 365] and in [115, 116, 121, 522].

© Springer International Publishing Switzerland 2014 191

P. Colli Franzone et al., Mathematical Cardiac Electrophysiology, MS&A 13,
DOI 10.1007/978-3-319-04801-7__7
192 7 Numerical Methods for the Bidomain and Reduced Models

Variational formulation. Finite element discretizations of the Bidomain and

Monodomain systems are based on the classical Galerkin procedure for the vari-
ational formulation of these systems.
Let V be the Sobolev space H 1 .˝H /=R and define by
Z
.'; / D ' dx 8'; 2 L2 .˝H /
˝H
Z
ai;e .'; / D .r'/T Di;e .x/ r dx 8'; 2 H 1 .˝H /
˝H
Z
a.'; / D .r'/T D.x/ r dx 8'; 2 H 1 .˝H /
˝H

the usual L2 inner product and elliptic bilinear forms. The variational formulation
of the Monodomain model reads as follows. Given v0 ; w0 2 L2 .˝H /; Iapp 2
L2 .˝H  .0; T //, find v 2 W 1;1 .0; T I V / and w 2 W 1;1 .0; T I L2 .˝H /M / such
that 8t 2 .0; T /
8
ˆ @
ˆ
ˆ Cm .v.t/; '/ C a.v.t/; '/ C .Iion .v; w/; '/ D .Iapp ; '/ 8' 2 V
< @t
ˆ
ˆ @
:̂ .w.t/; / D .R.v.t/; w.t//; / 8 2 L2 .˝H /M ;
@t
(7.1)

with appropriate initial conditions on v; w.

Analogously, the variational formulation of the Bidomain model (3.42) reads
as follows. Given v0 ; w0 2 L2 .˝H /; Iapp i;e
2 L2 .˝H  .0; T //, find ui;e 2
L .0; T I V / and w 2 L .0; T I L .˝H / / such that @v
2 2 2 M
@t 2 L .0; T I V /; @t 2
2 @w

L .0; T I L .˝H / /; and 8t 2 .0; T /

2 2 M

8
ˆ
ˆ
@
Cm .v.t/; uO i / C ai .ui .t/; uO i / C .Iion .v; w/; uO i / D .Iapp
i
; uO i / 8Oui 2 V
ˆ
ˆ
ˆ
ˆ @t
ˆ
ˆ
ˆ
<
@
Cm .v.t/; uO e / C ae .ue .t/; uO e /  .Iion .v; w/; uO e / D .Iappe
; uO e / 8Oue 2 V
ˆ
ˆ @t
ˆ
ˆ
ˆ
ˆ
ˆ
ˆ @
:̂ .w.t/; w/
O D .R.v.t/; w.t//; w/; O v.x; t/ D ui .x; t/  ue .x; t/ 8w O 2 L2 .˝H /M ;
@t
(7.2)
i;e
with the proper initial conditions on v; w and compatibility condition on Iapp .
Finite element discretizations. Let T be a uniform hexahedral triangulation of
h

˝H and V h the associated finite element space, see [414]. Choosing a finite element
basis f i g for V h and appropriate quadrature rule, we denote by
Z Z
M D fmrs D r s dxg; Ai;e D fai;e
rs D .r r/
T
Di;e r s dxg;
˝H ˝H
7.1 Space Discretization of Monodomain and Bidomain Models 193

the symmetric mass matrix and stiffness matrices, respectively, and by Ihion ; Ihapp
the finite element interpolants of Iion and Iapp , respectively. In the following, we
will denote by the same letters finite element functions and the vectors of their
nodal values. In the Monodomain model, the finite element approximation vh of the
transmembrane potential is the solution of

@vh
Cm M C Avh C MIhion .vh ; wh / D MIhapp ; (7.3)
@t
while in the Bidomain model, the finite element approximations ui;h ; ue;h of the intra
ed extracellular potentials are the solutions of the system
8
ˆ @vh
ˆ
< Cm M @t C Ai ui;h C MIion .vh ; wh / D MIapp
h i;h
ˆ
(7.4)
ˆ
ˆ @v
:̂ Cm M h C Ae ue;h  MIh .vh ; wh / D MIe;h ;
ion app
@t
where vh D ui;h  ue;h . In both cases, these equations are coupled with the
semidiscrete approximations of the gating and concentration system

@wh
D R.vh ; wh /:
@t
The Bidomain system can be written in compact form as
      !
@ ui;h ui;h MIhion .vh ; wh / MIi;h
Cm M CA C D app
; (7.5)
@t ue;h ue;h MIhion .vh ; wh / MIe;h
app

where

M M Ai 0
M D ; A D :
M M 0 Ae

The Bidomain system (7.4) can alternatively be written in a parabolic-elliptic

formulation in terms of vh ; ue;h . By adding the two equations and substituting
ui;h D vh C ue;h into the first equation, we obtain:
8
ˆ
ˆ @vh
< Cm M C Ai vh C Ai ue;h C MIhion .vh ; wh / D MIi;h
app
@t
(7.6)
ˆ
:̂ A v C .A C A /u D M.Ii;h C Ie;h /:
i h e i e;h app app

In the language of Differential-Algebraic equations (DAE), this formulation sep-

arates the differential variable (vh ) from the algebraic variable (ue;h ). Since this
parabolic-elliptic formulation lends itself to different operator splitting techniques,
treated in the next chapter, it has been and still is used by many researchers
in the field.
194 7 Numerical Methods for the Bidomain and Reduced Models

7.2 Time Discretization of Monodomain and Bidomain

Models

The time discretization of the Bidomain and Monodomain equations can be

performed by several techniques employing explicit, semi-implicit or implicit
schemes, requiring accordingly vector updates or the solution of linear or non-
linear systems at each time step. We refer, e.g., to Ethier and Bourgault [166] for
a comparative study of the stability and accuracy of several semi-implicit Bidomain
time discretizations.
Explicit methods. Explicit schemes, in particular the Forward Euler method,
have been used e.g. in [97, 368, 409, 520].
Operator splitting and decoupling PDE/ODE techniques. One of the most
popular technique is known as operator splitting, and is based on separating the
diffusion operator, associated with the conduction in the media, from the reaction
operator, associated with the ionic current, gating and ionic concentrations dynam-
ics. The advantage of operator splitting methods is to allow different numerical
schemes for the diffusion and the reaction terms in order to maximize computational
efficiency and eliminate complex dependencies between variables. The disadvan-
tage is a loss of accuracy, because the simultaneous dependencies between variables
is neglected. For the parabolic-elliptic formulation of the Bidomain model, a further
splitting method consists in solving sequentially the elliptic equation, followed by
the parabolic one. For more details on Bidomain operator splitting techniques, we
refer e.g. to the studies [319, 389, 399, 410, 501–503, 544, 545, 552]. Finally, we
mention the approximation results related to semi-discrete approximation in time of
the Bidomain model with the FHN membrane model. In [32] a priori and a posteriori
error estimates have been developed for the implicit Euler scheme.
Linear implicit methods. Additional methods that show a good compromise
between stability and efficiency are the linear implicit methods, which require at
each time step the solution of a few (2–4) linear systems. Bidomain linear implicit
methods have been studied e.g. in [124, 147, 340, 574].
We will now give some more details in case of implicit and semi-implicit dis-
cretizations of the Bidomain system in its parabolic–parabolic form. For simplicity
of notation, we will now drop the mesh size index h from all finite element functions,
writing e.g. vn instead of vnh .

7.2.1 Fully Implicit Methods

The advantage of implicit methods is that they do not require stability constraints
on the choice of the time step, but they can be quite expensive, because at each time
step one has to solve a non-linear system. Fully implicit methods in time have been
considered in [233, 338–340, 342, 574].
7.2 Time Discretization of Monodomain and Bidomain Models 195

Considering for simplicity the Backward Euler (BE) method with time step size
D tnC1  tn , the Bidomain system requires at each time step the solution of the
non-linear system

F .u/ D 0;

defined as
2 ! ! 3
unC1
i MIion .vnC1 ; wnC1 ; cnC1 /
6. M C A / C  7
6 unC1 MIion .vnC1 ; wnC1 ; cnC1 / 7
6 e
! ! 7
6 7
6 uni MIiapp 7
F .u/ D 6  M  7;
6 7
6 une MIeapp 7
6 7
4 5
wnC1  R.vn ; wnC1 / D wn ; cnC1 D cn C S.vn ; wnC1 ; cn /

where

Cm M M Ai 0
D ; M D ; A D :
M M 0 Ae

7.2.2 Decoupled Implicit Methods

A first attempt to reduce the computational cost of fully implicit methods is based
on decoupling the gating variables w from the potentials ui ; ue (hence v). This
strategy could be motivated by the nonzero pattern of the Jacobian of a fully-implicit
discretization when this shows a weak coupling between the gating and the potential
variables (for example using the LR1 ionic model the Jacobian is dominated by
its diagonal entries). Another motivation is that these two groups of variables are
associated to two very different submodels (the ionic model and the tissue model,
respectively) that describe different scales and are expressed mathematically by
different systems (an ODE and a PDE system, respectively).
We describe this decoupling technique here by employing for simplicity a
backward Euler scheme for the parabolic reaction-diffusion system.
Decoupled backward Euler (BE) method. Each time step of a decoupled
backward Euler (BE) time discretization of the Bidomain system (7.5) consists
of the following two steps. Analogous decoupled techniques could be applied to
high-order time discretizations, such as, e.g., third-order Rosenbrock methods (see
[338]). For simplicity, we will consider a fixed time step size .
Step 1: Updating gating and ionic concentration variables. Given vn , wn ,
cn , computed at the previous time step tn , first update the gating and ion
concentration variables at time tnC1 by solving
196 7 Numerical Methods for the Bidomain and Reduced Models

wnC1  R.vn ; wnC1 / D wn ; cnC1 D cn C S.vn ; wnC1 ; cn /;

where D tnC1  tn is the time step size.

Step 2: Updating intra- and extracellular potentials. Given the potentials
uni ; une ; vn computed at the previous time step and the gating and ionic con-
centration variables wnC1 , cn , computed at Step 1, find unC1
i ; uenC1 and hence
v nC1
D ui  ue , by solving the non-linear system
nC1 nC1

F.unC1
i ; unC1
e / D G; (7.7)

where

uinC1 MIion .vnC1 ; wnC1 ; cnC1 /

F.unC1 ; unC1 / D . M CA/ nC1 C  ;
i e
ue MIion .vnC1 ; wnC1 ; cnC1 /
Cm
D ;

and
" #
uni MIi;nC1
GD M C app
:
une MIe;nC1
app

7.2.3 Decoupled Semi-implicit Methods

Semi-implicit methods (also known as implicit-explicit methods) for the Bidomain

system have been studied e.g. in [104, 125, 166, 270, 455, 487, 494, 555]. We here
sketch a semi-implicit time discretization of the Bidomain system, using for the
diffusion term the implicit Euler method, while the non-linear reaction term Iion
is treated explicitly. The implicit treatment of the diffusion terms appearing in the
Monodomain or Bidomain models is essential in order to adaptively change the
time step according to the stiffness of the various phases of the heartbeat. The ODE
system for the gating variables is discretized by the semi-implicit Euler method
and the explicit Euler method is applied for solving the ODE system for the ion
concentrations. Given the potential vnM at the previous time-step, the Bidomain
system is decoupled by first solving for the gating and ion concentrations and then
by solving a linear system for the intra- and extracellular potentials.
Decoupled method for the parabolic–parabolic formulation.
Step 1: Updating gating and ionic concentration variables.

wnC1  R.vn ; wnC1 / D wn ; cnC1 D cn C S.vn ; wnC1 ; cn /:

7.2 Time Discretization of Monodomain and Bidomain Models 197

Step 2: Updating intra- and extracellular potentials. Find unC1 D

.unC1
i ; unC1
e / by solving the linear system

Abid unC1 D F (7.8)

where

M M Ai 0 Cm
Abid D M C A D C ; D ; (7.9)
M M 0 Ae

and
!
MŒIion .vn ; wnC1 ; cnC1 / C Ii;nC1
app 
F D : (7.10)
MΠIion .vn ; wnC1 ; cnC1 / C Ie;nC1
app 

As in the continuous model, vn is uniquely determined, while uni and une are
determined only up to the same additive time-dependent constant chosen by
imposing the condition 1T Mune D 0. Hence, at each time step we have to solve
the large linear system (7.8), that, as shown in [104], is very ill-conditioned and
increases considerably the computational costs of the simulations.
Following the techniques from [166], we now investigate the stability of this
semi-implicit scheme. For sake of simplicity, we consider a membrane model with
only one gating variable and without stimulus currents Iiapp ; Ieapp . Since the reaction
terms are taken explicitly, we suppose that Iion and F satisfy the following Lipschitz
condition:

jjIion .v; w/jj0  ion .jjvjj0 C jjwjj0 /

(7.11)
jjF .v; w/jj0  F .jjvjj0 C jjwjj0 /:

Using as test functions the solutions unC1

i , unC1
e and wnC1 in their respective
variational equations, we obtain
Z Z
vnC1  vn
unC1
i C ai .uinC1 ; unC1
i / C Iion .vn ; wnC1 /unC1
i D0
˝ZH ˝ZH
vnC1  vn
 uenC1 C ae .uenC1 ; unC1
e / Iion .vn ; wnC1 /uenC1 D 0
Z ˝H Z ˝H
wnC1  wn nC1
w  F .vn ; wnC1 /wnC1 D 0:
˝H ˝H

Adding these three equations and using the identity

2.anC1  an /anC1 D .anC1 /2 C .anC1  an /2  .an /2 ;

198 7 Numerical Methods for the Bidomain and Reduced Models

we get

jjvnC1jj20 C jjvnC1  vn jj20  jjvn jj20 C

2 ai .unC1
i ; unC1
i / C 2 ae .unC1
e ; unC1
e /C
Z jjw jj0 C jjw
nC1 2 nC1
 w jjZ0  jjwn jj20 
n 2

2 Iion .vn ; wnC1 /vnC1 C 2 F .vn ; wnC1 /wnC1 ;

˝H ˝H

which, thanks to the uniform ellipticity of the conductivity tensors Di;e and to the
Lipschitz condition (7.11), becomes

jjvnC1jj20 C jjvnC1  vn jj20  jjvn jj20 C

2 mi junC1
i j21 C 2 me juenC1 j21 C
jjw jj0 C jjwnC1  wn jj20  jjwn jj20 
nC1 2

2 ion jjv jj0 C . ion C 3 F /jjwnC1 jj20 C . ion C

nC1 2
F /jjv jj0 :
n 2

Denote now by M the index of the final time step (thus T D M ) and sum the
previous inequality for n going from 0 to m  1 with 1  m  M , to obtain

X
m1 X
m1 X
m1
jjvm jj20 C jjvnC1  vn jj20 C 2 m .juinC1 j21 C juenC1 j21 / C jjwm jj20 C jjwnC1  wn jj20 
nD0 nD0 nD0
X
m1
jjv0 jj20 C jjw0 jj20 C C1 .jjvm jj20 C jjwm jj20 / C C2 .jjvn jj20 C jjwn jj20 /;
nD0

where m D min.mi ; me /, C1 D max.2 ion ; ion C3 F/ and C2 D max.3 ion C

F ; ion C 3 F /. Hence it follows

X
m1
.1  C1 /.jjvm jj20 C jjwm jj20 / C 2 m .juinC1 j21 C junC1
e j21 / 
nD0
X
m1
jjv0 jj20 C jjw0 jj20 C C2 .jjvn jj20 C jjwn jj20 /:
nD0

Lemma 7.1 (Discrete Gronwall). Let fkn g, fpn g be two sequences of non-negative
real numbers, ˚ m a discrete real-valued function and g0 a non-negative real number
such that ˚ 0  g0 . Suppose also that 8m  1,

X
m1 X
m1
˚ m  g0 C pn C kn ˚ n :
nD0 nD0

Then the following property holds true:

!
X
m1 Pm1
˚  g0 C
m
pn e nD0 kn
:
nD0
7.2 Time Discretization of Monodomain and Bidomain Models 199

The lemma implies that, assuming < 1=C1 ,

 
1 T C2
max .jjvn jj20 C jjwn jj20 /  .jjv0 jj20 C jjw0 jj20 / e 1 C1 :
nD1;:::;M 1  C1

We can now use this inequality to bound the H 1 terms as

X
m1
1 T C2
.junC1
i j21 C junC1
e j21 /  .jjv0 jj20 C jjw0 jj20 / C max.jjvn jj20 C jjwn jj20 /:
nD0
2m 2m n

We shall now consider the stability of a semi-implicit second order BDF method
(SBDF). Taking unC1i ; uenC1 ; wnC1 as test functions, the associated variational
equations read as follows
Z
3vnC1  4vn C vn1 nC1
ui Cai .unC1
i ; unC1
i /C
˝H 2 Z
.2Iion .vn ; wn /  Iion .vn1 ; wn1 //unC1
i D0
Z ˝H
3vnC1  4vn C vn1 nC1
 ue Cae .unC1 e ; unC1
e /
˝H 2 Z
.2Iion .vn ; wn /  Iion .vn1 ; wn1 //unC1
e D0
Z ˝Z
H
3wnC1  4wn C wn1 nC1
w  .2F .vn ; wn /  F .vn1 ; wn1 //wnC1 D 0:
˝H 2 ˝H

Adding these three equations, multiplying by 4 and applying to the time discretiza-
tion terms the identity

.6anC1  8an C 2an1 /anC1 D .anC1 /2 C .2anC1  an /2 C .ıt t anC1 /2 

.an /2  .2an  an1 /2 ;

where ıt t anC1 D anC1  2an C an1 , we obtain

jjvnC1 jj20 C jj2vnC1  vn jj20  jjvn jj20  jj2vn  vn1 jj20 C

4 .ai .unC1
i ; unC1
i / C ae .uenC1 ; unC1
e //C
jjw jj0 C
nC1 2
Z jj2wnC1
 w jj
n 2
0  jjw jj
n 2
0  jj2w n
 wn1 jj20 
 8 .2Iion .vn ; wn /  Iion .vn1 ; wn1 //vnC1 C
˝H
Z
8 .2F .vn ; wn /  F .vn1 ; wn1 //wnC1 :
˝H
200 7 Numerical Methods for the Bidomain and Reduced Models

Thanks to the uniform ellipticity of the conductivity tensors Mi;e and to the Lipschitz
condition (7.11), we get

jjvnC1 jj20  jjvn jj20 C jj2vnC1  vn jj20  jj2vn  vn1 jj20 C

4 .mi juinC1j21 C me junC1
e j21 /C
jjw jj0  jjw jj0 C jj2w
nC1 2 n 2 nC1
 w jj0  jj2wn  wn1 jj20 
n 2

12. ion jjvnC1 jj20 C F jjwnC1 jj20 /C

4. ion C F /.jjvn jj20 C jjwn jj20 C jjvn1 jj20 C jjwn1 jj20 /:

Denoting now by M the index of the final time step (thus T D M ) and summing
the previous inequality for n going from 0 to m  1 with 1  m  M , we will see
the values v1 and w1 appear. They are used during the first time step of the SBDF
scheme and we take them v1 D v0 and w1 D w0 . Then we get

X
m1
jjvm jj20 C jj2vm  vm1 jj20 C 4 m .juinC1 j21 C juenC1 j21 / C jjwm jj20 C jj2wm  wm1 jj20 
nD0
X
m1
2.jjv0 jj20 C jjw0 jj20 / C C1 .jjvm jj20 C jjwm jj20 / C C2 .jjvn jj20 C jjwn jj20 /;
nD0

with C1 D max.12 ion ; 12 F / and C2 D max.24 ion C 12 F ; 12 ion C 24 F /.

Hence it follows

X
m1
.1  C1 /.jjvm jj20 C jjwm jj20 / C 4 m .juinC1 j21 C junC1
e j21 / 
nD0
X
m1
2.jjv0 jj20 C jjw0 jj20 / C C2 .jjvn jj20 C jjwn jj20 /:
nD0

From the Gronwall lemma, it follows that, assuming < 1=C1 ,

 
2 T C2
max .jjvn jj20 C jjwn jj20 /  .jjv0 jj20 C jjw0 jj20 / e 1 C1 :
nD1;:::;M 1  C1

We can now use this inequality to bound the H 1 terms as

X
m1
1 T C2
.junC1
i j21 C junC1
e j21 /  .jjv0 jj20 C jjw0 jj20 / C max.jjvn jj20 C jjwn jj20 /:
nD0
4m 4m n

A further decoupling approach consists of splitting the parabolic PDE from the
elliptic PDE in the parabolic-elliptic formulation of the Bidomain model (7.6). We
will now present some stability estimates originally developed by Fernandez and
Zemzemi [172] for this decoupled method. At each time step, the numerical scheme
consists of the following steps.
7.2 Time Discretization of Monodomain and Bidomain Models 201

Decoupled method for the parabolic–elliptic formulation.

Step 1: given the initial conditions vn , wn , cn computed at the previous time step,
find wnC1 , cnC1

wnC1  R.vn ; wnC1 / D wn ; cnC1 D cn C S.vn ; wnC1 ; cn /:

Step 2: given vn , update the extracellular potential une by solving the elliptic PDE

.Ai C Ae /une D Ai vn C Ii;n

app C Iapp :
e;n

Step 3: given une , wnC1 , cnC1 update the transmembrane potential vnC1 by solving
the parabolic PDE
 
Cm Cm
M C Ai vnC1 D Mvn  Ai une  Iion .vn ; wnC1 ; cnC1 / C Ii;nC1
app :

Following [172], let us investigate the stability of this decoupled semi-implicit

scheme in the case of a membrane model with only one gating variable, satisfying
the Lipschitz condition (7.11). For simplicity, we now disregard the presence of
the applied currents. Using as test functions the solutions vnC1 , une and wnC1 in
their respective variational equations, we obtain
Z Z
vnC1  vn
vnC1 C ai .vnC1 ; vnC1 / C ai .une ; vnC1 / C Iion .vn ; wnC1 /unC1
i D0
˝H ˝H
a .un ; une / C ae .une ; une / C ai .vn ; une / D 0
Zi e nC1 Z
w  wn nC1
w  F .vn ; wnC1 /wnC1 D 0:
˝H ˝H

Adding these three equations and using the identity

2.anC1  an /anC1 D .anC1 /2 C .anC1  an /2  .an /2 ;

we get

jjvnC1 jj20 C jjvnC1  vn jj20  jjvn jj20 C

2 ai .vnC1 ; vnC1 / C 2 ai .une ; vnC1 / C 2 ai .une ; une /C
2 ae .une ; une / C 2 ai .vn ; une /C
Z jjw jj0 C jjw  wn jjZ20  jjwn jj20 
nC1 2 nC1

2 Iion .vn ; wnC1 /vnC1 C 2 F .vn ; wnC1 /wnC1 :

˝H ˝H
202 7 Numerical Methods for the Bidomain and Reduced Models

By summing and subtracting the term 2 ai .vnC1 ; une / at the left hand side, we get

jjvnC1 jj20 C jjvnC1  vn jj20  jjvn jj20 C

2 ai .vnC1 C une ; vnC1 C une / C 2 ae .une ; une / C 2 ai .vn  vnC1 ; une /C
Z jjw jj0 C jjw  wn jjZ20  jjwn jj20 
nC1 2 nC1

2 Iion .vn ; wnC1 /vnC1 C 2 F .vn ; wnC1 /wnC1 :

˝H ˝H

Let us now focus on the term 2 ai .vn  vnC1 ; une / and perform the following
computations

2 ai .vn  vnC1 ; une / D 2 ai .vn  vnC1 ; une  vnC1 C vnC1 /

D 2 ai .vnC1  vn ; vnC1 / C 2 ai .vn  vnC1 ; vnC1 C une /
D ai .vnC1 ; vnC1 / C ai .vnC1  vn ; vnC1  vn /  ai .vn ; vn / C
2 ai .vn  vnC1 ; vnC1 C une /:

Moving the term 2 ai .vn  vnC1 ; vnC1 C une / to the right hand side and using the
Lipschitz condition (7.11), we finally get

jjvnC1 jj20 C jjvnC1  vn jj20  jjvn jj20 C

ai .vnC1 C une ; vnC1 C une / C 2 ae .une ; une / C ai .vnC1 ; vnC1 /  ai .vn ; vn /
jjwnC1 jj20 C jjwnC1  wn jj20  jjwn jj20 
2 ion jjv jj0 C . ion C 3 F /jjwnC1 jj20 C . ion C F /jjvn jj20 :
nC1 2

Denoting now by M the index of the final time step (thus T D M ), summing
the previous inequality for n going from 0 to m  1 with 1  m  M and using
the uniform ellipticity of the conductivity tensors Mi;e , we obtain

X
m1 X
m1
jjvm jj20 C jjvnC1  vn jj20 C .mi jvnC1 C une j21 C 2me june j21 / C
nD0 nD0
X
m1
mi jvm j21 C jjwm jj20 C jjwnC1  wn jj20 
nD0
X
m1
jjv0 jj20 C Ci jv0 j21 C jjw0 jj20 C C1 .jjvm jj20 C jjwm jj20 / C C2 .jjvn jj20 C jjwn jj20 /;
nD0

where C1 D max.2 ion ; ion C3 F/ and C2 D max.3 ion C F; ion C3 F /.

Hence it follows

X
m1
.1  C1 /.jjvm jj20 C jjwm jj20 / C .mi jvnC1 C une j21 C 2me june j21 / C mi jvm j21 
nD0
X
m1
jjv0 jj20 C Ci jv0 j21 C jjw0 jj20 C C2 .jjvn jj20 C jjwn jj20 /:
nD0
7.2 Time Discretization of Monodomain and Bidomain Models 203

The Gronwall lemma implies that, assuming < 1=C1 ,

 
1 T C2
max .jjvn jj20 C jjwn jj20 /  .jjv jj0 C Ci jv j1 C jjw jj0 / e 1 C1 :
0 2 0 2 0 2
nD1;:::;M 1  C1

We can now use this inequality to bound the H 1 terms as

X
m1
.mi jvnC1 C une j21 C 2me june j21 / C mi max jvn j21 
nD1;:::;M
nD0
jjv0 jj20 C Ci jv0 j21 C jjw0 jj20 C T C2 max.jjvn jj20 C jjwn jj20 /:
n

7.2.4 Operator Splitting Methods: Splitting ODEs and PDEs

In this section, we will present two operator splitting techniques for the Bidomain
model, based on splitting the ODEs of the membrane model from the PDEs. The
first method, Godunov splitting, is first order accurate in time, while the second
method, Strang splitting, is second order accurate in time. Both methods were first
introduced for the Bidomain model by Sundnes et al. [501]. The Godunov splitting
for the Monodomain model was developed by Qu and Garfinkel [410].
Godunov splitting.
Step 1: given the initial conditions vn , wn , cn computed at the previous time step,
find vQ nC1 , wnC1 , cnC1 by solving the ODEs system
8
ˆ dv
ˆ
ˆ Cm D Iion .v; w; c/ tn  t  tn C
ˆ
< dt
dw
D R.v; w/ tn  t  tn C
ˆ
ˆ
ˆ ddtc
:̂ D S.v; w; c/ tn  t  tn C :
dt

Step 2: given the initial condition vQ nC1 computed at Step 1, find unC1
i and uenC1
by solving the PDEs system
8
ˆ
< Cm
dv
C Ai ui D Iiapp tn  t  tn C
dt
:̂ Cm d v C Ae ue D Ieapp tn  t  tn C :
dt
Strang splitting.
Step 1: given the initial conditions vn , wn , cn computed at the previous time step,
find vnC1=2 , wnC1=2 , cnC1=2 by solving the ODEs system
204 7 Numerical Methods for the Bidomain and Reduced Models

8
ˆ dv
ˆ
ˆ Cm D Iion .v; w; c/ tn  t  tn C =2
ˆ
< dt
dw
D R.v; w/ tn  t  tn C =2
ˆ
ˆ
ˆ ddtc
:̂ D S.v; w; c/ tn  t  tn C =2:
dt

Step 2: given the initial condition vnC1=2 computed at Step 1, find uQ inC1 and uQ enC1
by solving the PDEs system
8
ˆ
< Cm
dv
C Ai ui D Iiapp tn  t  tn C
dt
:̂ Cm d v C Ae ue D Ieapp tn  t  tn C :
dt

Step 3: given the initial conditions vQ nC1 D uQ inC1  uQ enC1 , wnC1=2 , cnC1=2
computed at Step 1 and 2, find vnC1 , wnC1 , cnC1 by solving the ODEs system
8
ˆ dv
ˆ
ˆ Cm D Iion .v; w; c/ tn C =2  t  tn C
ˆ
< dt
dw
D R.v; w/ tn C =2  t  tn C
ˆ
ˆ
ˆ ddtc
:̂ D S.v; w; c/ tn C =2  t  tn C :
dt
In order to achieve the second order convergence for the Strang splitting scheme,
the sub-problems in each step must be solved to at least second-order accuracy.

7.3 Numerical Approximation of the Eikonal-Diffusion

Equation

Numerical methods for finding the evolving surface S.t/ based on a direct dis-
cretization of (4.23) encounters many difficulties, e.g. front–tracking techniques,
when high curvature and topological changes of the wavefronts occur. One way
to overcome the singularities due to collisions, merging and extinction is the level
set approach of (4.23) (see e.g. [365, 471]) which represents S.t/ as the zero level
surface of a function w.x; t/ formally solving

@w
D ˚.x; rw/.c C " div ˚ .x; rw//: (7.12)
@t
As observed before, during the excitation sequence in a fully recovered tissue
the wavefront surface S" .t/ admits a cartesian representation. Therefore " ./ D
1=jr j hence, dropping O."2 /, the eikonal-curvature equation (4.22) reduces to

˚.x; r /.c  " div ˚ .x; r // D 1; (7.13)

7.3 Numerical Approximation of the Eikonal-Diffusion Equation 205

and the eikonal-diffusion equation (4.26) to

 " div .˚.x; r /˚ .x; r // C c ˚.x; r / D 1: (7.14)

In both equations, the term of order " is related to the influence of the wavefront
curvature on the propagation in an anisotropic medium.
In a fully recovered tissue the propagation front admits a cartesian representation
and, from (4.17), (4.18), it is easy to verify that

p Q.x; p/p
˚.x; p/ D pT Q.x; p/p and ˚ .x; p/ D ;
˚.x; p/

with p D r . At the collision points, where r D 0, only a discontinuity appears

in the divergence term, thus the eikonal–diffusion equation (7.14) is more conve-
nient than the level set approach [365, 471] for the eikonal-curvature model, which,
under the same circumstances, exhibits singularities requiring a regularization of ˚.
Introducing boundary conditions, we have to solve the non-linear elliptic problem
8 p
< " divQ.x; p/p C c pT Q.x; p/p D 1 in ˝
H
(7.15)
: T
n QpD0 on ; .x/ D ta .x/ on Sa ;

where Sa is the boundary of the initial activated region and ta .x/ the corresponding
activation instants. The solution .x/ of this non-linear problem can be viewed as
the steady–state solution of the following parabolic problem associated with (7.15)
8 p
ˆ
ˆ
@w
ˆ
ˆ  " divQ.x; rw/rw C c rwT Q.x; rw/rw D 1 in ˝H
< @t
nT Q rw D 0 on ; (7.16)
ˆ
ˆ
ˆ
:̂ w.x; t/ D w .x/ on Sa w.x; 0/ D w0 :
a

Equation (7.16) belongs to a broad class of equations known as Hamilton-Jacobi

equations, with the Hamiltonian term given by c ˚.x; rw/ and a second order
non-linear diffusion term " divQ.x; rw/rw. We considered the following dis-
cretization in time obtained by applying a semi–implicit approximation for the
diffusive term and an explicit one for the transport term
wnC1  wn p
 " divQ.x; rwn /rwnC1 C c .rwn /T Q.x; rwn /rwn D 1;
n

with wn the approximate solution at the time tn and n D tnC1  tn . The space
discretization was carried out by the usual Galerkin finite element method. We
remark that the solution of the discrete problem can exhibit spurious secondary
fronts originating at the domain boundary, see [102]. This is due to the fact that
206 7 Numerical Methods for the Bidomain and Reduced Models

in Eq. (7.15) the transport term is dominant with respect to the non-linear diffusion
term. To overcome this problem the term ˚.x; rw/ required special treatment. A
way to avoid these numerical artifacts is to use the upwind scheme proposed by
Osher-Sethian for propagating fronts with curvature dependent speed. This upwind
method proved to be quite efficient and allowed us to solve our equation in every
case and using a mesh-size h of the order of 1 mm (see [102, 115] and also [522]).
Interested readers can find many results of numerical simulations with the eikonal
approach in [102, 114, 115, 117, 121] and [268, 271, 272, 369]. See [186] for recent
developments of numerical techniques to solve the eikonal equations.
Finally, we remark that the non-linear parabolic equation shares the same
non-linear diffusion term of the Relaxed Monodomain model. A similar implicit-
explicit time discretization could be applied for the numerical solution of the
Relaxed Monodomain model where the reaction term, modeling the ionic current
membrane, is treated explicitly instead of the Hamiltonian term. Therefore the semi-
discretization in time of the Relaxed Monodomain model reads:
8 nC1
ˆ
ˆ w  n R.vn ; wnC1 / D wn
< nC1
c D c n C n S.vn ; wnC1 ; c n /
ˆ v nC1
 vn
:̂ cm  div.Q.x; rvn /rvnC1 / C iion .vn ; wnC1 ; c nC1 / D Iapp
nC1
:
n
Chapter 8
Parallel Solvers for the Bidomain System

We have seen in Chap. 7 that most numerical approaches to solve the Bidomain
system lead to the solution of linear systems of equations. Linear solvers can be
classified as direct and iterative solvers. Direct solvers compute the solution of the
linear system in a finite number of steps, for instance by Gaussian elimination or
computing a factorization of the matrix, and in exact arithmetic would provide the
exact solution of the system. LU factorization is one of the most common direct
solvers, together with the Cholesky factorization for symmetric positive definite
matrices. On the other hand, in finite precision arithmetic, rounding errors and
the matrix ill-conditioning may spoil the approximate solution computed by direct
methods. Moreover, these direct methods have a high computational cost (cubic in
the number of unknowns of the linear system) and can suffer from fill-in, that is the
sparsity pattern of the original sparse matrix can be filled during the factorization
process, thus requiring large amounts of memory. One advantage of direct methods
in the context of evolution problems is that, if the system matrix does not change
during the time evolution, one needs to compute the factorization only once as a
preprocessing step.
The most popular methods to solve the linear systems arising from the dis-
cretization of Bidomain and Monodomain systems are iterative methods. In these
methods, approximate solutions of the linear system are computed iteratively until a
stopping criterion, typically involving the residual norm, is reached. Most published
approaches for the numerical discretisation of Monodomain and Bidomain models
leads to symmetric positive definite or semi-definite matrices. In this case, the
best choice of iterative method is the Conjugate Gradient method (CG). In some
cases, due to the formulation chosen and numerical scheme implemented, non-
symmetric matrices may arise, see e.g. [195,381,382], requiring alternative methods
as Krylov subspace solvers, in particular the Generalized Minimal Residual Method
(GMRES).

© Springer International Publishing Switzerland 2014 207

P. Colli Franzone et al., Mathematical Cardiac Electrophysiology, MS&A 13,
DOI 10.1007/978-3-319-04801-7__8
208 8 Parallel Solvers for the Bidomain System

In order to accelerate the convergence of iterative methods, several types of

preconditioners for the Bidomain system have been proposed in the literature: diag-
onal preconditioners [485], Symmetric Successive Over Relaxation [379], Block
Jacobi (BJ) preconditioners with incomplete LU factorization (ILU) for each block
[104, 371, 544], block triangular [195], optimized Schwarz [196], geometric multi-
grid [19, 499, 552], algebraic multigrid [381, 382, 389], and domain decomposition
preconditioners such as Multilevel Schwarz [341, 372, 373], Neumann-Neumann
and BDDC [581, 582]. We refer finally to [380] for substructuring preconditioners
applied to mortar discretizations of the Bidomain model.
In this chapter, we introduce Multilevel Schwarz methods for the Bidomain
system. Optimal convergence properties of the algorithms are established and
supported by numerical simulations in two and three dimensions. The numerical
tests were all performed on the Linux clusters described in Sect. 8.5. More details
and a proof of the main convergence rate bound can be found in [372]. Extensions
to hybrid and multiplicative Bidomain preconditioners can be found in [458, 459]
and to block preconditioners in [373]. For a general introduction to Domain
Decomposition methods and for Additive Schwarz preconditioners applied to
parabolic problems, we refer the interested reader to the monograph by Toselli and
Widlund [523] and to the papers by Cai [72, 73].

8.1 Bidomain Variational Setting

In this section, we briefly describe the variational formulation of the Bidomain

model. Assume that:
(H1) The cardiac region ˝H is a bounded Lipschitz connected open subset of R3 ;
(H2) The tensors Di;e .x/ satisfy the following uniform ellipticity condition:

9 ˛i;e ; Ci;e > 0 W ˛i;e jj2   T Di;e .x/  Ci;e jj2 ; 8 2 R3 ; x 2 ˝H I

(8.1)

(H3) The coefficients of Di;e .x/ are Lipschitz continuous.

Let V be the Sobolev space H 1 .˝H /, define the spaces
Z
VQ D fv 2 V W v D 0g and U D V  VQ D fu D .ui ; ue / W ui 2 V; ue 2 VQ g;
˝H

define the usual L2 -inner product

Z
.'; / D ' dx 8'; 2 L2 .˝H /
˝H
8.1 Bidomain Variational Setting 209

and the elliptic bilinear forms

Z
ai;e .'; / D .r'/T Di;e .x/r dx
˝H
Z
a.'; / D .r'/T D.x/r dx 8'; 2 H 1 .˝H /; with D D Di C De :
˝H

The variational formulation of the Bidomain model reads as follows. Given v0 ; w0 2

L2 .˝H /, Iappi;e
2 L2 .˝H  .0; T //, find ui 2 L2 .0; T I V /, ue 2 L2 .0; T I VQ /, w 2
@v
L2 .0; T I L2 .˝H /M / and c 2 L2 .0; T I L2 .˝H /M / such that 2 L2 .0; T I V /,
@t
@w @c
2 L2 .0; T I L2 .˝H /M /, 2 L2 .0; T I L2 .˝H /S / and 8t 2 .0; T / it holds
@t @t
8
ˆ @
ˆ
ˆ cm @t .v; uO i / C ai .ui ; uO i / C .Iion .v; w; c/; uO i / D .Iapp ; uO i / 8Oui 2 V
i
ˆ
<
@
cm .v; uO e / C ae .ue ; uO e /  .Iion .v; w; c/; uO e / D .Iapp e
; uO e / 8uOe 2 VQ
ˆ
ˆ @t
ˆ @ @
:̂ .w; w/ O  .R.v; w/; w/ O D 0; O  .S.v; w; c/; c/
.c; c/ O D 0 8w; O cO 2 V
@t @t
(8.2)

with the initial conditions (3.42).

We refer to [455] for a convergence analysis of finite element approximation of
the Bidomain model.
The anisotropic Bidomain model (8.2) is discretized by the finite element method
(see e.g. [61, 92]) in space and a semi-implicit method in time; see e.g. [414] for a
general introduction to these numerical methods.
Time discretization. The time discretization is performed by a semi-implicit
method using for the diffusion term the implicit Euler method, while the non-
linear reaction term Iion is treated explicitly. The implicit treatment of the diffusion
terms appearing in the Bidomain model is essential in order to adaptively change
the time step according to the stiffness of the various phases of the heart-beat.
The ODE system for the gating variables is discretized by the semi-implicit Euler
method and the explicit Euler method is applied for solving the ODE system for
the ion concentrations. As a consequence, the full evolution system is decoupled
by first solving the gating and ion concentrations system (given the transmembrane
potential vn D uni  une at the previous time-step)

.wnC1  t R.vn ; wnC1 /; w/

O D .wn ; w/
O 8w
O 2V
O D .c n C t S.vn ; wnC1 ; c n /; c/
.c nC1 ; c/ O 8cO 2 V;
210 8 Parallel Solvers for the Bidomain System

and then solving for .unC1

i ; unC1
e / 2 U the variational problem
(
.vnC1 ; uO i / C ai .uinC1 ; uO i / D . vn  Iion .vn ; wnC1 ; c nC1 / C Iapp
i
; uO i / 8Oui 2 V
 .vnC1
; uO e / C ae .ue ; uO e / D . v  Iion .v ; w
nC1 n n nC1 nC1
;c /  Iappe
; uO e / 8Oue 2 VQ
(8.3)
where vnC1 D unC1
i  uenC1 and D cm =t.
Lemma 8.1. The symmetric bilinear forms ..; //; ..; //0 W U U ! R defined by
Z Z Z
..u; v// WD rui  rvi C rue  rve C .ui  ue /.vi  ve /
˝H ˝H ˝H

and
Z Z
..u; v//0 WD u i vi C u e ve
˝H ˝H

are inner products.

Proof. Clearly ..u; u//  0; 8u 2 U:ZIf ..u; u// D 0 for some
Z u 2 U , then its
extracellular component would satisfy jrue j D 0: But
2
ue D 0, hence,
˝H ˝H
since (H1)
Z holds, ue 0. This in turn implies that ui 0, since ..u; u// D 0 also
implies u2i D 0; hence in conclusion u 0. The proof for ..; //0 is trivial.
˝H

We denote by jjj  jjj W U ! R the norm induced by the ..; // inner product, i.e.
Z Z Z
jjjujjj D 2
jrui j C
2
jrue j C
2
.ui  ue /2 :
˝H ˝H ˝H

Lemma 8.2. The bilinear form abid .; / W U  U ! R defined by

Z Z Z
abid .u; v/ WD Di rui  rvi C De rue  rve C .ui  ue /.vi  ve /
˝H ˝H ˝H

is continuous and elliptic with respect to the jjj  jjj-norm.

Proof. The lemma follows from the uniform ellipticity condition (8.1) and the
definition of jjj  jjj-norm. (a) Continuity:
Z Z Z
abid .u; v/ D Di rui  rvi C De rue  rve C .ui  ue /.vi  ve /
˝H ˝H ˝H
Z Z Z
 Ci rui  rvi C Ce rue  rve C .ui  ue /.vi  ve /
˝H ˝H ˝H

 max.Ci ; Ce ; /..u; v//  Cbid jjjujjj jjjvjjj:

8.1 Bidomain Variational Setting 211

(b) Ellipticity:
Z Z Z
abid .u; u/ D Di rui  rui C De rue  rue C .ui  ue /2
˝H ˝H ˝H
Z Z Z
 ˛i .rui / C ˛e
2
.rue / C 2
.ui  ue /2
˝H ˝H ˝H

 min.˛i ; ˛e ; /jjjujjj2  ˛bid jjjujjj2:

Given the previous notations, the stationary Bidomain system (8.3) can be rewritten
in the compact form:
Given vn , wnC1 , c nC1 , find unC1 2 U such that

abid .unC1 ; uO / D . vn  Iion .vn ; wnC1 ; c nC1 /; uO i  uO e /C .Iapp

i
; uO i /C .Iapp
e
; uO e / 8Ou 2 U;
(8.4)

which in a more general setting reads:

Given F D .Fi ; Fe / 2 L2 .˝H /  L2 .˝H /, find u 2 U such that

abid .u; v/ D< F; v > 8v 2 U; (8.5)

where

< F; v >D .Fi ; vi / C .Fe ; ve /:

To ensure the solvability of (8.5), F must satisfy the compatibility condition

< F; 1 >D 0; i.e.

Z Z
Fi C Fe D 0;
˝H ˝H

which in case (8.4) implies

Z
i
.Iapp C Iapp
e
/ D 0;
˝H

that is satisfied e.g. for

i
Iapp D Iapp
e
(8.6)

or for
Z
i
Iapp D 0; and e
Iapp D 0: (8.7)
˝H
212 8 Parallel Solvers for the Bidomain System

In the following, we assume (8.6), and setting f D Iapp

i
D Iapp
e
problem (8.5)
reduces to: find u 2 U such that

abid .u; v/ D .f; vi  ve / 8v 2 U: (8.8)

Theorem 8.1. Given f 2 L2 .˝H /, problem (8.8) is equivalent to: find u 2 U such
that

abid .u; v/ D .f; vi  ve / 8v 2 V  V: (8.9)

Proof. (8.9) ) (8.8) is trivial.

In order to prove (8.8) ) (8.9), it is sufficient to verify that (8.9) holds true taking
v D .0; 1/. Substituting v D .0; 1/ in (8.9), we obtain
Z Z
ui D f;
˝H ˝H

but this follows from (8.8) with v D .1; 0/, hence (8.8) and (8.9) are equivalent.
Remark 8.1. We note that, if (H1)–(H2)–(H3) hold, problem (8.8) is H 2 -
regular, i.e.

jui jH 2 .˝H / C jue jH 2 .˝H /  C jjf jjL2 .˝H / : (8.10)

Space discretization. The domain ˝H is discretized by introducing a quasi-

uniform finite element grid Th . Le V h be an associated conforming finite element
space. Define also the spaces
Z
VQ h D fvh 2 V h W vh D 0g and U h D V h  VQ h :
˝H

The discrete Bidomain system is obtained by applying a standard Galerkin proce-

dure. Given vh;n 2 V h , wh;n 2 V h , c h;n 2 V h we first find wh;nC1 and c h;nC1 solving

.wh;nC1  t R.vh;n ; wh;nC1 /; wO h / D .wh;n ; wO h / 8wO h 2 V h

.c h;nC1 ; cOh / D .c h;n C t S.vh;n ; wh;nC1 ; c h;n /; cOh / 8cOh 2 V h ;

and then .uh;nC1

i ; uh;nC1
e / 2 U h solving
8
ˆ
ˆ .vh;nC1 ; uO hi / C ai .uh;nC1 ; uO hi /
ˆ
<
i
D . v  Iion .v ; wh;nC1 ; c h;nC1 / C Iapp
h;n h h;n i
; uO hi / 8Ouhi 2 V h
(8.11)
ˆ
ˆ .v
h;nC1
; uO e / C ae .ue
h h;nC1
; uO e /
h
:̂ D . vh;n C I h .vh;n ; wh;nC1 ; c h;nC1 / C I e ; uO h / 8Ouh 2 V h
ion app e e

h
where Iion is the finite elements interpolants of Iion .
8.1 Bidomain Variational Setting 213

Choose a finite element basis f'i g for Vh and let M D .mrs / and Ai;e D .ai;e
rs / be the
symmetric mass and stiffness matrices defined by
XZ XZ
mrs D 'r 's dx; rs D
ai;e .r'r /T Di;e .x/ r's dx:
E E E E

Numerical quadrature with a simple trapezoidal rule in three dimensions is used

in order to compute these integrals. We can now enunciate the matrix formulation
of (8.11). Given vn , wn , cn , we first find wnC1 , cnC1 solving

wnC1  t R.vn ; wnC1 / D wn

cnC1 D cn C t S.vn; wnC1 ; cn /;

and then unC1

i ; unC1
e solving
    !
unC1 cm MŒ uni  une  MŒIhion .vn ; wnC1 ; cnC1 / C Iiapp 
Abid i D C ;
unC1
e t MŒuni C une  MŒ Ihion .vn ; wnC1 ; cnC1 / C Ieapp 
(8.12)
where

M M Ai 0
Abid D C
M M 0 Ae

and uni , une , wn , cn , unC1

i , unC1
e , wnC1 , cnC1 , Ihion , Ii;e
app are the vectors of nodal
h;n h;nC1
values of ui , uh;ne , w h;n
, c h;n
, u i , u h;nC1
e , wh;nC1
, c h;nC1 h
, Iion i;e
, Iapp respectively.
nC1 nC1
As in the continuous model, v is uniquely determined, while ui and uenC1
are determined only up to the same additive time-dependent constant chosen by
imposing the condition 1T MunC1 e D 0. Hence, at each time step we have to solve
the large linear system (8.12), that, as shown in [104], is very ill-conditioned and
increases considerably the computational costs of the simulations.
In order to introduce in the next chapter multilevel Schwarz preconditioners for
the discrete Bidomain system, as we did in (8.4), we need to rewrite (8.11) in the
compact form:
Given vh;n , wh;nC1 , c h;nC1 , find uh;nC1 2 U h such that 8Ouh 2 V h  V h

abid .uh;nC1 ; uO h / D . vh;n Iion

h
.vh;n ; wh;nC1 ; c h;nC1 /; uO hi Ouhe /C.Iapp
i
; uO hi /C.Iapp
e
; uO he /;
(8.13)
i
which, assuming as before Iapp D Iapp
e
D f , we reformulate the problem as:
Given f 2 L .˝H /, find u 2 U such that
2 h h

abid .uh ; vh / D .f; vhi  vhe / 8vh 2 V h  V h : (8.14)

214 8 Parallel Solvers for the Bidomain System

Remark 8.2. Clearly, (8.14) arises from the finite element discretization of (8.8).
By defining the linear operator Abid W U h ! U h as

..Abid uh ; vh //0 D abid .uh ; vh / 8vh 2 V h  V h ;

we have that (8.14) is equivalent to the linear operator equation

Abid uh D f h ; (8.15)

with right-hand side f h 2 U h defined as

..f h ; vh //0 D .f; vhi /  .f; vhe / 8vh 2 V h  V h :

We will construct in the next chapter multilevel Schwarz preconditioners for

problem (8.15) and, following the standard abstract Schwarz theory described e.g.
in [489, 523], we will prove their convergence.
In the proof of our main result on the optimality of the multilevel preconditioner,
we need an L2 -error estimate for the solutions of (8.8) and (8.14).
Theorem 8.2. Let u and uh be the solutions of (8.8) and (8.14), respectively. Then
the following error estimate holds

jjui  uhi jjL2 .˝H / C jjue  uhe jjL2 .˝H /  C h jjju  uh jjj: (8.16)

Proof. Consider the following dual problems:

D1. Given f 2 L2 .˝H /, find '.f / 2 U :

abid .v; '.f // D .f; vi  ve / 8v 2 U: (8.17)

R
D2. Given f 2 L2 .˝H / : ˝H f D 0, find '.f / 2 U :

abid .v; '.f // D .f; vi C ve / 8v 2 U: (8.18)

Let us define e h D .eih ; eeh / WD u  uh D .ui  uhi ; ue  uhe /. Taking f D eih and
v D e h in (8.17), one has

jjeih jj2L2 .˝H /  .eih ; eeh / D .eih ; eih  eeh / D abid .e h ; '.eih //:

Thanks to the Galerkin orthogonality, for all vh 2 U h it holds

jjeih jj2L2 .˝H /  .eih ; eeh / D abid .e h ; '.eih /  vh /  C jjje h jjj jjj'.eih /  vh jjj:
8.2 Abstract Convergence Theory for Schwarz Methods 215

Choosing vh D .Pi '.eih /; Pe '.eih //, where Pi W V ! V h and Pe W VQ ! VQ h are

standard finite element interpolation operators, and using (8.10), we get

jjeih jj2L2 .˝H /  .eih ; eeh /  C h jjje h jjj jjeih jjL2 .˝H / : (8.19)

Taking now f D eeh and v D e h in (8.18), and following the same arguments, we
have

jjeeh jj2L2 .˝H / C .eih ; eeh /  C h jjje h jjj jjeeh jjL2 .˝H / : (8.20)

Summing (8.19) and (8.20), we obtain

jjeih jj2L2 .˝H / C jjeeh jj2L2 .˝H /  C h jjje h jjj .jjeih jjL2 .˝H / C jjeeh jjL2 .˝H / /

and hence the thesis.

8.2 Abstract Convergence Theory for Schwarz Methods

In this section we recall the main results of the abstract Schwarz theory for Domain
Decomposition methods; for a more complete treatment, see the monographs
[489, 523].
Let U be a finite dimensional Hilbert space with inner product .; /, a.; / a
symmetric elliptic bilinear form on U  U and f 2 U . Consider the following
problem:
Find u 2 U such that

a.u; v/ D .f; v/ 8v 2 U: (8.21)

Define the linear operator A W U ! U by

.Au; v/ D a.u; v/ 8v 2 U:

Then (8.21) is equivalent to the linear operator equation

Au D f: (8.22)

Let Um , m D 0; : : : ; N , be a set of N C1 auxiliary spaces. In most application, these

subspaces are related to a decomposition of the domain ˝H into subdomains ˝m .
The subspace U0 plays a special role: it corresponds to a coarse mesh and provides
a global transportation of information. In addition, assume that there exists a set of
interpolation operators

Im W Um ! U:
216 8 Parallel Solvers for the Bidomain System

We will refer to the case when Um U as the nested case. In this case Im are simply
imbedding operators. Assume also that on each subspace Um there is a symmetric
positive definite bilinear form

am .; / W Um  Um ! R

and define the projections TQ m W U ! Um by

Q m u; v/ D a.u; Im v/
am .T 8v 2 Um :

This formalism allows us to consider inexact solvers for the subproblems. If we use
exact solvers, then am .; / D a.; /. Now define the operators

Tm D Im TQ m W U ! U:

We can now introduce the Additive Schwarz operator TAS given by

TAS D BAS A WD T0 C T1 C : : : C TN ; (8.23)

where BAS is the Additive Schwarz preconditioner. The Additive Schwarz method
replaces problem (8.22) by:
Find u 2 U such that

TAS u D BAS f; (8.24)

which can be solved by a Krylov subspace method, with BAS as preconditioner.

We will now describe the tools used in the convergence analysis of the Schwarz
methods, beginning with the following lemma.
Lemma 8.3. Let TAS be the Additive Schwarz operator defined in (8.23). Then
holds
X
a.T1
AS u; u/ D min
P am .um ; um /: (8.25)
um 2Um ;uD m Im um m

Proof. We first construct a particular decomposition of u which satisfies (8.25)

exactly, and then prove
X X
a.T1
AS u; u/  am .um ; um / 8um 2 Um ; Im um D u: (8.26)
m m
8.2 Abstract Convergence Theory for Schwarz Methods 217

P P
Let um D TQ m T1
AS u, then m Im um D. m Tm /T1
AS u and

X X
am .um ; um / D Q m T1
am .T Q 1
AS u; Tm TAS u/
m m
X
D a.T1 1
AS u; Tm TAS u/
m

D a.T1
AS u; u/:

We now prove (8.26),

X
a.T1 1
AS u; u/ D a.TAS u; Im um /
m
X
D a.T1
AS u; Im um /
m
X
D am .TQ m T1
AS u; um /
m
" #1=2 " #1=2
X X
 am .TQ m T1 Q 1
AS u; Tm TAS u/ am .um ; um /
m m
" #1=2 " #1=2
X X
D a.T1 1
AS u; Tm TAS u/ am .um ; um /
m m
" #1=2
X
D a.T1
AS u; u/
1=2
am .um ; um / :
m

Therefore it follows (8.26).

The abstract Schwarz convergence theory centers around three assumptions.

P decomposition). 9 C0 > 0 such that 8u 2 U there exists

Assumption 8.1 (Stable
a representation u D m Im um ; um 2 Um , with
X
am .um ; um /  C02 a.u; u/:
m

Assumption 8.2 (Strengthened Cauchy-Schwarz inequality). 9 0  Eml  1

such that

ja.Im um ; Il ul /j  Eml a.Im um ; Im um /1=2 a.Il ul ; Il ul /1=2

8um 2 Um ; ul 2 Ul m; l D 1; : : : ; N:

Define .E / to be the spectral radius of E D fEml g.

218 8 Parallel Solvers for the Bidomain System

Assumption 8.3 (Local stability). 9 ! > 0 such that

a.Im u; Im u/  !am .u; u/ 8u 2 Um ; m D 0; : : : ; N:

We recall for a linear operator L self-adjoint with respect to a.; / the Rayleigh
quotient characterization of the extreme eigenvalues,

a.Lu; u/ a.Lu; u/
min .L/ D min ; max .L/ D max :
u¤0 a.u; u/ u¤0 a.u; u/

We recall that for the condition number of L (.L/) in the norm induced by a, it
holds

max .L/
.L/ D :
min .L/

Note that by definition

min .L/a.u; u/  a.Lu; u/  max .L/a.u; u/ 8u 2 U:

Lemma 8.4. The operators Tm , and hence TAS , are self-adjoint with respect to
a.; /; that is

a.Tm u; v/ D a.u; Tm v/:

Proof. See [489].

We can now provide a bound on the condition number for the abstract Additive
Schwarz method.
Theorem 8.3. If Assumptions 8.1–8.3 are satisfied, then the condition number of
the abstract Additive Schwarz method is bounded by

.TAS /  !Œ1 C .E /C02 :

Proof. The bound on the smallest eigenvalue follows immediately from Lemma 8.3
and Assumption 8.1.
We now bound the largest eigenvalue. We first treat the subspaces Um , for m D
1; : : : ; N . The special subspace U0 is treated separately. Observe that

X
N X
N X
N
a. Tm v; Tm v/ D a.Tm v; Tl v/;
mD1 mD1 m;lD1

X
N
 Eml a.Tm v; Tm v/1=2 a.Tl v; Tl v/1=2 ;
m;lD1
8.2 Abstract Convergence Theory for Schwarz Methods 219

X
N
 .E / a.Tm v; Tm v/;
mD1

X
N
 !.E / am .TQ m v; TQ m v/;
mD1

X
N
 !.E / a.v; Tm v/;
mD1

X
N
 !.E /a.v; Tm v/;
mD1

X
N X
N
 !.E /a.v; v/1=2 a. Tm v; Tl v/1=2 :
mD1 lD1

It follows
X
N X
N X
N
a.ΠTm 2 v; v/ D a. Tm v; Tl v/;
mD1 mD1 lD1

 ! 2 2 .E /a.v; v/:
P
Thus the largest eigenvalue of . N 2 2 2
mD1 Tm / is bounded by !  .E /. Hence the
PN
largest eigenvalue of mD1 Tm is bounded by !.E / and

X
N
a. Tm v; v/  !.E /a.v; v/: (8.27)
mD1

To estimate the largest eigenvalue of T0 we use

a.T0 v; T0 v/  !a0 .TQ 0 v; TQ 0 v/;

D !a.v; T0 v/;
 !a.v; v/1=2 a.T0 v; T0 v/1=2 :

Thus

a.T0 v; T0 v/  ! 2 a.v; v/;

which implies

a.T0 v; v/  !a.v; v/:

Combine this with (8.27) and the proof is completed.

220 8 Parallel Solvers for the Bidomain System

8.3 Two-Level Additive Schwarz Methods for the Bidomain

System

In this section, we construct and analyze a two-level overlapping Additive Schwarz

method for problem (8.14). Let T0 D TH be a coarse shape-regular triangulation of
˝H consisting of N non-overlapping hexahedral subdomains ˝m , m D 1; : : : ; N
of diameter Hm and let H D maxm Hm . Let T1 D Th be a fine shape-regular
triangulation nested in T0 , consisting of hexahedral elements j , j D 1; : : : ; Ne
of diameter hj and let h D maxj hj . An overlapping partition of ˝H is then
constructed using the standard technique of adding to each subdomain ˝m all the
fine elements j 2 T1 within a distance ı from its boundary @˝m . We denote
by ˝m0 the overlapping subdomain obtained by such extensions of ˝m . With each
subdomain ˝m0 , we associate the following finite element spaces

Vm WD fv 2 V h W v.x/ D 0 x 2 ˝H n˝m0 g and Um WD Vm  Vm :

Let V H be the space of trilinear finite elements associated to the coarse triangulation
T0 and define
Z
VQ H WD fv 2 V H W v D 0g; U0 D U H WD V H  VQ H :
˝H

We remark that U0 U h instead Um is not a subset of U h , m D 1; : : : ; N . Define

the interpolation operators Im W Um ! U h , m D 1; : : : ; N , as
 Z Z 
given u D .ui ; ue / 2 Um ; Im u D .Im;i u; Im;e u/ WD ui  ue ; ue  ue :
˝H ˝H

I0 W U0 ! U h is simply the imbedding operator. Defining the projection operators

TQ m W U h ! Um

by

abid .TQ m u; v/ D abid .u; Im v/ 8v 2 Um ;

and Tm D Im TQ m , we can define the Additive Schwarz operator as

TAS D BAS Abid WD T0 C T1 C : : : C TN : (8.28)

where BAS is the two-level Additive Schwarz preconditioner. As in the scalar elliptic
case, it is easy to see that the matrix form BAS of BAS is

X
N X
N
BAS D Bm D T
Rm A1
m Rm : (8.29)
mD0 mD0
8.3 Two-Level Additive Schwarz Methods for the Bidomain System 221

Here, for m D 1; : : : ; N , Rm are boolean restriction matrices and Am the local

stiffness matrices for the Bidomain problems restricted to the subdomain ˝m0 ,
while for m D 0 R0 is the fine-to-coarse restriction matrix and A0 is the coarse
Bidomain stiffness matrix associated with the coarse space U0 D U H . More
general projection-like operators TQ m associated with approximate bilinear forms
and inexact local solvers could be used as well, see [489, 523]. The use of
the BAS preconditioner (8.29) for the iterative solution of the Bidomain discrete
system (8.15) can also be regarded as replacing (8.15) with the preconditioned
system

TAS u D g;

where g D BAS f , which can be accelerated by a Krylov subspace method.

The following result extends the classical overlapping Schwarz analysis to the
condition number of TAS .
Theorem 8.4. The condition number of the 2-level additive Schwarz operator TAS
defined in (8.28) is bounded by
 
H
.TAS /  C 1 C ;
ı

with C independent of h; H; ı.
Proof. We apply the general abstract Schwarz theory of the previous section based
on verifying the three assumptions stable decomposition, strengthened Cauchy-
Schwarz inequality and local stability, that provide an upper and lower bound on
.TAS /, see also [523, p. 40].
(a) Since we use exact local solvers, the local stability assumption holds true with
a unit constant.
(b) Using a standard coloring argument, see [523, Ch. 2], we obtain that the spectral
radius .E / of the matrix in the strengthened Cauchy-Schwarz inequality is
bounded from above by the number of colors, hence we have a constant upper
bound.
(c) It only remains to prove a stable decomposition for our subspace decomposition.
We will prove that this assumption is satisfied with a constant C02 D C.1 C H
ı /,
with C independent of h; H and ı.
Let
• QH W V h ! V0 be the L2 -projection onto V0 ,
• I h the linear interpolation operator onto V h and P
• m a partition of unity such that m 2 C01 .˝m0 /, 0  m  1, NmD1 m D 1,

C
jrm j2L1  ; (8.30)
ı2
222 8 Parallel Solvers for the Bidomain System

and
 2
h
km  Nm k2L1 .K/  C ; (8.31)
ı

where Nm is the average of m on the element K 2 Th .

Given u 2 U h , we define:

u0 2 U0 by u0;i WD QH ui and u0;e WD QH ue I

w 2 U h by wi WD ui  u0;i and we WD ue  u0;e I
um 2 Um by um;i WD I h .m wi / and um;e WD I h .m we /:

Thanks to the definition of Im and because I h is a linear operator, it follows that

X
N
uD Im um :
mD0

We can bound each scalar component as in the standard proof for scalar elliptic
problems (see e.g. [489, 523])

X
N   XN
H
jum;i j2H 1 .˝H / C 1C jui j2H 1 .˝H / and jum;e j2H 1 .˝H /
mD0
ı mD0
 
H
C 1C jue j2H 1 .˝H / : (8.32)
ı

Moreover, for all m D 1; : : : ; N and K 2 Th , we have

kum;i  um;e k2L2 .K/ D kI h .m wi /  I h .m we /k2L2 .K/ D

D kI h .m .wi  we //k2L2 .K/  kwi  we k2L2 .K/ :

Since a finite bounded number (independent of h, H and ı) of um;i and um;e are
nonzero for any element K, summing over m we obtain

X
N
kum;i  um;e k2L2 .K/  C kwi  we k2L2 .K/ ;
mD1

and summing over all the elements K

X
N
kum;i  um;e k2L2 .˝H /  C kwi  we k2L2 .˝H / :
mD1
8.4 Multilevel Additive Schwarz Methods for the Bidomain System 223

Since

kwi  we k2L2 .˝H / D kui  ue C QH .ue  ui /k2L2 .˝H /  4kui  ue k2L2 .˝H / ;

ku0;i  u0;e k2L2 .˝H / D kQH .ui  ue /k2L2 .˝H /  kui  ue k2L2 .˝H / ;

we obtain

X
N
kum;i  um;e k2L2 .˝H /  C kui  ue k2L2 .˝H / :
mD0

Adding this and the bounds (8.32), we have

X
N  
H
.jum;i j2H 1 Cjum;e j2H 1 Ckum;i um;e k2L2 /  C 1 C .jui j2H 1 Cjue j2H 1 Ckui ue k2L2 /;
ı
mD0

hence

X
N  
H
jjjumjjj  C 1 C
2
jjjujjj2:
mD0
ı

From the continuity and coercivity of abid .; /, it follows the stable decomposition

X
N  
H
abid .um ; um /  C 1 C abid .u; u/:
mD0
ı

8.4 Multilevel Additive Schwarz Methods for the Bidomain

System

The method described in the previous section uses two levels with a coarse and a
fine mesh. The smaller is H , the smaller are the local problems, but the algebraic
linear system corresponding to the problem in U0 becomes larger. We can then apply
recursively this two-level technique to the coarse problem and obtain an additive
multilevel method for the Bidomain system. We refer to the original works by
Dryja and Widlund [151], Zhang [584] and later Dryja, Sarkis and Widlund [153]
for an overview of these methods for scalar elliptic problems, including multilevel
diagonal scaling variants and different choices of coarse spaces.
Let us consider L  2 rather than just two mesh levels: on each level k D
0; 1; : : : ; L  1, ˝H is discretized with a shape-regular mesh Tk with elements of
characteristic size hk . Tk is a refinement of Tk1 , with level L  1 being the finest,
hence hL1 D h and h0 D H . On each level k D 0; : : : ; L  1, we define a finite
224 8 Parallel Solvers for the Bidomain System

element space V hk , with V hL1 D V h and V h0 D V H , and the spaces VQ hk , U hk as in

0 Nk
the previous section. We introduce L  1 sets of overlapping subdomains f˝km gmD1
for k D S1; : : : ; L  1, such that on each level there is an overlapping decomposition
0
˝H D N k
mD1 ˝km and we denote with ık the overlap at level k. As in [584], we
0
make the following assumption about the sets f˝km g.
S k 0
Assumption 8.4. On each level the decomposition ˝H D N mD1 ˝km satisfies:
0 0
(a) @˝km aligns with the boundaries of level k elements, i.e. ˝km is the union of
0
level k elements. The diameter(˝km ) D O.hk1 /.
0 Nk
(b) The subdomains f˝km gmD1 form a finite covering of ˝H , with a covering
0 Nk
constant Nc , i.e. we can color f˝km gmD1 using at most Nc colors in such a
way that subdomains of the same color are disjoint.
0 Nk
(c) There exists a partition of unity fkm g, associated with f˝km gmD1 , which satisfies
X
km D 1; with km 2 H01 .˝km / \ C 0 .˝km /; 0  km  1;
m

jrkm jL1  C =ık and jjkm  Nkm jjL1 .K/  C.hk =ık /; (8.33)

where K 2 Tk and Nkm is the average of km on K.

For k D 1; : : : ; L  1 and m D 1; : : : ; Nk , we define the spaces
0
Vkm WD V hk \ H 1 .˝km /; Ukm D Vkm  Vkm :

Define the interpolation operators Ikm as in the two-level case. Defining the
Q km W U h ! Ukm by
projections T

abid .TQ km u; v/ D abid .u; Ikm v/ 8v 2 Ukm ;

and Tkm D Ikm TQ km , we can introduce the Multilevel Additive Schwarz (MAS)
operator

XX
L1 Nk
TMAS D BMAS Abid WD T0 C Tkm : (8.34)
kD1 mD1

As mentioned before, approximate local solvers could be used as well. The matrix
form BMAS of the preconditioner BMAS is given by

XX
L1 Nk
BMAS D R0T A1
0 R0 C
T
Rkm A1
km Rkm ;
kD1 mD1

0
where Akm is the local stiffness matrix of subdomain ˝km at level k and Rkm is the
0
restriction matrix from the finest level to subdomain ˝km , see [489] for more details.
We can then prove our main result for the multilevel case.
8.4 Multilevel Additive Schwarz Methods for the Bidomain System 225

Theorem 8.5. The condition number of the multilevel additive Schwarz operator
TMAS defined in (8.34) is bounded by
 
hk1
.TMAS /  C max 1C ;
kD1;:::;L1 ık

where C is a constant independent of the mesh sizes hk and the number of levels L.
Proof. Upper bound. The proof of the optimal upper bound follows from the steps
of the original proof by Zhang [584], from Lemma 3.2 to Lemma 3.5. These
results can be applied to our case since they hold also for general conforming
finite elements (Remark 3.2 in [584]) and uniform elliptic operators (Remark 3.3
in [584]).
Lower bound. In order to prove the left inequality, we need a stable decompo-
sition of u 2 U h . For each level k D 0; : : : ; L  2, we introduce the projections
Qk W U h ! U hk induced by the abid .; / bilinear form, i.e.

abid .Qk u; '/ D abid .u; '/ 8' 2 U hk :

We denote by Qik and Qek the two components of Qk , i.e. Qk u D .Qik u; Qek u/. It
follows from Theorem 8.2 that, given u 2 U h

jjui  Qik ujjL2 C jjue  Qek ujjL2  C hk jjju  Qk ujjj: (8.35)

Let u0 D Q0 u, uk D .Qk Qk1 /u for k D 1; : : : ; L2 and uL1 D .I QL2 /u.

8k let uk;i and uk;e be the two components of uk . It follows from (8.35) that

jjuk;i jjL2 C jjuk;e jjL2  C hk1 jjjuk jjj: (8.36)

8u 2 U h we use the abid .; /-orthogonal decomposition

u D u0 C u1 C : : : C uL1 :

Consider then for each k D 1; : : : ; L  1 a partition of unity fkm g as in

Assumption 8.4. Let Ihk be the linear interpolation operator onto V hk and ukm the
function in Ukm such that

ukm;i D Ihk .km uk;i / and ukm;e D Ihk .km uk;e /;

PNk1
hence uk D mD1 Ikm ukm . We prove first that for each k

X
Nk
jjjukmjjj2  C jjjuk jjj2 :
mD1
226 8 Parallel Solvers for the Bidomain System

Thanks to a well-known inverse inequality, it holds that

jukm;i j2H 1 .K/ D jIhk .km uk;i /j2H 1 .K/ D jIhk .Nkm uk;i /  Ihk ..km  Nkm /uk;i /j2H 1 .K/ 

 2jNkm uk;i j2H 1 .K/ C C h2 N

k jjIhk ..km  km /uk;i /jjL2 .K/ :
2

The last term above is zero for all elements K in the interior of an ˝km , since km D
Nkm D 1 on the non overlapping regions. Therefore, summing over all the elements,
the last term only includes the elements in the overlapping regions ık ;m . Hence

X
Nk X
Nk X
jukm;i j2H 1 .˝H /  C.juk;i j2H 1 .˝H / C h2 N
k jjIhk ..km  km /uk;i jjL2 .K/ /
2

mD1 mD1 K2ık ;m

X
Nk
 C.juk;i j2H 1 .˝H / C ık2 jjuk;i jj2L2 .ı / /: (8.37)
k ;m
mD1

We can bound the last terms using the following result by Dryja and Widlund [152]:
Lemma 8.5. Let ˝H be a shape regular region, in R2 or R3 , of diameter H and let
ı be a strip along its boundary of width ı > 0. Then

H 1
jjujj2L2 .ı /  C ı 2 Œ.1 C /juj2H 1.˝H / C jjujj2L2 .˝H / :
ı Hı
Applying this lemma to (8.37), we obtain

X
Nk
hk1 1
jukm;i j2H 1 .˝H /  C.juk;i j2H 1 .˝H / C .1 C /juk;i j2H 1 .˝H / C jjuk;i jj2L2 .˝H / /;
mD1
ık hk1 ık

and using (8.36)

X
Nk
hk1 X
Nk
hk1
jukm;i j2H 1 .˝H /  C.1C /jjjuk jjj2 ; jukm;e j2H 1 .˝H /  C.1C /jjjuk jjj2 :
mD1
ık mD1
ık

Following the same arguments as in the two-level case, we get

X
Nk
jjukm;i  ukm;e jj2L2 .˝H /  C jjuk;i  uk;e jj2L2 .˝H /  C jjjuk jjj2 ;
mD1

X
Nk
hk1
jjjukm jjj2  C.1 C /jjjuk jjj2 :
mD1
ık
8.5 Numerical Results for Multilevel Schwarz Preconditioners 227

Using the coercivity and the continuity of the Bidomain bilinear form abid .; / and
summing over the levels, we finally obtain the stable decomposition

XX
L1 Nk
hk1
abid .u0 ; u0 / C abid .ukm ; ukm /  C max .1 C /abid .u; u/:
kD1;:::;L1 ık
kD1 mD1

8.5 Numerical Results for Multilevel Schwarz

Preconditioners

The numerical experiments were performed on the Linux Cluster IBM CLX/1024 of
the Cineca Consortium (www.cineca.it), with 1024 processors (Intel Xeon Pentium
IV 3 GHz, 512 KB cache), grouped into 512 biprocessor nodes. The total RAM
is 1 TB and the peak performance is declared to 6.1 Tflops. We were allowed
to use up to 240 processors in each run. Our FORTRAN code is based on the
parallel library PETSc from the Argonne National Laboratory [24, 25]. We solve
the Bidomain system on a domain ˝H that is either a cartesian slab (see Fig. 8.2,
left) or the image of a cartesian slab using ellipsoidal coordinates, yielding a portion
of a truncated ellipsoid (see Fig. 8.1). This second choice can be used in simulations
with idealized ventricular geometries, see e.g. [104, 125] These two choices allow
us to also test the performance of our multilevel preconditioner in presence of severe
domain deformations.
The values of the conductivity coefficients used in the numerical tests are
reported in Table 8.1 and the capacitance per unit volume is set to cm D 1 mF cm3 .
The values of the coefficients and parameters in the LR1 model are given in the
original paper [308].
For both types of domains, we denote a level mesh by the triplet T D Ti Tj Tk
indicating the number of subdivisions in each coordinate direction. When we scale
up the mesh, for brevity we define cT D cTi  cTj  cTk . In all tests, the the
finest mesh size is h D 0:1 mm, which has been shown to accurately resolve the
steep excitation fronts in the Bidomain–LR1 model without numerical artifacts, see
[104, 125].
We perform 5 time steps of 0.05 ms each with the semi-implicit Euler scheme
described in Chap. 7 and at each time step we solve iteratively the discrete Bidomain
system (8.12) by the Preconditioned Conjugate Gradient (PCG, see [445]) method
with the L-level MAS preconditioner (8.34). The PCG initial guess is the discrete
solution of the previous time step and the stopping criterion is a 104 residual
reduction. The problem on the coarsest level is solved by PCG with 108 residual
reduction as stopping criterion. In all runs, we report the PCG condition number,
extreme eigenvalues, iteration counts and cpu timings relative to the 5th time step,
in order to avoid special right-hand side configurations associated with the first few
time steps.
228 8 Parallel Solvers for the Bidomain System

32 = 4⋅ 4⋅ 2

8 = 2⋅ 2⋅ 2 16 = 4⋅ 2⋅ 2

240 = 15⋅ 8⋅ 2
64 = 8⋅ 4⋅ 2 128 = 8⋅ 8⋅ 2

Fig. 8.1 Portions of ellipsoidal domain decomposed in 8, 16, 32, 64, 128, 240 subdomains for
scaled speedup test

Table 8.1 Conductivity coefficients used in the numerical tests

Conductivity coefficients
Normal tissue
li D 3  103 1 cm1 le D 2  103 1 cm1
ti D 3:1525  104 1 cm1 te D 1:3514  103 1 cm1
ni D 3:1525  105 1 cm1 ne D 6:757  104 1 cm1
Ischemic tissue
li D 3  104 1 cm1 le D 1  103 1 cm1
ti D 6:305  106 1 cm1 te D 3:3785  104 1 cm1
ni D 6:305  107 1 cm1 ne D 1:6892  104 1 cm1

8.5.1 Additive Preconditioner

Test MAS-3D-1. Increasing levels and increasing global size, cartesian slab. We
first consider in Table 8.2 a cartesian slab, see Fig. 8.2, left panel, and increase the
number of levels L in our MAS preconditioner from 3 to 6, while increasing the fine
meshes as Ti C1 D 2Ti , starting with a fixed coarse mesh T0 D 8  8  2. The number
of processors and subdomains is fixed at 128, but the global problem size increases
8.5 Numerical Results for Multilevel Schwarz Preconditioners 229

Table 8.2 Test MAS-3D-1. Increasing number of levels L for cartesian slabs on Fig. 8.2 with
increasing fine meshes TiC1 D 2Ti starting with a fixed coarse mesh T0 D 8  8  2; fixed number
of processors D 128
L Meshes d.o.f.  D max = min it. Time
3 T0 ; 2T0 ; 4T0 19 602 7.76 D 4.217/0.543 15 1.05
4 T0 ; 2T0 ; 4T0 ; 8T0 143 650 7.97 D 4.501/0.564 17 2.74
5 T0 ; 2T0 ; 4T0 ; 8T0 ; 16T0 1.1 M 7.28 D 5.241/0.720 19 5.20
6 T0 ; 2T0 ; 4T0 ; 8T0 ; 16T0 ; 32T0 8.6 M 6.82 D 5.783/0.847 18 11.29

20

18
it.
16

14

12

10
κ
8 2

4 λ
max

2 λmin

0
3 4 5 6
LEVELS

Fig. 8.2 Left: Cartesian slab subdivided into 128 D 8  8  2 subdomains. Right: plot of
min ; max ; ; it. Reported in Table 8.2

from 19 602 d.o.f. (for L D 3, fine mesh T2 D 4T0 ) to 8.6 millions d.o.f. (for
L D 6, fine mesh T5 D 32T0 ). The MAS preconditioner has overlap size ık D hk
and ILU(0) local solvers for the subdomains of each level. The main results of this
table are also plotted in Fig. 8.2, right panel. The condition numbers and iteration
counts do not grow with increasing L (actually they even decrease for higher values
of L), in spite of the strong growth of the problem size. Of course, the cpu timings
reported in the last column grow because the global problem size grow, due to the
growth of the subdomain size on the finest mesh from 43 .L D 3/ to 323 .L D 6/.
These results confirm the optimality result of Theorem  8.5, with
 constant bounds
hk1
independent of L and depending only on max 1C :
kD1;:::;L1 ık
Test MAS-3D-2. Scaled speedup on ellipsoidal domains. In Table 8.3, we then
consider a scaled speedup test for our MAS operator with 2, 3, 4 levels. Again,
we fix the overlap size ık D hk and ILU(0) local solvers on the subdomains
of each level. The local size of each subdomain on the finest mesh is kept fixed
at the value 483 and each subdomain is assigned to one processor. The number
of subdomains (hence number of processors) is increased from 8 D 2  2  2 to
240 D 15  8  2 (first and second columns), forming increasing domains ˝H that are
portions of a truncated ellipsoidal domain as shown in Fig. 8.1. These subdomains
230 8 Parallel Solvers for the Bidomain System

Table 8.3 Test MAS-3D-2. Condition numbers, extreme eigenvalues, iteration counts, cpu timings
for 2, 3, 4 level MAS, ık D hk , ILU(0) local solvers: scaled speedup with fixed subdomain size
48  48  48 on the ellipsoidal domains of Fig. 8.1
2 levels: T0 ; T1 D 48T0
Procs. T0 T1 D 48T0 d.o.f. (M)  D max = min it. Time
8 222 96  96  96 1.8 183.50 D 2.268/0.012 45 53.1
16 422 192  96  96 3.6 172.44 D 2.265/0.013 45 65.0
32 442 192  192  96 7.2 228.63 D 2.455/0.011 50 65.8
64 842 384  192  96 14.4 226.95 D 2.479/0.011 50 74.7
128 882 384  384  96 28.7 150.31 D 2.413/0.016 43 66.2
240 15  8  2 720  384  96 53.8 149.98 D 2.415/0.016 43 68.1
3 levels: T0 ; T1 D 4T0 ; T3 D 48T0
Procs. T0 T2 D 48T0 d.o.f. (M)  D max = min it. Time
8 222 96  96  96 1.8 21.70 D 2.903/0.134 23 23.9
16 422 192  96  96 3.6 23.33 D 3.092/0.132 24 30.4
32 442 192  192  96 7.2 25.07 D 3.263/0.130 24 34.2
64 842 384  192  96 14.4 25.56 D 3.269/0.128 25 32.0
128 882 384  384  96 28.7 30.70 D 3.347/0.109 26 40.2
240 15  8  2 720  384  96 53.8 30.99 D 3.381/0.109 26 40.8
4 levels: T0 ; T1 D 4T0 ; T2 D 16T0 ; T3 D 48T0
Procs. T0 T3 D 48T0 d.o.f. (M)  D max = min it. Time
8 222 96  96  96 1.8 10.81 D 3.815/0.348 20 27.2
16 422 192  96  96 3.6 11.65 D 4.141/0.355 20 29.6
32 442 192  192  96 7.2 10.18 D 4.137/0.406 19 31.0
64 842 384  192  96 14.4 9.95 D 4.117/0.414 19 33.1
128 882 384  384  96 28.7 9.96 D 4.089/0.411 19 34.8
240 15  8  2 720  384  96 53.8 9.95 D 4.083/0.410 19 36.0

are the elements of the coarse mesh T0 . The finest mesh is then 48T0 , since the
local mesh size is kept fixed at 483 . The intermediate level meshes in case of 3
and 4 levels are T1 D 4T0 and T2 D 16T0 . With these data, the global size of
the discrete Bidomain system increases then from 1.8 million d.o.f. for the smallest
domain with 8 subdomains to 53.8 million d.o.f. for the largest domain with 240
subdomains. In each of these cases, the last columns of Table 8.3 report the condition
number , extreme eigenvalues max ; min , iteration counts it: and cpu timings of the
PCG solver with MAS preconditioner. The same data are plotted in Fig. 8.3 vs.
the processor count. The results clearly show the scalability of the MAS operator,
since all the reported quantities seem to approach constant values as the number
of subdomains (hence processors) increases. As expected, these asymptotic values
improve with the number of levels, since for increasing L the MAS preconditioner
becomes more powerful.
Table 8.4 reports the results of a scaled speedup test run recently on the BlueGene
Q cluster of the CINECA consortium (www.cineca.it). The preconditioner consid-
ered is the MAS(4), the domains are portions of a truncated ellipsoid, the number
8.5 Numerical Results for Multilevel Schwarz Preconditioners 231

103 55
50
45 2 LEVELS
2 LEVELS
MAS CONDITION N.

MAS ITERATIONS
40
102
35

3 LEVELS 30 3 LEVELS
25
20
101
4 LEVELS 15 4 LEVELS
10
5
100 0
0 50 100 150 200 250 0 50 100 150 200 250
PROCESSORS (SUBDOMAINS) PROCESSORS (SUBDOMAINS)

Fig. 8.3 Test MAS-3D-2. Scalability of 2, 3, 4 level MAS: condition numbers and iteration counts
reported in Table 8.3

Table 8.4 Test MAS-3D-2. MAS(4)

Weak scaling for MAS(4)
Procs d.o.f.  D M= m it. Time
solvers on ellipsoidal
structured meshes, run on the 64 4 319 890 41.85 D 8.70/2.08e-1 43 5.65
BlueGene Cluster 128 8 553 474 33.41 D 6.79/2.03e-1 39 5.57
256 17 040 642 36.37 D 6.81/1.87e-1 40 5.70
512 33 949 186 27.37 D 5.16/1.88e-1 36 5.48
1 024 67 766 274 29.53 D 5.16/1.75e-1 36 5.69
2 048 135 268 866 27.56 D 5.08/1.84e-1 34 8.50
4 096 270 274 050 28.91 D 5.09/1.76e-1 34 16.39
8 192 540 021 250 25.03 D 5.10/2.04e-1 32 16.51
16 384 1 079 515 650 26.55 D 5.11/1.92e-1 32 17.39
Average condition number (), extreme eigenvalues ( M , m ),
PCG iteration count (it) and CPU time in seconds (time) per time
step

Table 8.5 Test MAS-3D-3. Increasing number of levels L for fixed coarse mesh T0 D 8  8  2
(486 d.o.f.) and fine mesh TL1 D 48T0 D 384  384  96 (28.7M d.o.f.), intermediate meshes
TiC1 D 2Ti ; ellipsoidal domain of Fig. 8.1, fixed number of processors D 128
L Meshes  D max = min it. Time
2 T0 ; 48T0 150.31 D 2.413/0.016 43 66.2
3 T0 ; 2T0 ; 48T0 86.55 D 3.375/0.039 38 158.5
4 T0 ; 2T0 ; 4T0 ; 48T0 43.31 D 4.374/0.101 33 124.9
5 T0 ; 2T0 ; 4T0 ; 8T0 ; 48T0 17.82 D 5.118/0.287 24 139.1
6 T0 ; 2T0 ; 4T0 ; 8T0 ; 16T0 ; 48T0 10.94 D 5.892/0.539 20 32.7

of processors varies from 64 to 16 384, while the number of d.o.f. increases from
4 million to about 1 billion.
Test MAS-3D-3. Increasing levels and fixed global size, ellipsoidal domain.
In Table 8.5, we fix the coarse mesh T0 D 8  8  2 (486 d.o.f.) and fine mesh
TL1 D 48T0 D 384  384  96 (28.7M d.o.f.), hence the number of subdomains
232 8 Parallel Solvers for the Bidomain System

and processors are fixed to 128 (the number of coarse elements). We then vary the
number of levels L from 2 to 6, with intermediate meshes defined by Ti C1 D 2Ti .
The ellipsoidal domain is the one of Fig. 8.1 decomposed into 128 subdomains. Both
the condition numbers and iteration counts improve considerably when the number
of levels increases, because the improvement of the minimum eigenvalues more than
compensate a slight increase of the maximum eigenvalue.
Test MAS-3D-4. Scaled speedup with variable overlap and ILU/LU local
solvers, cartesian slab. We now focus on the effect of two variables that we have
kept fixed so far: the overlap size ık and the use of inexact/exact local solvers.
Therefore, we repeat a scaled speedup test as in Table 8.1 but on slab domains and
with smaller subdomain size 163 in order to be able to run our MAS preconditioner
with exact LU local solvers. The finest mesh is then always TL1 D 16T0 , while
the intermediate meshes for L D 3; 4 are 2T0 and 4T0 . Table 8.6 reports the

Table 8.6 Test MAS-3D-4. Iteration counts and condition numbers of 2, 3, 4 level MAS: scaled
speedup with fixed subdomain size 16  16  16 and varying overlap size ık D 1; 2; 4  hk , with
inexact ILU(0) local solvers (in brackets exact LU local solvers)
2 levels: T0 ; T1 D 16T0
ık D hk ık D 2  hk ık D 4  hk
Procs. T0 it.  it.  it. 
8 2  2  2 38 (23) 104.5 (36.1) 39 (17) 85.3 (14.6) 34 (14) 58.7 (10.2)
18 3  3  2 47 (24) 143.4 (51.5) 49 (18) 129.1 (19.6) 37 (16) 72.0 (12.3)
32 4  4  2 45 (21) 123.3 (33.1) 41 (17) 96.9 (17.2) 39 (17) 80.3 (12.6)
50 5  5  2 45 (21) 148.1 (30.1) 41 (17) 96.8 (16.7) 43 (18) 95.9 (13.9)
72 6  6  2 41 (20) 117.2 (27.5) 42 (18) 97.9 (19.1) 43 (18) 97.3 (13.9)
128 8  8  2 39 (20) 92.0 (24.5) 42 (18) 100.8 (17.5) 44 (17) 95.4 (11.5)
3 levels: T0 ; T1 D 2T0 ; T2 D 16T0
ık D hk ık D 2  hk ık D 4  hk
Procs. T0 it.  it.  it. 
8 2  2  2 18 (14) 11.6 (6.5) 18 (12) 10.5 (4.6) 17 (12) 9.7 (3.7)
18 3  3  2 19 (14) 13.3 (6.6) 18 (13) 12.0 (4.8) 18 (13) 11.6 (5.1)
32 4  4  2 19 (14) 13.6 (6.4) 19 (13) 12.9 (5.0) 19 (13) 12.2 (4.4)
50 5  5  2 19 (14) 13.6 (6.1) 19 (14) 12.9 (4.9) 19 (13) 12.3 (4.9)
72 6  6  2 19 (14) 13.4 (6.1) 19 (14) 12.8 (4.8) 19 (13) 12.1 (5.2)
128 8  8  2 19 (14) 13.5 (6.0) 19 (13) 12.8 (4.7) 19 (14) 12.0 (4.6)
4 levels: T0 ; T1 D 2T0 ; T2 D 4T0 ; T3 D 16T0
ık D hk ık D 2  hk ık D 4  hk
Procs. T0 it.  it.  it. 
8 2  2  2 18 (16) 8.9 (8.1) 18 (16) 8.9 (7.3) 18 (16) 8.3 (6.7)
18 3  3  2 20 (18) 14.6 (12.3) 20 (18) 14.4 (10.9) 19 (17) 13.9 (8.2)
32 4  4  2 17 (17) 7.5 (7.9) 17 (16) 7.2 (7.8) 17 (16) 7.0 (7.8)
50 5  5  2 17 (17) 7.4 (7.5) 17 (16) 7.1 (7.3) 17 (16) 6.8 (7.2)
72 6  6  2 17 (17) 7.3 (7.1) 16 (16) 7.0 (6.9) 16 (16) 6.7 (6.8)
128 8  8  2 16 (16) 7.3 (6.6) 16 (16) 7.0 (6.3) 16 (16) 6.7 (6.1)
8.5 Numerical Results for Multilevel Schwarz Preconditioners 233

150 55
50
MAS CONDITION NUMBER (ILU(0))

2 LEVELS (δ = 1,2,4)
45

MAS ITERATIONS (ILU(0))

40
100
35
30
2 LEVELS (δ = 1,2,4)
25
3 LEVELS (δ = 1,2,4)
20
50
15
4 LEVELS (δ = 1,2,4)
* = 3 LEVELS, = 4 LEVELS (δ = 1,2,4) 10
5
0 0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140

PROCESSORS (SUBDOMAINS) PROCESSORS (SUBDOMAINS)

Fig. 8.4 Test MAS-3D-4. Scalability of 2, 3, 4 level MAS with ILU(0) local solvers and variable
overlap ık D 1; 2; 4  hk : condition numbers (left) and iteration counts (right) reported in Table 8.6

55 30

50
25
45 o = 2 LEVELS (δ = 1,2,4)
MAS CONDITION NUMBER (LU)

= 2 LEVELS (δ = 1,2,4)
40 * = 3 LEVELS (δ = 1,2,4)
MAS ITERATIONS (LU)

20
35 = 4 LEVELS (δ = 1,2,4)

30
15
25

20 * = 3 LEVELS, = 4 LEVELS (δ = 1,2,4)

10
15

10 5
5

0 0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
PROCESSORS (SUBDOMAINS) PROCESSORS (SUBDOMAINS)

Fig. 8.5 Test MAS-3D-4. Scalability of 2, 3, 4 level MAS with LU local solvers and variable
overlap ık D 1; 2; 4  hk : condition numbers (left) and iteration counts (right) reported in Table 8.6

iteration counts and condition numbers of 2, 3, 4 level MAS with inexact ILU(0)
local solvers; in brackets are the same quantities for MAS with exact LU local
solvers. We consider three cases of increasing overlap size ık D 1; 2; 4  hk . The
ILU(0) results are then plotted in Fig. 8.4, and the LU results in Fig. 8.5, where the
2, 3, 4 level data are denoted by ı; ?; : : :, respectively. The scalability of the MAS
preconditioner is evident in all cases; as expected, the LU results are better than
the ILU(0) results and increasing the overlap size improves the conditioning of the
preconditioned operator but only by a marginal amount when L > 2 (in fact the
iteration counts are basically independent of ı for L > 2), except in the 2-level
ILU(0) case, where the results are a bit erratic. Increasing the levels improve the
results in the ILU(0) case, but in the LU case the 3-level MAS seems to be slightly
better conditioned than the 4-level MAS for this choice of meshes and parameters.
234 8 Parallel Solvers for the Bidomain System

8.5.2 Multiplicative and Hybrid Preconditioners

In this section, we study numerically the behavior of a hybrid variant BMHS of

the multilevel additive Schwarz preconditioner BMAS . This variant is multiplicative
among the levels and additive in the levels (see [489]). We recall the same notations
used in Sect. 8.4. Let Tk , k D 0; : : : ; L  1 be a family of nested triangulations
of ˝H , coarsening from L  1 to 0, and Ak the matrix obtained discretizing (8.2)
on Tk : so AL1 D Abid . Rk are the restriction operators from TkC1 to Tk . We
0
decompose each grid Tk into Nk overlapping subdomains ˝km for m D 1; : : : ; Nk ,
such that the overlap ık at level k D 1; : : : ; L  1 is equal to the mesh size hk of
0
the grid Tk . Let Rkm be the restriction operator from Tk to ˝km and Akm the local
0
stiffness matrix of subdomain ˝km . The action of this Multilevel Hybrid Schwarz
(MHS) preconditioner on a given residual r D rL1 is given by:

NX
L1

uL1 T
RL1m A1
L1m RL1m rL1
mD1

rL2 RL2 .rL1  AL1 uL1 /

X
NL2

uL2 T
RL2m A1
L2m RL2m rL2
mD1

:::
u0 A1
0 r0

u1 u1 C R0T u0
X
N1
u1 u1 C T
R1m A1
1m R1m .r1  A1 u1 /
mD1

:::
uL1 uL1 C RL2
T
uL2
X
NL1

uL1 uL1 C T
RL1m A1
L1m RL1m .r
L1
 AL1 uL1 /
mD1

u uL1

In the following, we will denote the MHS preconditioner with L levels MHS(L).
8.5 Numerical Results for Multilevel Schwarz Preconditioners 235

TRANSMEMBRANE POTENTIAL EXTRACELLULAR POTENTIAL

10 30

20
−10

−20
10
−30

−40
0

−50

−60 −10

−70
−20
−80

−84.38 17.91 5.11 −25.10 35.48 3.03

Fig. 8.6 Test MHS-2D-1. Patterns of level lines of the transmembrane and extracellular potentials
during the excitation phase (t D 25 ms after the stimulus). Reported below each panel are the
minimum, maximum and step in mV of the displayed map

8.5.2.1 MHS: Two-Dimensional Domains

Test MHS-2D-1. Fixed global size, increasing number of subdomains, square

domain. We simulate the initial depolarization of a square section of cardiac tissue,
having dimensions of 2:56  2:56 cm2, applying a stimulus of 200 mA=cm3 for 1 ms
on a small area of 2  2 elements at a vertex of the domain, see Fig. 8.6. The two-
dimensional conductivity tensors are given by

li;e 0
Di;e D :
0 ni;e

The global mesh is fixed to be of 256256 elements (132 098 unknowns), 512512
elements (526 338 unknowns) and 1 024  1 024 elements (2 101 250 unknowns).
The cases of 4, 16, 64 subdomains (processors) and 2–7 levels are considered. The
model is run for 200 time steps of 0.05 ms, i.e. for a time interval of 10 ms on
the CLX Linux cluster. In Tables 8.7 and 8.8, we report the average number of
PCG iterations per time step, the average condition number and the average time
needed to solve the linear system. The results show the scalability of the MHS
preconditioner, since both the condition numbers and the iteration counts remain
almost constant when the number of subdomains and levels increase.
Test MHS-2D-2. Increasing levels and increasing global size, square domain.
The aim of this test is to study the behavior of the preconditioner increasing the
number of levels from 2 to 9. The size of the fine mesh varies form 4  4 elements
(50 unknowns) with 2 levels to 512  512 elements (526 338 unknowns) with 9
levels, hence the dimensions of the square domain increase from 0:04  0:04 to
5:12  5:12 cm2 . The number of subdomains (and processors) is kept fixed to 4. As
236 8 Parallel Solvers for the Bidomain System

Table 8.7 Test MHS-2D-1. Condition numbers, iteration counts, cpu timings for 2, 3, 4 level
MHS. Fixed global size, increasing the number of subdomains, square domain
2 levels 3 levels 4 levels
Procs. Mesh  it. Time  it. Time  it. Time
4 2562 1.05 4 2:93 1.05 4 1:17 1.05 4 0.98
16 2562 1.04 4 0:84 1.05 4 0:43 1.05 4 0.35
64 2562 1.04 4 0:68 1.05 4 0:43 1.05 4 0.35
4 5122 1.04 2 15:04 1.05 3 3:60 1.05 3 3.14
16 5122 1.04 2 4:02 1.05 3 1:25 1.05 3 1.15
64 5122 1.04 2 1:44 1.05 3 0:61 1.05 3 0.50
4 1 0242 1.06 2 81:89 1.07 2 21:96 1.07 2 14.20
16 1 0242 1.06 2 20:71 1.07 2 6:08 1.07 2 3.91
64 1 0242 1.06 2 6:91 1.07 2 1:98 1.07 2 1.29

Table 8.8 Test MHS-2D-1. Same format as in Table 8.7 for 5, 6, 7 level MHS
5 levels 6 levels 7 levels
Procs. Mesh  it. Time  it. Time  it. Time
4 2562 1.05 4 0:98 1.06 4 0:98 1.06 4 0.98
16 2562 1.05 4 0:38 1.05 4 0:33 1.05 4 0.36
64 2562 1.05 4 0:37 1.12 5 0:36 – – –
4 5122 1.05 3 3:10 1.05 3 3:08 1.05 3 3.09
16 5122 1.05 3 0:84 1.05 3 0:95 1.05 3 0.95
64 5122 1.05 3 0:47 1.05 3 0:43 1.05 3 0.57
4 1 0242 1.07 3 14:55 1.07 3 13:20 1.07 3 13.26
16 1 0242 1.07 3 3:44 1.07 3 3:23 1.07 3 3.28
64 1 0242 1.07 3 1:19 1.07 3 1:30 1.07 3 1.28

in the previous tests, we run the model for 10 ms after the stimulus. Table 8.9 reports
the average minimum and maximum eigenvalues, the average condition number, the
average number of PCG iterations and the average solving time per time step. The
condition numbers and iteration counts do not grow with increasing L, in spite of
the strong growth of the problem size. Of course, the cpu timings reported in the
last column grow because the global problem size grow, due to the growth of the
subdomain size on the finest mesh.
Test MHS-2D-3. Scaled speedup on square domains. In this test, we vary the
number of subdomains from 4 to 64, keeping fixed the local mesh in each subdomain
to 64  64 elements (8 450 unknowns), hence varying the global number of degrees
of freedom (d.o.f.) from 33 800 in the smallest case with 4 subdomains to 540 800 in
the largest with 64 subdomains. The number of levels varies from 2 to 7. As in test
MHS-2D-1, we simulate the initial depolarization of a myocardial section, running
the model for 200 time steps on the CLX cluster of CINECA. Tables 8.10 and 8.11
report the average number of PCG iterations and the average solving time per time
8.5 Numerical Results for Multilevel Schwarz Preconditioners 237

Table 8.9 Test MHS-2D-2. Increasing number of levels L for square domains with increasing fine
meshes TiC1 D 2Ti starting with a fixed coarse mesh T0 D 2  2; fixed number of processors D 4
L Meshes d.o.f.  D max = min it. Time
2 T0 ; 2T0 50 1.16 D 1.05/0.90 4 0.01
3 T0 ; 2T0 ; 4T0 162 1.05 D 1.01/0.97 3 0.01
4 T0 ; 2T0 ; 4T0 ; 8T0 578 1.04 D 1.00/0.97 3 0.02
5 T0 ; 2T0 ; 4T0 ; 8T0 ; 16T0 2 178 1.04 D 1.00/0.97 3 0.03
6 T0 ; 2T0 ; 4T0 ; 8T0 ; 16T0 ; 32T0 8 450 1.05 D 1.01/0.96 4 0.08
7 T0 ; 2T0 ; 4T0 ; : : : ; 32T0 ; 64T0 33 282 1.05 D 1.01/0.96 4 0.25
8 T0 ; 2T0 ; 4T0 ; : : : ; 64T0 ; 128T0 132 098 1.05 D 1.01/0.96 4 0.99
9 T0 ; 2T0 ; 4T0 ; : : : ; 128T0 ; 256T0 526 338 1.01 D 1.00/0.99 3 3.12

Table 8.10 Test MHS-2D-3. Condition numbers, iteration counts, cpu timings for 2, 3, 4 level
MHS. Scaled speedup with fixed subdomain size 64  64 on square domains
2 levels 3 levels 4 levels
Procs. d.o.f.  it. Time  it. Time  it. Time
4 33 800 1.04 4 0.46 1.05 4 0.25 1.05 4 0.25
9 76 050 1.04 4 0.63 1.04 4 0.33 1.05 4 0.32
16 135 200 1.05 4 0.87 1.05 4 0.43 1.05 4 0.45
25 211 250 1.05 4 1.07 1.05 4 0.51 1.05 4 0.45
36 304 200 1.05 4 1.43 1.05 4 0.53 1.05 4 0.55
49 414 050 1.05 4 1.57 1.05 4 0.66 1.05 4 0.66
64 540 800 1.05 4 1.94 1.05 4 0.86 1.05 4 0.75

Table 8.11 Test MHS-2D-3. Same format as in Table 8.10 for 5, 6, 7 level MHS
5 levels 6 levels 7 levels
Procs. d.o.f.  it. Time  it. Time  it. Time
4 33 800 1.05 4 0.23 1.05 4 0.25 1.05 4 0.26
9 76 050 1.04 3 0.29 1.05 4 0.28 1.05 4 0.30
16 135 200 1.05 4 0.37 1.05 4 0.33 1.05 4 0.36
25 211 250 1.05 4 0.40 1.05 4 0.41 1.05 4 0.43
36 304 200 1.05 4 0.44 1.05 4 0.44 1.07 4 0.43
49 414 050 1.05 4 0.53 1.05 4 0.48 1.08 4 0.50
64 540 800 1.05 4 0.66 1.05 4 0.55 1.09 5 0.58

step. We observe that, in this case too, both condition numbers and iteration counts
remain constant when increasing the number of subdomains and levels.

8.5.2.2 MHS: Three-Dimensional Domains

Test MHS-3D-1. Increasing levels and increasing global size, cartesian slab. As
in test MAS-3D-1 of the previous section, the behavior of the MHS preconditioner
238 8 Parallel Solvers for the Bidomain System

Table 8.12 Test MHS-3D-1. Increasing number of levels L for cartesian slabs with increasing fine
meshes TiC1 D 2Ti starting with a fixed coarse mesh T0 D 4  4  2; fixed number of processors
D 32
L Meshes d.o.f.  D max = min it. Time
3 T0 ; 2T0 ; 4T0 5 202 1.97D1.18/0.60 7 0.13
4 T0 ; 2T0 ; 4T0 ; 8T0 37 026 1.53D1.03/0.67 5 0.27
5 T0 ; 2T0 ; 4T0 ; 8T0 ; 16T0 278 850 1.46D1.01/0.68 5 0.94
6 T0 ; 2T0 ; 4T0 ; 8T0 ; 16T0 ; 32T0 2.1M 1.34D1.01/0.74 5 5.9

Table 8.13 Test MHS-3D-2. Condition numbers, iteration counts, cpu timings for 3, 4 level MHS.
Scaled speedup with fixed subdomain size 48  48  48 on the ellipsoidal domains of Fig. 8.1
3 levels 4 levels
Procs. d.o.f. (M)  it. Time  it. Time
8 1:8 1.17 4 19:11 1.18 4 16.67
16 3:6 1.17 4 21:21 1.18 4 19.82
32 7:2 1.18 4 21:86 1.20 5 20.57
64 14:4 1.18 4 23:8 1.20 5 21.32
128 28:7 1.30 5 25:83 1.30 5 22.02
240 53:8 1.30 5 28:27 1.30 5 23.13

is studied here increasing the number of levels from 3 to 6. We consider again a

slab domain. The size of the fine mesh varies form 16  16  8 elements (5 202
d.o.f.) with 3 levels to 128  128  64 elements (2 163 330 unknowns) with 6
levels, hence the dimensions of the slab domain increase from 0:16  0:16  0:08 to
1:28  1:28  0:64 cm3. The number of subdomains (and processors) is kept fixed
to 32 D 4  4  2. We run the model for 0.5 ms after the stimulus. Table 8.12
reports the average minimum and maximum eigenvalues, the average condition
number, the average number of PCG iterations and the average solving time per
time step. As in the two-dimensional case, the condition number and the iteration
counts are independent of the number of levels, result that confirms the optimality
of the MHS(L) preconditioner.
Test MHS-3D-2. Scaled speedup on ellipsoidal domains. We consider in this
case an ellipsoidal geometry (See Fig. 8.1). We vary the number of subdomains
from 8 to 240, keeping fixed the local mesh in each subdomain to 48  48  48
nodes (221 184 unknowns), hence varying the global number of degrees of freedom
(d.o.f.) from 1:7  106 in the smallest case with 8 subdomains to 5:3  107 in
the largest with 240 subdomains. We simulate the initial depolarization phase for
0.5 ms, i.e. 10 time steps. The number of levels varies from 3 to 6. Tables 8.13
and 8.14 report the average condition number, the average number of PCG iterations
and the average solving time per time step. These results show the scalability of the
MHS preconditioner also in a three-dimensional deformed domain. The condition
numbers and iteration counts are essentially the same for MHS with 3,4,5 levels
and only marginally larger with 6 levels. The best performance in term of CPU time
8.5 Numerical Results for Multilevel Schwarz Preconditioners 239

Table 8.14 Test MHS-3D-2. 5 levels 6 levels

Same format as in Table 8.13
Procs. d.o.f. (M)  it. Time  it. Time
for 5, 6 level MHS
8 1:8 1.18 4 15.98 1.38 5 16.86
16 3:6 1.18 4 16.97 1.38 5 17.07
32 7:2 1.21 5 21.71 1.43 6 21.32
64 14:4 1.21 5 21.62 1.39 5 20.96
128 28:7 1.30 5 21.36 1.56 6 22.81
240 53:8 1.30 5 22.82 1.56 6 24.65

Table 8.15 Test MHS-3D-3. Presence of transmural ischemia: comparison of the three precon-
ditioners BJ, MG(5) and MHS(5) in presence of discontinuous conductivity coefficients due to
ischemia
BJ MG(5) MHS(5)
Procs.  it. time  it. Time  it. Time
8 3 060 156 33:61 2.86 9 16:58 2.33 8 16.73
16 3 384 163 17:48 2.96 10 10:64 2.39 8 8.58
32 3 500 167 10:16 2.95 10 5:14 2.41 8 5.47
64 4 268 183 5:68 3.35 11 3:47 2.83 9 3.44

seems to be attained with 5 levels, hence in the following tests we focus on the
MHS(5).
Test MHS-3D-3. Presence of transmural ischemia. The domain considered
in this case is an anisotropic slab of dimensions 1:28  1:28  0:32 cm3 , with a
transmural ischemic region of dimensions 0:320:320:32 cm3 located at the center
of the slab. The portion of tissue is discretized by a cartesian grid of 128  128  32
elements (1 098 306 unknowns). The excitation process is started by applying a
stimulus of 200 A=cm3 for 1 ms on a small area of 2  2  2 elements at a vertex of
the endocardium. The ischemic condition is modeled by increasing the extracellular
concentration of potassium in the LR1 model from 5.4 mV (control) to 20 mV
(ischemia) (see [569]) and reducing the conductivity coefficients in the ischemia
region as indicated in Table 8.1 (see [235]), hence the conductivity coefficients
present discontinuities on the boundaries of the ischemic region and this makes the
resolution more difficult. See also Chap. 3 for more details on modeling ischemic
conditions.
We compare MHS(5) with the standard one-level Block Jacobi preconditioner
(BJ) and the multigrid preconditioner with 5 levels (MG(5)) proposed in [552].
We observe that the hybrid method is also in this case scalable and comparable
to (sometimes better than) the multigrid preconditioner MG(5) (Table 8.15).
MHS-3D-4. Complete cardiac cycle. In this last test, we simulate a complete
heartbeat (400 ms, about 3 000 time steps) in a portion of ventricle having dimension
1:92  1:92  0:48 cm3 , discretized by a cartesian grid of 192  193  49 nodes
(3:6  106 d.o.f.). We run the simulation on 36 processors of the Linux cluster of
Milan (cluster.mat.unimi.it). Table 8.16 reports the average PCG iterations per time
240 8 Parallel Solvers for the Bidomain System

Table 8.16 Test MHS-3D-4. Prec. it. Time (s) Total time
Complete cardiac cycle
BJ 205 46:02 29 h 49 min
MG(5) 8 11:11 7 h 21 min
MHS(5) 6 9:67 6 h 26 min
it average PCG iterations per time step,
time average execution time per time step
in seconds, total time total simulation time

260 20

240
18 MG(5)
MHS(5)
220
16
ITERATIONS

200
14
180
12
160
10
PCG

140
8
120

100 6

80 4
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400

TIME (ms) TIME (ms)

Fig. 8.7 Test MHS-3D-4. Time evolution of the PCG iterations with BJ preconditioners (left) and
multilevel preconditioners MG(5), MHS(5) (right)

TRANSMEMBRANE POTENTIAL EXTRACELLULAR POTENTIAL

−96.04 38.41 2.00 −30.77 37.91 2.00

−80 −60 −40 −20 0 20 −20 −10 0 10 20 30

Fig. 8.8 Test MHS-3D-4. Patterns of level lines of the transmembrane and extracellular potentials
during the excitation phase (t = 40 ms). Reported below each panel are the minimum, maximum
and step in mV of the displayed map

step, the average execution time per time step and the total simulation time. The
detailed iteration counts as function of time during the complete heartbeat are shown
in Fig. 8.7 (left panel for BJ and right panel for MG(5) and MHS(5)). Figure 8.8
shows the spatial maps of the transmembrane and extracellular potentials computed
40 ms after the stimulus was given at a vertex of the domain, i.e. during the
excitation phase. Figure 8.9 shows the transmembrane and extracellular potentials
8.6 Block Preconditioners for the Bidomain System 241

TRANSMEMBRANE POTENTIAL EXTRACELLULAR POTENTIAL

50 30
BJ
MG(5) 20
MHS(5)
0 10
mV

−50 −10

−20

−100 −30
0 100 200 300 400 0 100 200 300 400
TIME (ms) TIME (ms)

Fig. 8.9 Test MHS-3D-4. Time evolution at a fixed point of the transmembrane and extracellular
potentials, computed with the three methods

computed in a fixed point of the domain by the three methods (the three plots are
indistinguishable because superimposed). MHS(5) reduces the computational time
of about 78 and 12 %, compared to BJ and MG(5) respectively.

8.6 Block Preconditioners for the Bidomain System

Block preconditioners can be used, and combined with Schwarz preconditioners,

for both the parabolic-parabolic (PP, see 3.42) and parabolic-elliptic (PE, see 3.43)
formulations of the Bidomain system. Such block preconditioners for the Bidomain
system have been studied e.g. in [373, 379, 381, 382].
We recall that after the time and space discretization described in Chap. 7, the
Bidomain discrete systems in both PP (7.5) and PE (7.6) formulations involve
matrices with a 2  2 block structure

M C Ai  M M C Ai Ai
APP D ; APE D ;
 M M C Ae Ai A
Cm
where D is proportional to the inverse time step size, see (7.9).

8.6.1 Block-Diagonal and Block-Factorized Bidomain

Preconditioners

Denoting by

A11 A12
AD
A12 A22
242 8 Parallel Solvers for the Bidomain System

either one of APP , APE , we will consider the following classical block precondition-
ers for A (see Axelsson [20]):
• The block-diagonal preconditioner

B1 0
BD D I (8.38)
0 B2

• The block-factorized preconditioner

I 0 B1 A12
BF D : (8.39)
A12 B1
1 I 0 B2

The main abstract results for these preconditioners are given in the following two
propositions, see Axelsson [20] for a proof. For both propositions, we define the
constant

vT A12 A1
11 A12 v
2
D sup T
(8.40)
v2Rn nKer.A22 / v A22 v
q
and the function .x/ D 12 .1 C x/ C 1
4 .1  x2/ C x 2.

Proposition 8.1 (Axelsson [20, Th. 9.3]). If ˛1 A11  B1  ˛2 A11 , ˇ1 A22  B2 

ˇ2 A22 , with ˛2  ˇ2 , then
   
1 ˛2 ˛1 ˇ2
.BD A/  :
˛1 .1  2/ ˇ1 ˛2

Proposition 8.2 (Axelsson [20, Th. 9.5]). If ˛1 A11  B1  ˛2 A11 , ı1 A22  S2 

ı2 A22 , where S D B2 C A12 B1
1 A12 , with ˛2  1  ˛1 >
2
, ı2  1  ı1 > 2 ,
then
 1
1 maxf˛2 ; ı2 g  1
min .B A/  1 C .r 2 / ;
F
1 2

1 ı2 1
where r2 D minf ˛ı221 ; ˛2 1 g, and ˛2 > 1 and/or ı2 > 1,

 1
1 1  minf˛1 ; ı1 g
max .BF A/  1 .r1 / ;
1 2

where r1 D minf 1˛

1ı1 ; 1˛1 g, and ˛1 < 1 and/or ı1 < 1.
1 1ı1
8.6 Block Preconditioners for the Bidomain System 243

We now define the diagonal blocks B1 ; B2 of our block precondition-

ers (8.38), (8.39) as scalar overlapping Schwarz preconditioners for each scalar
component. Therefore, for the PP formulation, we define:

B1 1
1 D scalar BMAS preconditioner for ct M C Ai ;

B1 1
2 D scalar BMAS preconditioner for ct M C Ae ; (8.41)

while for the PE-formulation, we define

B1 1
1 D scalar BMAS preconditioner for ct M C Ai ;

B1 1
2 D scalar BMAS preconditioner for Ai C Ae : (8.42)

We could also build our scalar blocks using the hybrid multilevel preconditioner
1 1
BMHS instead of BMAS .
The following convergence rate bound for the block-diagonal preconditioner can
be found in [373].
Lemma 8.6. For both Bidomain PP and PE formulations (A D APP or A D APE ),
the condition number of the block-diagonal preconditioned operator with MAS
scalar blocks (8.41), (8.42) is bounded by
 
1 hk1 1 C
.BD A/  c max 1C :
1kL1 ık 1

This result shows that in addition to the standard domain decomposition param-
eters, convergence with the block-diagonal preconditioner BD depends on the
parameter , which in turn depends only on the original Bidomain blocks. The
following estimates on 2 can be found in [381]:
(a) For the PP formulation, 2  .1 C min .Ae ; M//1 , where min .Ae ; M/ is
the minimum eigenvalue of Ae M1 (see [381, Lemma 4.3]) and numerical
experiments show that is very close to 1, so the bound of Lemma 8.6 might
be large and convergence slow;
(b) For the PE formulation, 2  .1 C min .Ae ; Ai //1 , where min .Ae ; Ai / is
the minimum eigenvalue of Ae A1 i (see [381, Lemma 4.1]) and numerical
experiments seem to indicate that is close to 1=2, so the bound of Lemma 8.6
is satisfactory.
The following convergence rate bound for the block-factorized preconditioner
can be found in [373].
Lemma 8.7. For both Bidomain PP and PE formulations (A D APP or A D APE ),
the extreme eigenvalues of the block-factorized preconditioned operator with MAS
244 8 Parallel Solvers for the Bidomain System

scalar blocks (8.41), (8.42) are bounded by

0   11
hk1
B c max 1 C C N c C
1 B 1kL1 ık C ;
min .BF A/  @1 C 2 A
1 2

 1
1 1  .Nc C 1/1
max .B A/  1  :
F
1

These bounds are pessimistic because, due to Prop. 8.2, they require that 2 < ı1
and predict a large condition number when 2 approaches ı1 , while our numerical
results seem to indicate that the same considerations on for the block-diagonal
case also hold for the block-factorized case.

8.6.2 Numerical Results with Block Preconditioners

In this section, we present the results of parallel numerical experiments performed

on Linux Clusters using the parallel library PETSc [24, 25], from the Argonne
National Laboratory.
The Parabolic-Parabolic (PP) and Parabolic-Elliptic (PE) formulations of the
Bidomain system coupled to the LR1 model are integrated by the Implicit-Explicit
Method described in the previous sections. The values of the coefficients and
parameters in the LR1 model are given in the original paper [308]. The linear
system at each time step is solved by the preconditioned conjugate gradient (PCG)
method, using as stopping criterion a 104 reduction of the residual l 2 -norm.
We precondition the PCG iteration by the multilevel Schwarz preconditioner in
either its additive version MAS(L) with L levels defined in (8.34) or its hybrid
version MHS(L), studied in [458, 459]. These preconditioners are employed both
as coupled preconditioners for the whole PP and PE Bidomain systems and as block
preconditioners in the block-diagonal and block-factorized preconditioners. Inexact
ILU(0) local solvers are used for the local problems on the subdomains.
Domain geometry and fiber structure. The domain ˝ is the image of a
cartesian slab using ellipsoidal coordinates, yielding a portion of truncated ellipsoid
(see Fig. 8.10). The family of truncated ellipsoids is described by the parametric
equations
8
< x D a.r/ cos  cos min   max ;
y D b.r/ cos  sin min    max ;
:
z D c.r/ sin  0  r  1;
8.6 Block Preconditioners for the Bidomain System 245

procs. 8 16 32 64 128 256

2·2·2 4·2·2 4·4·2 8·4·2 8·8·2 16 · 8 · 2
96·96·96 192·96·96 192 ·192 ·96 384 ·192 ·96 384 ·384 ·96 768 ·384 ·96
d.o.f. 1.8M 3.6M 7.2M 14.4M 28.7M 57.4M

Fig. 8.10 Mesh parameters for weak scaling Tests 1 and 2

where a.r/ D a1 C r.a2  a1 /; b.r/ D b1 C r.b2  b1 /; c.r/ D c1 C r.c2  c1 /; and

ai ; bi ; ci ; i D 1; 2 are given coefficients determining the main axes of the ellipsoid.
The fibers rotate intramurally linearly with the depth for a total amount of 120ı
proceeding counterclockwise from epicardium to endocardium. More precisely, in
a local ellipsoidal reference system .ue ; ue  ; ue r /, the fiber direction al .x/ at a point
x is given by

2 
al .x/ D ue cos ˛.r/Cue  sin ˛.r/; with ˛.r/ D .1r/ ; 0  r  1:
3 4
Conductivity coefficients. The values of the conductivity coefficients used in all
the numerical tests are the following:

li D 3  103 1 cm1 le D 2  103 1 cm1

t D 3:1525  10  cm t D 1:3514  103 1 cm1
i 4 1 1 e

ni D 3:1525  105 1 cm1 ne D 6:757  104 1 cm1 :

Mesh hierarchy. We denote the cartesian mesh used to discretize our domains
by T D Ti Tj Tk , indicating the number of elements in each coordinate direction.
This notation applies to both fine and coarse meshes. When we scale up the mesh by
a factor c, for brevity we define cT D cTi  cTj  cTk , i.e. the number of elements
in cT is c 3 times the number of elements in T . Figure 8.10 shows the domains
used in our parallel tests described below, reporting for each test the number of
processors (procs.), the coarse mesh (T0 ), the fine mesh (TL1 ), and the number of
degrees of freedom (d.o.f). Our tests will range from 1.8 million d.o.f. on 8 procs.
to 57.4 million d.o.f. on 256 procs.
Stimulation site, initial and boundary conditions. The depolarization process
is started by applying a stimulus of Iapp D 200 mA=cm3 lasting 1 ms on the face of
the domain modeling the endocardial surface. The initial conditions are at resting
values for all the potentials and LR1 gating variables, while the boundary conditions
are for insulated tissue. In all simulations, the fine mesh size is h D 0:01 cm. The
time step size is t D 0:05 ms.
246 8 Parallel Solvers for the Bidomain System

1e5 400
diagonal
350
diagonal
1e4
factorized 300

1e3 250
κ2

it.
200
1e2 150 factorized

100
10
50
coupled coupled
1 0
816 32 64 128 256 816 32 64 128 256
procs. procs.

Fig. 8.11 Test 1, weak scaling on ellipsoidal domains of Fig. 8.10 for PP formulation with
coupled, diagonal and factorized MHS(4) preconditioners. Top: table with condition numbers (),
extreme eigenvalues ( max ; min ), iteration counts (it.) for each preconditioner as a function of the
number of processors/subdomains (procs). Bottom: plots of condition numbers (left) and iteration
counts (right) from the table above

8.6.2.1 Test 1: Weak Scaling for Coupled, Diagonal, Factorized MHS

Preconditioners Applied to the PP Formulation

In this test, we compare the performance of coupled, diagonal, factorized MHS

preconditioners applied to the PP formulation. In the table of Fig. 8.11, a weak
scaling test on increasing ellipsoidal domains is considered. The number of levels is
kept fixed to 4. The results show that the MHS-coupled preconditioner is completely
scalable, with the PCG iterations which remain constant when increasing the
number of processors. On the other hand, both diagonal and factorized MHS
preconditioners are not scalable, because condition numbers and PCG iterations
increase considerably when the number of processors grows.

8.6.2.2 Test 2: Weak Scaling for Coupled, Diagonal, Factorized MHS

Preconditioners Applied to the PE Formulation

As in Test 1, we compare here the performance of coupled, diagonal, factorized

MHS preconditioners, but applied to the PE formulation. The same weak scaling
test on increasing ellipsoidal domains is considered, as in the previous Test 1.
The number of levels is kept fixed to 4. The results in the table of Fig. 8.12
show that now the coupled, diagonal and factorized MHS preconditioners are
all completely scalable, with the PCG iterations which remain bounded when
increasing the number of processors. The best convergence is achieved by the MHS-
coupled preconditioner (4 PCG iterations), the worst by the MHS-diagonal (21 PCG
iterations).
8.6 Block Preconditioners for the Bidomain System 247

18

16 20 diagonal
diagonal
14
factorized
factorized 15
12

10
κ2

it.
8 10

4 5

2 coupled coupled

0 0
816 32 64 128 256 816 32 64 128 256
procs. procs.

Fig. 8.12 Test 2, weak scaling on ellipsoidal domains of Fig. 8.10 for PE formulation with
coupled, diagonal and factorized MHS(4) preconditioners. Top: table with condition numbers (),
extreme eigenvalues ( max ; min ), iteration counts (it.) for each preconditioner as a function of the
number of processors/subdomains (procs). Bottom: plots of condition numbers (left) and iteration
counts (right) from table above

8.6.2.3 Test 3: Strong Scaling for Coupled, Diagonal, Factorized MHS

Preconditioners Applied to the PP and PE Formulations

In this strong scaling (standard speedup) test, we compare again the performance
of coupled, diagonal, factorized MHS preconditioners applied to both PP and PE
formulations. The number of levels is kept fixed to 4. The cardiac domain ˝ is the
portion of ellipsoid in Fig. 8.10, discretized by a fine mesh of 128  128  64 finite
elements, for a total amount of 2 163 330 dof. Since this is a strong scaling test,
the fine mesh is fixed, while the number of subdomains (D number of processors)
increases from 8 to 512. In this way, the subdomain size is reduced when the number
of subdomains is increased. The simulation is run for 0.5 ms during the excitation
phase, i.e. for 10 time steps.
Table 8.17 reports the average condition number, PCG iterations count and CPU
time per time step. The results show that the coupled preconditioners for both PP
and PE formulations have comparable performances and they are better than the
block preconditioners, especially for the PP formulation. The speedups, defined
with respect to the base 8 processors run, are good for the coupled preconditioners
up to 128 processors (up to 64 processors they are even superlinear for the PP
formulation), while they degenerate with 256 and 512 processors, because the local
problems become too small and the increasing communication costs deteriorate the
parallel performance. The block preconditioners have a much worse performance in
the PP formulation (as already seen in Test 1), showing reasonable speedups up to 64
processors but degenerating rapidly afterward, with even negative return (decreasing
speedups) for 512 processors. For the PE formulation the block preconditioners
show better timings, initially even better than the coupled preconditioner since
each iteration of the latter is more expensive, but such timings worsen rapidly
after 128 processors and become triple the coupled preconditioner timings for 512
248 8 Parallel Solvers for the Bidomain System

Table 8.17 Test 3, strong scaling test on ellipsoidal domain for PP and PE formulations with
coupled, diagonal and factorized MHS(4) preconditioners
MHS(4)-coupled MHS(4)-diagonal MHS(4)-factorized
Procs.  it. Time  it. Time  it. Time
PP formulation
8 2.56 7 8.4 1.1e C 5 239 88.6 3.6e C 3 90 49.2
16 2.63 7 3.6 1.3e C 5 255 51.0 3.4e C 3 89 26.5
32 2.63 7 1.9 1.3e C 5 254 24.8 3.4e C 3 89 13.0
64 2.63 7 1.0 1.3e C 5 254 16.5 3.4e C 3 89 6.8
128 3.20 8 0.6 2.1e C 5 319 10.9 4.4e C 3 96 4.8
256 3.21 8 0.5 2.1e C 5 317 10.1 4.5e C 3 96 4.7
512 3.21 8 0.5 2.1e C 5 316 12.8 4.5e C 3 96 7.9
PE formulation
8 2.61 7 8.2 27.62 25 5.9 21.82 21 6.1
16 2.68 7 4.0 27.60 25 3.3 21.62 22 3.3
32 2.68 7 2.0 27.60 25 1.7 21.62 22 1.6
64 2.68 7 1.4 27.60 25 1.0 21.62 22 0.9
128 3.27 8 0.8 27.16 25 0.9 19.67 23 0.8
256 3.29 8 0.6 27.16 25 1.1 19.70 23 0.9
512 3.29 8 0.5 27.16 25 1.5 19.70 23 1.4
Condition numbers (), iteration counts (it.) for each preconditioner, CPU times (time) in seconds
as a function of the number of processors/subdomains (procs)

processors. This is also confirmed by the low speedups that degenerate rapidly and
start decreasing already at 256 processors.
In conclusions, these results show that block preconditioners for the Bidomain PP
formulation are not scalable, while they are scalable for the PE formulation, but with
higher iteration counts and computational costs than the coupled preconditioner.
Therefore, in our parallel tests (with idealized ventricular geometries and structured
finite elements) multilevel Schwarz preconditioners seem to be more efficient when
applied as coupled rather than block Bidomain preconditioners.
Chapter 9
Simulation Studies of Cardiac Bioelectrical
Activity

Anisotropic Bidomain computer simulations have been used to study cardiac

conduction since the beginning of the 1980s. At that time, computational power
limited investigations to simple geometries corresponding to two-dimensional
sheets [28, 392] and idealized three-dimensional cylindrical domains, which for
symmetry reduce to two-dimensional domains [426, 434]. With the evolution
of computer technology and in particular the development of parallel computer
platforms, Bidomain numerical simulations of the cardiac bioelectrical activity
have become three-dimensional [101, 102, 226, 295, 397] and increasingly more
realistic [399, 452, 526, 530], allowing a wider range of investigations to elucidate
the mechanisms underlying the propagation of electrical impulse in cardiac tissue
under normal and pathological conditions.
In this chapter, we illustrate how the orthotropic Bidomain system described in
the previous chapters can be used to simulate and study the space-time evolution
of the most important phenomena in cardiac electrophysiology. We consider three-
dimensional blocks modeling a portion of ventricular wall. Three-dimensional
models have been developed also for atrial tissue, accounting for both spatial
heterogeneity and anisotropy and they have been coupled with a human torso to
simulate body surface maps and electrocardiograms; for studies focusing on the
propagation in normal and fibrillating atria, we refer to e.g. [17, 219, 253, 466].
In the following, we will focus on:
• The genesis of cardiac excitation and virtual electrode polarization;
• The anisotropic propagation of excitation and recovery fronts;
• The effects of cardiac heterogeneities on fronts propagation and APD
distribution;
• The morphology of electrograms;
• The computation of excitation and repolarization time markers;
• The effects of ischemic regions;
• The simulation of cardiac reentry phenomena.

© Springer International Publishing Switzerland 2014 249

P. Colli Franzone et al., Mathematical Cardiac Electrophysiology, MS&A 13,
DOI 10.1007/978-3-319-04801-7__9
250 9 Simulation Studies of Cardiac Bioelectrical Activity

9.1 Cardiac Excitation and Virtual Electrode Phenomena

It is well known that the macroscopic electrical properties of the cardiac muscle
are markedly anisotropic. Experimental studies have confirmed that the Bidomain
representation of the cardiac tissue as superimposition of two continuous conductive
media, the intracellular and extracellular spaces, exhibits different anisotropy ratios
on the order of 2 and 10, with the intracellular domain the most anisotropic one; see
e.g. [100,277,420,421]. The development of optical mappings of cardiac transmem-
brane action potential, starting in the mid 1990s, has led to experimental studies by
several groups that have investigated the effects of unipolar extracellular cathodal or
anodal stimulations of cardiac tissue, see e.g. [156, 157, 281, 349, 480, 559]. These
studies have established that the stimulation by a unipolar electrode produces a
characteristic transmembrane pattern called virtual electrodes response.
After an anodal stimulus, the transmembrane potential distribution exhibits
a virtual anode (VA), i.e. negatively polarized (hyperpolarized) volume around
the stimulating electrode having a dog-bone shape, and by two virtual cathodes
(VCs), i.e. positively polarized (depolarized) regions adjacent to the concave part
of the hyperpolarized anodal dog-bone boundary, see e.g. Fig. 9.4. The central
dog-bone VA is aligned across the fiber direction, while the two adjacent VCs
are aligned along the fiber direction. Conversely, after a cathodal stimulus, the
polarity is reversed, i.e. the transmembrane potential pattern exhibits a central dog-
bone shaped VC aligned across fiber and two adjacent VAs aligned along fiber. It
is well known that only Bidomain models with unequal anisotropy ratios of the
intra- and extracellular media are able to generate the observed virtual electrode
polarization regions, see e.g. [469, 557]. Therefore, the experimental detection of
virtual electrode responses by [281, 349, 559] strongly supports the evidence for the
unequal anisotropy ratios assumption in Bidomain models; see the recent survey of
the Bidomain theory of pacing in [256].
The heart can be stimulated electrically in four different ways, see e.g. [146,207,
300, 301]:
• By turning on a negative current (cathode make, CM);
• By turning on a positive current (anode make, AM);
• By turning off of negative current (cathode break, CB);
• By turning off of positive current, (anode break, AB).
See Fig. 9.1 for a schematic illustration of these four cardiac stimulations.
In [427, 435], the Bidomain model with unequal anisotropic ratio was first
proposed and used to establish a theoretical framework able to explain the make
and break mechanisms of excitability in terms of the underlying virtual electrodes
polarization. The effects and features of make and break excitation mechanisms,
generated by unipolar extracellular cathodal or anodal stimulations, have been
subsequently investigated by simulation studies in [164,390,416,428–430,432,436,
450,451,486] and by experimental studies in [156,157,480,559]; see also the recent
surveys [256, 558] and the book [158].
9.1 Cardiac Excitation and Virtual Electrode Phenomena 251

CATHODE MAKE CATHODE BREAK

applied current

applied current
excitation
excitation

time time

ANODE MAKE ANODE BREAK

excitation excitation
applied current

applied current

time time

Fig. 9.1 Schematic illustration of the cathode make (CM), cathode break (CB), anode make (AM),
anode break (AB) excitation mechanisms

We briefly recall the main activation features of these mechanisms. With the
make stimulation, activation occurs with the stimulus onset either at the central VC
during cathodal stimulation (cathode make CM) or at the two VCs along the fiber
direction during anodal stimulation (anode make AM). With the break stimulation,
activation occurs after the stimulus termination at the two VAs during cathodal
stimulation (cathode break CB) or at the central VA during anodal stimulation
(anode break AB).
The break modes CB and AB generated by local electrode stimulation have been
confirmed by optical recordings and Bidomain simulations when a short stimulation
is delivered during the relative refractory period (systolic phase), whereas the make
modes CM and AM have been observed experimentally and simulated during the
diastolic phase. Many studies have underlined the important role played by the
anode break excitation in reentry induction mechanisms [429] and defibrillation
[11, 156, 157, 486, 525].
Here we focus on the 3D investigation of the excitation mechanisms associated
with an anodal stimulation, which can be generated only in tissues with unequal
anisotropy ratio. Additional simulations and results can be found in [131, 132].
252 9 Simulation Studies of Cardiac Bioelectrical Activity

9.1.1 Methods and Parameter Calibration

The anisotropic Bidomain model. Let us consider a three-dimensional domain ˝H

modeling the cardiac tissue, in contact with a conducting medium ˝0 , representing
either the intracavitary blood or an extracardiac bath. We define ˝ D ˝H [ ˝ 0
and S D @ \ @˝0 the common not insulated interface. The remaining interface
@˝ is assumed insulated. The evolution of the transmembrane potential v.x; t/,
extracellular potential ue .x; t/, extracardiac potential ub .x; t/, gating variables
w.x; t/ and ionic concentrations c.x; t/ is described by the macroscopic Bidomain
model:
8
ˆ
ˆ cm @t v  div.Di rv/  div.Di rue / C iion .v; w; c/ D 0 in ˝H
ˆ
ˆ
ˆ
ˆ @ w  R.v; w/ D 0; @t c  S.v; w; c/ D 0 in ˝H
ˆ t
ˆ
ˆ
ˆ  div.D C D /ru D div D rv C i e
in ˝H
ˆ T
<
i e e i app
n Di r.v C ue / D 0 on @˝H
(9.1)
ˆ
ˆ  div b rub D iapp
e
in ˝ n ˝H
ˆ
ˆ
ˆ
ˆ ue D ub on @˝H
ˆ
ˆ
ˆ
ˆ nT De rue D nT b rub on @˝H
:̂ T
n b rub D 0 on @˝

with appropriate initial conditions on v.x; 0/; w.x; 0/ and c.x; 0/. Here cm and iion
denote the capacitance and the ionic current of the membrane per unit volume,
iRapp represents the applied current per unit volume with the compatibility condition
˝ iapp D 0, and b is the conductivity coefficient of the extracardiac medium.
e

Choice of reference potential. The extracellular potential ue is defined up to

an independent constant R.t/ determined by the choice of the reference potential.
In this paper, we consider as a reference potentialR the average of the extracellular
potential over the cardiac volume, i.e. we impose ˝H ue .x; t/dx D 0.
Membrane model. In our implementation of the Bidomain model for cardiac
tissue with normal bioelectric properties, the Luo-Rudy phase I (LR1) action poten-
tial model is used to represent the nonlinear kinetics of the ventricular membrane
current according to the original parameters of [308], with the addition of
(i) The funny current If , modeled as in [31, 81, 82, 149, 579];
(ii) The electroporation current Ie , modeled as in [145];
(iii) The outward current Ia activated upon induced depolarization, modeled as
in [88].
We remark that the incorporation of (i) and (ii) have also been carried out in [16]
in order to investigate by means of 2D Bidomain simulations the cardiac tissue
response to electric shocks. The augmented LR1 model with the incorporation of
If , Ie and Ia currents in the following is called full membrane model.
9.1 Cardiac Excitation and Virtual Electrode Phenomena 253

Myocardial conductivity tensors and fiber architecture. We recall that the

macroscopic fiber structure of the cardiac tissue induces anisotropic conductiv-
ity properties along and across the fiber directions. Assuming axial-symmetric
anisotropy, the macroscopic conductivity tensors Di .x/ and De .x/ at any point
x 2 ˝H are defined as

Di;e .x/ D li;e al .x/aTl .x/ C ti;e .I  al .x/aTl .x//; (9.2)

i;e
where al .x/ is a unit vector parallel to the local fiber direction and l;t are the
effective intra and extracellular conductivity coefficients measured along (l) and
across (t) the fiber direction.
Computational domain. The cardiac domain ˝H considered in this study is
a cartesian slab of dimensions 0:96  0:96  0:32 cm3 , modeling a portion of the
left ventricular wall. The endocardial surface is in contact with a smaller slab of
dimensions 0:96  0:96  0:16 cm3 , modeling the extracardiac bath, where b D
6e 3 1 cm1 similar to the blood conductivity.
Stimulation site. Diastolic cathodal or anodal stimuli are delivered at the center
of the epicardial surface in a small region of dimensions 0:06  0:06  0:03 cm3
of a tissue at rest. In order to ensure the compatibility condition, we inject a total
stimulation current in a strip of dimensions 0:960:960:08 cm3 in the extracardiac
bath with equal strength and opposite polarity of that used at the subepicardial
level. Moreover, in order to reduce the effects of the delivered bath current on
the potential field in the subepicardial layer, we have lowered the excitability of
the subendocardial layer by strongly decreasing the maximal sodium conductance
during the stimulation interval. In order to simulate the break excitation mechanisms
during systole, we perform an S1-S2 stimulation protocol. We first apply an initial
S1 cathodal stimulus to the resting tissue and simulate the resulting action potential
propagation, that sweeps the cardiac domain until a suitable time during the relative
refractory period (RRP). At this instant, a subsequent S2 cathodal or anodal stimulus
is delivered and we simulate the resulting action potential propagation where the
surrounding tissue has recovered its excitability properties.
Activation time isochrones. During the Bidomain simulations, we process the
distribution of the transmembrane potential v.x; t/ in order to define the activation
time at.x/ as the first time instant for which v.x; at.x// D vup ; we choose vup D
50 mV, a value above the threshold value of v capable of initiating the upstroke
of the action potential. Then the excitation wavefronts are displayed by drawing the
isochrone surfaces of the activation time.
Numerical methods. In all computations, a structured grid of 96  96  48
hexahedral isoparametric Q1 finite elements of size h D 0:1 mm is used in space,
while the time discretization is based on the following double operator splitting
procedure, based on splitting both the ODEs from the PDEs and the elliptic PDEs
from the parabolic one, similar to the second decoupled semi-implicit method
254 9 Simulation Studies of Cardiac Bioelectrical Activity

described in Sect. 7.2.3, with the addition of the blood volume. Thus, at the general
time step t nC1 , given the transmembrane potential vn at the previous time step, we
find:
1. The gating variables wnC1 , by solving the related ODEs with the Backward Euler
method;
2. The ionic concentration variables c nC1 , by solving the related ODEs with the
Forward Euler method;
3. The extracellular and extracardiac potentials uenC1 and unC1 b , by solving the
algebraic linear system resulting from the finite element discretization of the
elliptic PDEs in (9.1);
4. The transmembrane potential vnC1 , by solving the algebraic linear system
resulting from the finite element discretization of the parabolic PDE in (9.1).
This operator splitting strategy yields two large linear systems of algebraic
equations that must be solved at each time step. In order to ensure parallelization and
portability of our Fortran code, we use the PETSc parallel library [24, 25], a suite of
data structures and functions for building large-scale parallel scientific applications,
based on the MPI communication library. The parallel strategy employed assigns
each subdomain to one processor and the information associated with the interior
of the subdomain is uniquely owned by that processor. The processor stores all
subvectors and a block of the matrices (mass, stiffness) associated to each over-
lapping subdomain. Our parallel code employs different DA (Distributed Arrays)
PETSc objects for representing v on the domain ˝H (tissue) and for unC1 e ; unC1
b on
the domain ˝ (tissue and bath). The two large linear systems at each time step are
solved by a parallel conjugate gradient method, preconditioned by the Multilevel
Hybrid Schwarz preconditioner, described in Chap. 8, for the ill-conditioned elliptic
system (461 041 d.o.f.) and the Block Jacobi preconditioner for the well conditioned
parabolic system (310 497 d.o.f.). These preconditioners are based on the multilevel
PETSc objects PCMG (MultiGrid) with ILU(0) local solvers.

9.1.2 Anode and Cathode Make Mechanisms

Figures 9.2 and 9.3 show the epicardial distribution of the intracellular potential
(top), extracellular potential (middle) and transmembrane potentials (bottom) 2 ms
after the anodal stimulus elicited in a slab without and with transmural fiber rotation,
respectively. In each figure, the panels on the right column refer to a tissue with
the unequal anisotropy ratios of Table 9.1, while the panels on the left column
refer to a tissue slab with equal anisotropy ratio. In the left panel of Fig. 9.2, the
intra- and extracellular potential elliptical patterns have the same major to minor
semi-axis ratio, yielding a transmembrane potential pattern (bottom left panel) with
more rounded elliptical lines around the anodal electrode. Conversely, in a slab
with unequal anisotropy ratio, the intra- and extracellular equipotential contour
lines, displayed in top and middle right panels of Fig. 9.2, have elliptical shapes
9.1 Cardiac Excitation and Virtual Electrode Phenomena 255

intracellular potential intracellular potential

20 20

0
0
−20
−20
−40

−40
−60

−80 −60

−100
−80

−120
−100
−140
−120

−157.37 270.81 10.00 −123.88 215.32 10.00

extracellular potential extracellular potential

100 100

80
80

60
60
40

40
20

0 20

−20
0

−40
−20
−60

−71.49 503.33 5.00 −37.84 491.53 5.00

transmembrane potential transmembrane potential

−55
−100
−60

−120 −65

−70
−140

−75
−160
−80

−180
−85

−200 −90

−95
−220
−100
−232.52 −85.07 10.00 −276.21 −54.83 5.00

Fig. 9.2 Slab without transmural fiber rotation. Epicardial distribution of the intracellular potential
(top), extracellular potential (middle) and transmembrane potentials (bottom) 2 ms after an anodal
stimulus. Right panels: tissue with the unequal anisotropy ratios of Table 9.1. Left panels: tissue
with equal anisotropy ratio. Below each panel are reported the minimum, maximum and step of
the displayed map and the colorbar denotes the range of values of the displayed equipotential lines
256 9 Simulation Studies of Cardiac Bioelectrical Activity

intracellular potential intracellular potential

20 20

0 0

−20 −20

−40
−40

−60
−60

−80
−80

−98.55 220.16 10.00 −93.87 120.38 10.00

extracellular potential extracellular potential

30 30

25 25

20 20

15
15

10
10
5
5
0

0
−5

−10 −5

−12.68 452.39 5.00 −7.95 334.97 5.00

transmembrane potential
transmembrane potential
−86
−70

−88
−75
−90

−80
−92

−85
−94

−96 −90

−98 −95

−100
−232.22 −85.07 2.00 −214.59 −66.81 2.00

Fig. 9.3 Slab with transmural fiber rotation. Epicardial distribution of the intracellular potential
(top), extracellular potential (middle) and transmembrane potentials (bottom) 2 ms after an anodal
stimulus. Right panels: tissue with the unequal anisotropy ratios of Table 9.1. Left panels: tissue
with equal anisotropy ratio. Below each panel are reported the minimum, maximum and step of
the displayed map and the colorbar denotes the range of values of the displayed equipotential lines
9.1 Cardiac Excitation and Virtual Electrode Phenomena 257

with different ratio of the two semiaxis. The difference of these two epicardial
patterns yields a transmembrane potential distribution with the typical virtual
electrode response (Fig. 9.2, bottom right panel). In fact, the tissue within and
around the anodal electrode is negatively polarized (hyperpolarized) and it exhibits
an epicardial dog-bone shape, developing perpendicularly to the epicardial fiber
direction of 45ı . Two regions of positive polarization (depolarization), i.e. two
virtual cathodes, develop along the fiber direction adjacent to the concave parts of
the epicardial virtual anode boundary.
The same anodal stimulation was applied to a tissue slab with transmural fiber
rotation. In case of unequal anisotropy ratio, see right panels of Fig. 9.3, the
epicardial transmembrane potential pattern exhibits a virtual electrode response
with a twisted hyperpolarized dog-bone shaped region, which is not symmetric
with respect to the epicardial fiber direction because of the counterclockwise
fiber rotation from epicardium to endocardium. In fact, the equipotential surfaces
inside the virtual cathode volumes, shown in Fig. 9.4, are shaped as two horns
pointing counterclockwise when proceeding from the upper (epicardial) face to
the lower (endocardial) face. On intramural sections parallel to the epicardial
face, the equipotential lines inside the virtual anode preserve the same dog-bone
shape as on the epicardium but with a counterclockwise rotation, thus reducing

Table 9.1 Conductivity coefficients of the intra and extracellular media in S cm1 and anisotropy
ratios lti;e D li;e =ti;e , tn
i;e
D ti;e =ni;e
Unequal anisotropy
Medium li ti ni lti i
tn
Intra 2.31724e3 2.43504e4 5.69083e5 9.51622 4.27889
Medium le te ne lte e
tn
Extra 1.54483e3 1.04385e3 3.7221e4 1.47993 2.80447
Equal anisotropy
Medium li ti ni lti i
tn
Intra 1.34511e3 3.36278e4 8.40695e5 4 4
Medium le te ne lte e
tn
extra 5.35282e3 1.3382e3 3.34551e4 4 4

Fig. 9.4 Slab with transmural fiber rotation. Anodal stimulation applied to a slab with unequal
anisotropy ratio. Two different views of the isopotential surfaces of the transmembrane potential
distribution of values 88 and 75 mV, 2 ms after the beginning of the stimulation
258 9 Simulation Studies of Cardiac Bioelectrical Activity

their area and yielding a twisted tote bag; see Fig. 9.4. On the contrary, the same
anodal stimulus applied to a slab with equal anisotropy ratios and transmural fiber
rotation does not yield virtual cathodal regions but only an slightly twisted elliptical
hyperpolarized region centered at the stimulation site; see Fig. 9.3, bottom left panel.
This confirms that virtual electrodes can only be generated by bidomain models
with unequal anisotropy ratios of the intra- and extracellular media. The formation
of the two virtual cathodes elicited by the anodal stimulation explains why an anodal
stimulation is able to excite the myocardium, as shown by the excitation sequence
shown in Figs. 9.5 and 9.6 described below.
Anode make mechanism of excitability. We apply an anodal stimulus, lasting
10 ms and with an amplitude of 0:1112 mA, to the center of the epicardial face
of a tissue slab at rest with unequal anisotropy and with the full membrane
model. Figure 9.5 shows the snapshots of the simulated transmembrane potential
distributions on the epicardial surface and on the transmural diagonal sections.
The densely packed equipotential transmembrane lines indicate the location of the
excitation layers, propagating on the epicardial face and transmurally. The snapshots
in Fig. 9.5 reveal two distinct wavefronts propagating on the epicardial face and
transmurally, which subsequently merge at epicardial and intramural levels. In fact,
distinct activation wavefronts arise from two epicardial locations inside the virtual
cathodes and first propagate in all directions on the epicardial face. In particular,
they propagate outward and inward along the diagonal parallel to the fiber direction,

4 ms 6 ms 8 ms
50 50 50

0 0 0

−50 −50 −50

−100 −100 −100

−368.54 24.43 5.00 −341.37 29.12 5.00 −330.11 29.31 5.00

4 ms 6 ms 8 ms
50 50 50
0 0 0
D1

−50 −50 −50

−100 −100 −100
−371.33 −78.38 5.00 −341.82 −77.69 5.00 −330.11 29.40 5.00

50 50 50
0 0 0
D2

−50 −50 −50

−100 −100 −100
−371.33 23.45 5.00 −341.82 28.73 5.00 −330.11 28.34 5.00

Fig. 9.5 Diastolic anode make excitation. Anodal stimulation during the diastolic interval with
total current Iapp D 0:1177 mA and 10 ms duration, applied to a slab with unequal anisotropic
calibration and funny current. Top panel. Transmembrane potential distributions on the epicardial
section at 3 time instants during the diastolic interval for cathode make stimulation. Bottom panel.
Transmembrane potential distributions on the two diagonal transmural sections of the slab d1 (top)
and d2 (bottom) at 3 time instants during the diastolic interval. Same conditions as in the top panel
(Reproduced with permission from [131])
9.1 Cardiac Excitation and Virtual Electrode Phenomena 259

a b
50 50 30
A1
C
2

A C 0 0 20
2 2

mV

ms
C1
−50 10
−50
C1 A1
A2

−100 0
−100 0 5 10 15 20
−401.13 −50.90 5.00
ms

c d
D1
20 20
18 15
16 10
14 5

12 6.94 21.38 1.00

10 D2
20
8
15
6
10
4
5

2.06 21.38 1.00 2.17 17.91 1.00

Fig. 9.6 Diastolic anode make excitation. Anodal stimulation during the diastolic interval with
total current Iapp D 0:1177 mA and 10 ms duration, applied to a slab with unequal anisotropic
calibration and funny current. (a) Transmembrane potential distribution on the epicardial surface
0:5 ms after the beginning of the stimulation. (b)-left. Transmembrane potential waveforms in
points C1 (blue continuous), A1 (red continuous), A2 (red dashed) and C2 (blue dashed), indicated
in figure (a). (b)-right. Activation time profiles on the diagonal across (blue) and along (red) fibers.
The black line indicates the instant (10 ms), when the stimulus ends. (c). Epicardial isochrones of
activation time. (d). Isochrones of activation time on the two transmural diagonals. Same format
as in Fig. 9.6 (Reproduced with permission from [131])

but, when the excitation isochrones reach points lying on a boundary of the virtual
anode, a block of the inward propagation takes place during the stimulation interval,
as also shown by the large slope of the activation time of Fig. 9.6b inside the anodal
area. Therefore, from about 2 to 6 ms, two distinct activation wavefronts propagate
outward, moving faster along fibers. At about 12 ms, the central epicardial area of
the virtual anode is suddenly activated (0.5 ms), since it becomes excitable only
after the stimulus turn off. Subsequently, all fronts finally merge in a unique large
wavefront.
260 9 Simulation Studies of Cardiac Bioelectrical Activity

Excitation starts at about 2 ms, as shown by the activation time plot along
the diagonal parallel to the epicardial fiber direction, displayed in Fig. 9.6b, thus
inducing an anode make excitation. The isochrones of activation time on the
epicardial surface and transmural diagonal sections (Fig. 9.6c, d) show clearly
that the first triggered areas are the two epicardial sites located at the center
of the virtual cathodes, generating two distinct excitation wavefronts propagating
intramurally and separated by a central region within the virtual anode volume,
which is inexcitable until the stimulus is turned off. These two wavefronts are open
surfaces, each having a closed curve as rim lying on the epicardial surface. There
are epicardial and intramural excitation pathways that, starting from the virtual
cathodes, proceed turning around the boundary of the virtual anode volume and
point toward its epicardial center. These pairs of pathways, coming from the virtual
cathodes, cause the merging of the two wavefronts forming an activated tissue
volume bounded by two intramural surfaces, one consisting of a wavefront moving
toward the epicardial boundary and intramurally toward the endocardial face, and
the other being the conduction block surface surrounding the inexcitable region,
which is suddenly activated after the stimulus is turned off. From the inspection
of the transmural isochrones pattern displayed in Fig. 9.6d, we observe that the
major transmural effect of the virtual electrode polarization is the propagation of
two wavefronts, separated by the transmural part of the virtual anode region, that
merge 5 ms after the beginning of epicardial excitation at 1 mm depth.
Cathode make mechanism of excitability. We now consider the application of
a cathodal stimulus, lasting 1 ms and with an amplitude of 0:0378 mA, to a slab
with unequal anisotropy and full membrane model. Figure 9.7 shows the snapshots
of the transmembrane potential distribution on the epicardial surface and on the two

2 ms 4 ms 6 ms
50 50 50

0 0 0

−50 −50 −50

−100 −100 −100

−84.63 22.10 5.00 −84.51 23.27 5.00 −84.52 25.39 5.00

2 ms 4 ms 6 ms
50 50 50
0 0 0
D1

−50 −50 −50

−100 −100 −100
−84.08 22.10 5.00 −83.74 22.15 5.00 −83.69 23.54 5.00

50 50 50
0 0 0
D2

−50 −50 −50

−100 −100 −100
−84.54 22.10 5.00 −84.39 20.28 5.00 −84.23 21.65 5.00

Fig. 9.7 Diastolic cathode make excitation. Cathodal stimulation with total current Iapp D
0:0378 mA and 1 ms duration, applied to a slab with unequal anisotropic calibration and funny
current. Same format as in Fig. 9.5 (Reproduced with permission from [131])
9.1 Cardiac Excitation and Virtual Electrode Phenomena 261

diagonal transmural sections (bottom panel) at instants subsequent to the end of the
stimulus pulse. The densely packed equipotential transmembrane lines indicates the
location of the excitation layer, propagating on the epicardial face and transmurally.
This excitation wavefront has an approximate ellipsoidal shape with the major
axis and one of the two minor axes parallel and perpendicular, respectively, to the
epicardial fiber direction of 45ı ; the remaining minor axis develops intramurally
through the wall thickness and transversally to fiber layers.
Figure 9.8a illustrates the characteristic pattern of the epicardial transmembrane
potential at 0:5 ms (at the beginning of the stimulation interval), with a dog-bone
shaped virtual cathode, located around the site of the stimulation electrode, and
two adjacent virtual anode areas. A cathode make excitation mechanism occurs,
because activation arises before the end of stimulation as confirmed by the activation
time plots at points A1 ; A2 ; C1 ; C2 along the two epicardial diagonals, displayed in
Fig. 9.8b.

a b
−80 50 30

−81 A
1

A1 C2 −82 0 C 20
1
mV

ms

−83
A2
−50 10
−84 C2
C1 A2
−85

−100 0
−86 0 5 10
−85.83 −66.32 0.20 ms

c d
D1
18

16 15

14 10
5
12

10 0.29 19.09 1.00

8 D2
6 15
4 10
2 5

0.29 19.09 1.00 0.29 17.67 1.00

Fig. 9.8 Diastolic cathode make excitation. Cathodal stimulation during the diastolic interval
with total current Iapp D 0:0378 mA and 1 ms duration, applied to a slab with unequal anisotropic
calibration and funny current. Same format as in Fig. 9.6 (Reproduced with permission from [131])
262 9 Simulation Studies of Cardiac Bioelectrical Activity

In order to describe in detail the origin, development and shape of the excitation
wavefronts, we plot in Fig. 9.8c, d the activation time isochrones on the epicardial
face and on two transmural diagonal sections. Excitation originates at locations
surrounding the physical cathode and propagates outward in all epicardial and
intramural directions, moving faster along fibers and assuming an approximate
semi-ellipsoidal shape. The epicardial velocities of propagation along and across
fibers are about 0:060 and 0:035 cm/ms, respectively, for unequal anisotropy, as also
confirmed by the activation time slope in Fig. 9.8b.

9.1.3 Anode and Cathode Break Mechanisms

In order to study the anodal and cathodal virtual electrode phenomena during
systole, an S1  S 2 stimulation protocol has been considered. The cathodal S1
stimulus is applied at the center of the epicardial surface and the entire activation-
repolarization process is simulated. At 380 ms after the S1 stimulus, during the
relative refractory period (RRP) of the repolarization phase, a cathodal or anodal
S 2 stimulus is applied at the same site of the S1 stimulus pulse. In the following,
we denote by T 1 anisotropy the anisotropic calibration of Table 9.1, and by T 2
anisotropy a calibration with a 20 % reduction of intracellular anisotropy ratios, see
also [131].
Anode break mechanism of excitability. We consider an S 2 anodal stimulus of
amplitude 0:2916 mA and duration 10 ms, in a slab with T 1 anisotropy with the full
membrane model. Figure 9.9 shows the snapshots of the transmembrane potential

10 ms 18 ms 22 ms
50 50 50

0 0 0

−50 −50 −50

−100 −100 −100

−346.99 56.49 5.00 −81.57 10.48 5.00 −82.64 6.95 5.00

10 ms 18 ms 22 ms
50 50 50
0 0 0
D1

−50 −50 −50

−100 −100 −100
−348.80 −49.18 5.00 −81.20 10.48 5.00 −81.50 5.16 5.00
50 50 50
0 0 0
D2

−50 −50 −50

−100 −100 −100
−341.77 56.10 5.00 −80.95 10.24 5.00 −81.57 4.95 5.00

Fig. 9.9 Systolic anode break excitation. Anodal stimulation during the systolic interval (S1-S2
protocol, S2 at 380 ms) with total current Iapp D 0:2916 mA and 10 ms duration applied to a slab
with unequal anisotropy and membrane model with the addition of If ; Ie ; Ia . Same format as in
Fig. 9.5 (Reproduced with permission from [131])
9.1 Cardiac Excitation and Virtual Electrode Phenomena 263

distribution at 10, 18, 22 ms after the end of the S 2 anodal pulse. During the 10 ms
of stimulation, a large dog-bone shaped region of strongly hyperpolarized tissue
(virtual anode) is generated, while two virtual cathodes develop along the fiber
direction near the concave portion of the anodal region, see Fig. 9.10. Although
the virtual cathode volumes are depolarized above threshold (above 20 mV in
Fig. 9.10a), anode make stimulation does not occur, because the surrounding regions
are in the refractory period and still inexcitable, since the sodium inactivation gates
are closed. When the stimulus is turned off, the combined effects of the membrane
funny current If , of the electroporation current Ie and of the outward current Ia ,
depolarize the tissue inside the virtual anode, inducing anode break excitation with
a delay of about 2 ms. The densely packed equipotential lines in the snapshots of
Fig. 9.9 reveal the development of two epicardial excitation wavefronts propagating
mainly in the direction across the epicardial fibers. The intramural diagonal sections
exhibit a strongly different shape of the excitation layers, that originate from

a b 50 30
A1
0 C
1

C
−20 0 A 2 20
A2 C 2
2
mV

ms

−40
C1 A −50 10
1
−60

−80
−100 0
0 10 20 30
−100
ms
−389.42 17.26 5.00

c d D1
30
30 20
10
25

20
11.89 34.75 1.00

15 D2

10 30
20
5
10

0.13 34.75 1.00 0.13 35.98 1.00

Fig. 9.10 Systolic anode break excitation. Anodal stimulation during the systolic interval (S1-
S2 protocol, S2 at 380 ms) with total current Iapp D 0:2916 mA and 10 ms duration applied to a
slab with unequal anisotropy and membrane model with the addition of If ; Ie ; Ia . Same format as
in Fig. 9.6 (Reproduced with permission from [131])
264 9 Simulation Studies of Cardiac Bioelectrical Activity

the center of the virtual anode and then reach and collide on the epicardial and
intramural boundary of the virtual cathode volumes, which are inexcitable. Then
the wavefront rims move around these obstacles until the activated tissue fully
surrounds them.
In order to describe in detail the origin, development and shape of the excitation
wavefronts, the isochrone lines drawn in Fig. 9.10c, d show clearly that the first
activated point is the epicardial central site of the anodal region. Moreover, there
are intramural excitation pathways that, starting from the epicardial anodal areas,
first point toward the endocardium, progressively bend and proceed around the
intramural boundary of the virtual cathodes, and finally point toward the epicardial
face where a wavefront collision happens.
The initial phase of this anode break mechanism is due to the interplay between
the If ; Ie currents, yielding a transmembrane potential above the threshold
excitability inside the virtual anode region, and the sodium channels dynamics.
Therefore, we have a membrane-based trigger excitation mechanism. Apart from
the different triggering process, the excitation wavefront pattern is strongly similar
to the activation sequence generated by a cathodal pulse applied at the center of the
epicardial surface in the presence of two obstacles (i.e. non conducting volumes),
located inside the virtual cathode regions, which are inexcitable.
Figure 9.11e reports the transmembrane potential distributions after the end of an
S 2 anodal stimulus of 0:1512 mA amplitude, lower than in the previous case, and
10 ms duration. After the pulse turns off, the transmembrane potential in the virtual
cathode regions decays toward the resting value, while that in the virtual anode
area, due to the effect of the membrane funny current, depolarizes with a delay of
about 9 ms from the end of the stimulus pulse. The propagating wavefront assumes
an approximate semi-elliptical shape, as in the diastolic anode break excitation,
because the virtual cathodes at the end of the stimulation interval are excitable
again and do not behave like obstacles. The transition between the two systolic
break excitation patterns happens smoothly when decreasing the strength of the
anodal impulse. This is confirmed by the patterns displayed in Fig. 9.11, showing
that decreasing the pulse amplitude the cathodal volumes become re-excitable and
a dimple like inflection appears in the epicardial isochrones associated with slow
propagation along the fiber direction (panel c). Decreasing further the intensity of
the anodal stimulus allows an almost full re-excitability of the virtual cathodes after
the stimulus end, yielding approximate semi-ellipsoidal propagating wavefronts
with flat isochrone lines in the direction along fibers (panel e). Below the threshold
of 0:1512 mA, anode break excitation does not occur.
The systolic anode break mechanism of excitation has been also simulated for a
slab with T 2 anisotropy and a strongly similar excitation sequences, depending on
the amplitude of the stimulus pulse, was observed (not shown). The velocities of
propagation along (across) fibers are 0:042 cm/ms (0:031) for T 1 anisotropy and
0:040 (0:030) for T 2. The degree of anisotropy influences weakly the stimulus
threshold of the anode break excitation, which is slightly higher for T 2 anisotropy
(0:1809 mA) than for T 1 anisotropy (0:1512 mA).
9.1 Cardiac Excitation and Virtual Electrode Phenomena 265

a b
50 30
35

30 A C
1 2

25 0 20

20

mV

ms
A C
2 1
15
−50 10
10

5
−100 0
0.19 35.95 1.00 0 10 20 30
ms

c d
50 30

35
C
2

30 0 A1 20
C
1
mV

ms
25 A2
−50 10

20

−100 0
15 0 10 20 30
14.79 37.72 1.00
ms
e f 50 30
C2

40

0 C 20
35 A 1
1
mV

ms

30
−50 10
A2

25

20 −100 0
0 10 20 30
18.78 43.92 1.00
ms

Fig. 9.11 Systolic anode break excitation. Anodal stimulations during the systolic interval (S1-
S2 protocol, S2 at 380 ms) with total current Iapp D 0:2160 mA (first row), Iapp D 0:1728 mA
(second row), Iapp D 0:1512 mA (third row), and 10 ms duration applied to a slab with
unequal anisotropy and membrane model with the addition of If ; Ie ; Ia . Panels (a-c-e): Epicardial
isochrones of activation time. Panels (b-d-f)-left: Transmembrane potential waveforms in points
C1 (blue continuous), A1 (red continuous), A2 (red dashed) and C2 (blue dashed), indicated in
Fig. 9.8a. Panel (b-d-f)-right: Activation time profiles on the diagonal across (blue) and along
(red) fibers. The black line indicates the instant (10 ms) when the stimulus ends (Reproduced with
permission from [131])
266 9 Simulation Studies of Cardiac Bioelectrical Activity

Cathode break mechanism of excitability. An S 2 cathodal stimulus of

0:2160 mA in amplitude and lasting 10 ms is applied to a slab with T 1 anisotropy
and full membrane current model. The tissue under the cathode is strongly
depolarized and at the end of the S 2 stimulus pulse, i.e. at 380 C 10 D 390 ms,
the transmembrane potential reaches a maximum value of about 185 mV, see
Fig. 9.12. The epicardial transmembrane potential pattern exhibits a cathodal region
around the site of stimulation with a pair of hyperpolarized regions (virtual anodes)
developing along fiber near the concave parts of the cathodal region. Because the
S 2 stimulus is still applied during the relative refractory period, the depolarized
tissue under the cathode is not excitable, thus cathode make excitation does not
occur. After the termination of the stimulus, the hyperpolarized tissue around the
virtual anode areas removes the inactivation of the sodium channels, rendering the
tissue excitable. The electrotonic currents (charge diffusion), coming from strong
depolarization under the cathodal area and directed toward the virtual anode, are
able to originate simultaneously two activation wavefronts arising from the concave
portions of the epicardial boundary of the virtual cathode. Thus, a cathode break
excitation mechanism is induced immediately after the end of the stimulus pulse
with a delay of 0:05 ms, as also confirmed by the inspection of the activation time
plot of Fig. 9.13b.
Figure 9.12 displays the transmembrane potential distribution on the epicardial
surface and on transmural sections at three instants after the start of the S 2
cathodal stimulus, showing that two almost symmetrical wavefronts propagate on
the epicardial face and at subepicardial levels, mainly along the fiber direction,
assuming a semi-dumbbell shape.

10 ms 14 ms 18 ms
50 50 50

0 0 0

−50 −50 −50

−100 −100 −100

−116.68 186.08 5.00 −82.48 13.90 5.00 −83.95 9.05 5.00

10 ms 14 ms 18 ms
50 50 50
0 0 0
D1

−50 −50 −50

−100 −100 −100
−88.01 186.08 5.00 −81.25 2.05 5.00 −81.19 −18.00 5.00

50 50 50
0 0 0
D2

−50 −50 −50

−100 −100 −100
−116.68 185.78 5.00 −81.86 12.25 5.00 −81.74 3.66 5.00

Fig. 9.12 Systolic cathode break excitation. Cathodal stimulation during the systolic interval
(S1-S2 protocol, S2 at 380 ms) with total current Iapp D 0:2160 mA and 10 ms duration, applied
to a slab with unequal anisotropy and membrane model with the addition of If ; Ie ; Ia . Same format
as in Fig. 9.5 (Reproduced with permission from [131])
9.1 Cardiac Excitation and Virtual Electrode Phenomena 267

b 50 30
A1

A
a 50
2
0 20

C1

mV

ms
A C2 C
2 0 2

−50 10
C A
1 1
−50

−100 0
0 10 20 30
−100 ms
−118.42 185.88 5.00

d D1
c
35 30
20
30
10
25
0.05 35.08 1.00
20

15 D2
30
10
20
5
10

0.05 35.08 1.00 0.05 30.72 1.00

Fig. 9.13 Systolic cathode break excitation. Cathodal stimulation during the systolic interval
(S1-S2 protocol, S2 at 380 ms) with total current Iapp D 0:2160 mA and 10 ms duration, applied
to a slab with unequal anisotropy and membrane model with the addition of If ; Ie ; Ia . Same format
as in Fig. 9.6 (Reproduced with permission from [131])

From the isochrone lines displayed in Fig. 9.13c, we clearly see that the excitation
sequence between 10 and 21 ms consists of two separated epicardial isochrones,
being the rims of two distinct wavefronts, which collide at 12 ms. As a major
transmural effect of the virtual electrode polarization, Fig. 9.13d shows that each
of the two propagating open wavefront surfaces has a rim partly coinciding with
a curve lying on the transmural boundary surrounding the virtual cathode volume.
The transmural wavefronts merge at about 25 ms in the midwall, then, after collision,
we have a unique wavefront propagating around the virtual cathode volume. Apart
from the virtual cathode volume, the activation patterns associated to the diastolic
anode make and to the systolic cathode break excitation mechanisms are similar,
especially after the merging of the separated wavefronts, as shown by comparing
the isochrones displayed in Figs. 9.13 and 9.6. Despite this qualitative similarity,
Fig. 9.13 shows that, for cathode break excitation, the first activated sites are not
268 9 Simulation Studies of Cardiac Bioelectrical Activity

located in the epicardial central sites in the virtual anodes, but at the two concave
portions of the boundary surrounding the virtual cathode volume. The trigger of
the systolic cathode break mechanism is due to the interplay between the outward
electrotonic current from the cathodal region and the membrane current dynamics at
the locations around the virtual anode. The wavefront propagation coincides with an
activation sequence moving around an obstacle (i.e. unexcitable volume) associated
with the depolarized region under the cathode.
Cathode break excitability mechanism is simulated also for a slab with T 2
anisotropy and exhibits a strongly similar excitation sequence (not shown). The
epicardial velocities of propagation along (across) fibers are 0:049 cm/ms (0:030)
for T 1 anisotropy and 0:040 (0:030) for T 2. These values are lower than the corre-
sponding diastolic velocities, because the tissue during the RRP is less excitable than
at rest. The degree of anisotropy influences the stimulus threshold of the cathode
break excitation, which is higher for the weaker T 2 anisotropy (0:1075 mA) than
for the stronger T 1 anisotropy (0:0578 mA). Finally, the cathode break mechanism
occurs, as expected, independently of the membrane funny current incorporation
(not shown).

9.1.4 Cathodal and Anodal Strength-Interval S-I Curves

In order to investigate the influence of the electroporation current Ie on the shape

of the S-I curves, we have computed the S-I curves with both strong and weak Ie
electroporation currents, associated to the electroporation characteristic voltages of
Vep D 126 mV and Vep D 258 mV, respectively. Figure 9.14 reports the cathodal and
anodal S-I curves for an orthotropic three-dimensional slabs using the augmented
LRd model with the strong and weak electroporation currents.
S1-S2 stimulation protocol and S-I curves. We first apply at a small volume
centered on the epicardial surface of the resting tissue an initial S1 cathodal
e
(iapp < 0) stimulus of 0:0324 mA amplitude and 1 ms duration super-threshold
initiating a propagating excitation wavefronts that sweeps the cardiac domain. At
a suitable time (coupling interval) during the relative refractory period (RRP) a
e e
subsequent premature S2 cathodal (iapp < 0) or anodal (iapp > 0) stimulus (with
duration 10 ms and amplitude specified case by case in the Results) is delivered at
the same subepicardial thin volume and we simulate the resulting action potential
propagation where the surrounding tissue has recovered its excitability properties.
We compute, for the augmented LRd model and the orthotropic anisotropic slab,
the cathodal and the anodal strength-interval (S-I) curve evaluating the minimal
amplitude S2 cathodal or anodal pulse able to elicit an excitation propagating
response thus, under which the tissue failed to respond. The values Istim of S2
e
stimulation amplitudes reported are given in mA, obtained multiplying iapp times
the volume of the stimulation site. We first determine the S-I curve for S1-S2
coupling interval increasing with 5 ms steps. Since propagating excitation can be
elicited before and after the end of an S2 pulse, as usual, we call make and break
9.1 Cardiac Excitation and Virtual Electrode Phenomena 269

Cathodal S−I curve, strong electroporation Anodal S−I curve, strong electroporation
0.5 0.5

0.4 0.4
stimulation threshold (mA)

stimulation threshold (mA)

0.3 0.3

0.2 0.2

0.1 0.1

0 0
180 190 200 210 220 230 240 250 180 190 200 210 220 230 240 250
S2 stimulation onset (ms) S2 stimulation onset (ms)

Cathodal S−I curve, weak electroporation Anodal S−I curve, weak electroporation
0.5 0.5

0.4 0.4
stimulation threshold (mA)

stimulation threshold (mA)

0.3 0.3

0.2 0.2

0.1 0.1

0 0
180 190 200 210 220 230 240 250 180 190 200 210 220 230 240 250
S2 stimulation onset (ms) S2 stimulation onset (ms)

Fig. 9.14 Cathodal (left) and anodal (right) strength-interval curves for strong (top) and weak
(bottom) electroporation. Bold points and empty circles denote make and break stimulation
responses, respectively (Reproduced with permission from [132])

stimulation responses the two type of responses. When computing the anodal S-I
curves described above, we refine the step from 5 to 1.25 ms around the break/make
transition in order to better resolve the S-I curve in this interval.
S-I curves. Both cathodal S-I curves display a monotonically decreasing behav-
ior in the portions corresponding to the cathode break stimulation response and the
subsequent phase corresponding to the cathode make stimulation response, with an
almost constant threshold stimulus strength. The use of strong electroporation has
a weak effect on the make portion of the S-I curve, while on the break portions
it induces a slight decrease of the threshold stimulus strength with a shift of the
effective refractory period which is increased from 180 to about 190 ms.
The anodal S-I curves displayed in Fig. 9.14 exhibit the following main features:
a dip close to the effective refractory period, a subsequent plateau or dome phase
corresponding to the anode break stimulation response and a sudden decrease close
to the end of the relative refractory period, separating the anode break and make
mechanisms. By relative refractory period, we mean the time interval, between the
effective refractory period and the resting phase, where another action potential can
270 9 Simulation Studies of Cardiac Bioelectrical Activity

be elicited only by a sufficiently strong stimulus. The comparison of the strong

and weak electroporation S-I curves shows no significant threshold variation in the
make portion of the S-I curve, while the presence of weak electroporation increases
the threshold values in the break phase, with a more sudden and large jump in the
transition from break to make stimulation responses. Weak electroporation also
produces a slight increase of the relative refractory period, (from 20 to 27 ms). The
break portion of the anodal S-I curves exhibits a typical dip followed by a dome
and a subsequent sudden fall sharing the same features of the S-I curves observed
experimentally in [146, 324, 480] and computed in [255, 256, 428]; we refer to [428]
for an explanation of these features.
In the cathodal S-I curve, the transition from break to make stimulation responses
occurs during the coupling interval between 210 and 215 ms. In the anodal S-I curve,
the same transition occurs between 215 and 220 ms with weak electroporation and
between 210 and 212.5 ms with strong electroporation. We remark that any pulse
amplitude above threshold yields the same type of response (make or break) as
the threshold response, except during a short interval around the transition between
break and make stimulation responses.

9.2 Anisotropic Propagation of Excitation and Recovery

Fronts

In this section, we present some simulations of excitation and repolarization

sequences on myocardial slabs of different sizes, showing how the distribution of
the action potential durations (APD) is influenced by both the anisotropic electrical
conduction and the fiber rotation. This influence occurs in spite of the homogeneous
intrinsic properties of the cell membrane. The APD dispersion patterns are closely
correlated to the anisotropic curvature of the excitation wavefront. More detailed
results can be found in [123].
Parameter calibration. In the following, we shall assume that both the intra-
cellular coupling and the extracellular matrix are uniform, i.e. the intra- and
i;e
extra-cellular conductivity coefficients along and across fibers l;t;nu are independent
of position. The conductivity coefficients given in Table 9.2 have been calibrated so
that the associated propagation velocities (l ; t ;  ) of ideal plane wavefronts can be
conservatively estimated by (60, 25, 10) cm s1 , respectively. These estimates have
been obtained by simulating the activation sequence elicited by a vertex stimulation
of a 2D rectangular sheet of dimensions 3  3 cm2 with fibers parallel to a side.
The excitation wave front propagated along fiber with a velocity of 58.7 cm s1 ,
while the velocity across fiber was 24.6 cm s1 if the conductivity coefficient was
ti;e or 9.3 cm s1 if the conductivity coefficient was i;e . Due to the convexity of
the fronts, i.e with a negative curvature sign, the wave front velocities were lower
than those of plane fronts propagating parallel or perpendicular to fiber direction,
as expected. In the following, we will consider 3D parallelepipedal slabs and we
9.2 Anisotropic Propagation of Excitation and Recovery Fronts 271

Table 9.2 Parameters calibration for numerical tests

 D 103 cm1 ; Cm D 1 F=cm2
le D 2  103 S cm1 ; li D 3  103 S cm1
te D 1:3514  103 S cm1 ; ti D 3:1525  104 S cm1
e D te =1 ; ni D ti =2
1 D 2 D 1 axisymmetric case, 1 D 2; 2 D 10 orthotropic case

will identify the upper and lower face planes with the epicardial and endocardial
surfaces, respectively. The minimal velocity  D 10 cm s1 corresponds to a
propagation across fibers on intramural planes parallel to the epicardial surface,
while the velocity t D 25 cm s1 corresponds to a propagation across fibers in
directions perpendicular to the epicardial plane. The initial conditions are .u0i ; u0e / D
.84; 0/ mV, so that v0 D 84 mV and we apply a large stimulus of 200 A for
1 ms on a small volume (3 or 5 mesh points in each direction). The slow inward
current Isi in the LR1 model is scaled by a factor 2/3. This yields an action potential
duration (APD) of about 265 ms.
The fibers rotate intramurally linearly with depth for a total amount of 90ı , i.e.
al .x/ D uex cos ˛.r/ C uey sin ˛.r/; ˛.r/ D .1  2r/=4; r 2 Œ0; 1. The fiber
direction on the epicardium is 45ı with respect to the horizontal side of the panels
displayed in Figs. 9.15–9.17, while on the endocardium is 45ı . In the orthotropic
anisotropy case, we choose a D uex sin ˛.r/  uey cos ˛.r/ and at D uez . In order
to strengthen the role played by the rotational anisotropy, the counter-clockwise
(CCW) intramural fiber rotation amounts to 90ı on a slab thickness of 0.5 cm,
with direction varying linearly from epicardium to endocardium. The stimulus was
applied at an appropriate epicardial vertex or at the center of the slab. In all cases,
the finite element mesh size is h D 0:1 mm.

9.2.1 Excitation and Repolarization Sequences

In order to study the influence of fiber rotational anisotropy on the activation and
repolarization sequences and on the APD distribution, we simulate the electrical
activity in a slab of cardiac tissue, from the onset of excitation to the end of
repolarization. We consider the following test problems:
• Bidomain-LR1 model on a 2  2  0:5 cm3 slab, vertex stimulation (orthotropic
(left Panel) and axisymmetric (right Panel) in Fig. 9.15 top);
• Monodomain-LR1 model on a 220:5 cm3 slab, vertex stimulation (orthotropic
(left Panel) and axisymmetric (right Panel) in Fig. 9.15 bottom);
• Monodomain-LR1 model on a 4  4  0:5 cm3 slab (central stimulation,
axisymmetric in Fig. 9.16, orthotropic in Fig. 9.17).
272 9 Simulation Studies of Cardiac Bioelectrical Activity

ACTI REPO APD ACTI REPO APD

80 340 270 60 320
60 268 270
320
EPI

EPI
40
40 266 300
300 265
264 20
20
280 262 280 260
0.09 80.44 2.50 271.52 341.02 2.50 260.59 271.79 0.30 0.10 72.15 2.50 274.74 331.63 2.50 259.49 274.65 0.30

268 268

266 266
MID

MID
264 264

262 262
9.73 81.49 2.50 279.15 343.15 2.50 261.05 269.41 0.30 10.91 64.27 2.50 280.09 327.25 2.50 260.98 269.17 0.30

268
266
ENDO

ENDO
266
264
264
262
262
20.63 86.94 2.50 285.84 347.83 2.50 260.89 268.87 0.30 21.41 64.62 2.50 285.32 324.81 2.50 260.18 267.38 0.30

270
268 270
266
265
264
262
260

0.09 86.94 2.50 271.52 347.83 2.50 260.59 271.79 0.30 0.10 72.15 2.50 274.74 331.63 2.50 259.49 274.68 0.30

ACTI REPO APD

ACTI REPO APD
275
80 270 60 320
340 270
268
EPI
60 40
EPI

320 266
40 300 265
300 264 20
20 262
280 280 260
0.12 87.99 2.50 271.82 348.36 2.50 260.37 271.97 0.30 0.12 76.07 2.50 275.28 335.49 2.50 259.43 275.17 0.30

268
268
266
MID

266
MID

264
264
262
262
9.89 90.52 2.50 279.27 354.16 2.50 261.15 269.38 0.30 11.15 68.33 2.50 280.50 331.06 2.50 260.83 269.35 0.30

268 266
ENDO

266
ENDO

264
264
262
262
21.67 68.03 2.50 285.77 328.35 2.50 260.32 267.56 0.30
20.81 99.38 2.50 286.06 359.62 2.50 260.25 268.79 0.30
275
270
270
268
266 265
264
262 260

0.12 76.07 2.50 275.28 335.49 2.50 259.43 275.19 0.30

0.12 99.38 2.50 271.82 359.62 2.50 260.25 271.97 0.30

Fig. 9.15 Orthotropic (left Panels) and axisymmetric (right Panels) Bidomain Models (top row)
and Monodomain Models (bottom row) on a slab 2  2  0:5 cm3 with vertex stimulation and
homogeneous cellular membrane properties. Patterns of level lines of the depolarization time (first
column ACTI), repolarization time (second column REPO) and action potential duration (third
column APD) on epi, midwall and endocardial planes and at the bottom 3D view. Reported below
each panel are the maximum, minimum and step in ms of the displayed map (Reproduced with
permission from [123])

9.2.1.1 Bidomain Model: Orthotropic Versus Axisymmetric Anisotropy

In order to contain the computational cost required by the simulation of an entire QT

interval (excitation plus recovery) using the Bidomain model, we consider a slab of
dimensions 2  2  0:5 cm3 .
Previous experimental studies (see e.g. Burgess et al. [67], Osaka et al. [364],
Gotoh et al. [208], Taccardi et al. [512]) indicate that the main effects of anisotropy
on the APD occur in regions where the wave fronts move across fibers. To best
illustrate cross-fiber propagation, we oriented the epicardial fibers along a diagonal
and we stimulated a vertex situated at one end of the other diagonal, so that the
elicited wave fronts move mainly across fiber on the epicardium (see Fig. 9.15 top).
9.2 Anisotropic Propagation of Excitation and Recovery Fronts 273

ACTI REPO APD

275
60 320
270
EPI
40
300 265
20
280 260
0.14 76.43 2.50 276.20 335.81 2.50 259.38 276.06 0.30

268
266
MID

264
262
11.70 68.66 2.50 280.69 331.30 2.50 260.78 268.99 0.30

266
ENDO

264
262
260
22.03 64.13 2.50 285.63 326.92 2.50 259.50 267.42 0.30

275

270

265

260
0.14 76.43 2.50 276.20 335.81 2.50 259.38 276.06 0.30

Fig. 9.16 Axisymmetric monodomain model with intramural fiber rotation on a slab 4  4 
0:5 cm3 with central epicardial stimulation. Same format as in Fig. 9.15 (Reproduced with
permission from [123])

The macroscopic sequences of excitation and recovery and the spatial distribu-
tion of the APDs are illustrated in Fig. 9.15 top, that relates to a Bidomain model
implemented in a slab with orthotropic and axisymmetric anisotropy, respectively.
Both anisotropic slabs exhibit similar qualitative features of the patterns of excita-
tion, recovery and APD distribution.
Activation sequence. It is well known that in the presence of rotational
anisotropy, intramural excitation, starting from an epicardial stimulation site, first
proceeds toward the endocardium but subsequently, due to fiber rotation, comes
back pointing toward the epicardial plane (see e.g. [67,102,116,184,506,508,509]).
Due to these intramural return pathways, propagation undergoes an acceleration,
in particular in epicardial areas where the excitation moves mainly across fibers.
In the orthotropic slab, the lower epicardial velocity across fiber (as compared
to axisymmetric model) causes the appearance of more pronounced dimple-like
inflections.
Repolarization sequence. Repolarization wave fronts exhibit a somewhat
smoother shape and faster propagation compared with the excitation sequence, as
shown by the isochronal lines on the epi, midwall and endocardial planes. Indeed,
the epicardial plane is fully activated in about 80.35 ms and repolarizes in 69.50 ms,
274 9 Simulation Studies of Cardiac Bioelectrical Activity

ACTI REPO APD

80 340 274
272
60 320 270
EPI
268
40 266
300
20 264
262
280
0.14 85.67 2.50 274.81 345.64 2.50 259.97 274.69 0.30

268
266
MID

264
262
11.02 75.68 2.50 280.00 339.82 2.50 261.04 268.98 0.30
268
266
ENDO

264
262

21.48 71.91 2.50 285.60 333.66 2.50 260.11 268.15 0.30

274
272
270
268
266
264
262

0.14 85.67 2.50 274.81 345.64 2.50 259.92 274.69 0.30

Fig. 9.17 Orthotropic monodomain model with intramural fiber rotation on a slab 4  4  0:5 cm3
with central epicardial stimulation. Same format as in Fig. 9.15 (Reproduced with permission from
[123])

giving rise to a total amount of APD dispersion of 11.2 ms in the orthotropic case. In
the axisymmetric case, the same section is fully activated and repolarized in about
72 and 57 ms, respectively, with an APD dispersion of 15 ms.
APD distribution. The main common features, displayed by the APD on the
epicardial, intramural and endocardial planes, in both the anisotropic slabs are: a
maximum located at the epicardial stimulation site, higher APD values in regions
where propagation is mainly along fibers, an intramural valley configuration of APD
reduction, a relative minimum in areas of front-boundary collision, and a maximum
located on the endocardial plane. A detailed comparison between orthotropic and
axisymmetric cases shows that the former emphasizes some anisotropic features
displayed by the APD distribution pattern. In particular, the presence of a dimple-
like inflection exhibited by the excitation isochronal lines on intramural horizontal
planes ranging from the epicardium to the midwall, and of a corresponding finger-
shaped valley of low APD are more pronounced in the orthotropic slab, as shown
e.g. by comparing the midwall panels of Fig. 9.15 top.
We remark that the axisymmetric and the orthotropic slabs are fully excited
in about 72 and 87 ms, respectively. The extinction area of excitation is located
at a vertex on the endocardium in the orthotropic slab, whereas it appears at the
9.2 Anisotropic Propagation of Excitation and Recovery Fronts 275

epicardium in the axisymmetric slab. This is the result of the reduced velocity
t across fibers on planes parallel to epicardium being lower than the transmural
velocity  across the slab thickness, combined with the effects of the intramural
return pathways of the excitation toward the epicardium. The main quantitative
discrepancy between the two slabs lies at the epicardium, where different activation
and repolarization isochrone shapes in the front-boundary collisions are present;
they are associated with the appearance, in the APD pattern of the orthotropic
slab, of an epicardial relative maximum located at a side of the finger-shaped
valley of APD reduction which started from a relative minimum located at the
boundary.
The APD distributions show several common features in our different simu-
lations, namely: APDs are longest near the pacing site and tend to be shorter
near the front-boundary collisions. Moreover, the spatial distribution of APDs
are influenced by the anisotropic motion of the wave front. Thus, in a slab with
homogeneous cellular membrane properties, i.e. where all individual cells have the
same intrinsic transmembrane action potential, the rotational anisotropy produces
an APD variability displaying an anisotropic pattern. The orthotropic anisotropy
further increases the distortion of the propagating wavefront, emphasizing the
corresponding anisotropic features of the APD dispersion of the axisymmetric case.

9.2.1.2 Vertex Stimulation of Anisotropic Models: Bidomain Versus

Monodomain Models

Due to the high computational cost of the Bidomain model on a slab with
dimensions 2  2  0:5 cm3 , we performed simulations including both excitation
and recovery using the Monodomain model on the same slab with orthotropic and
axisymmetric anisotropy. We focus the comparison on the shape and propagation
of the excitation and recovery isochrones, as well as the spatial distribution of the
APD.
Activation and repolarization sequences. The comparison of Fig. 9.15 top and
bottom for both orthotropic and axisymmetric slabs evidences a strong qualitative
similarity between the simulated isochrones of the Bidomain and Monodomain
models. This fairly good P match is confirmed by the values of the P correlation
coefficient CC.x; y/ D n1 . .xi  x/.yi  y//=.x y /; where x D n1 xi is the
q P
arithmetic mean of x, x D 1
n
.xi  x/ is the standard deviation of x and
2

analogously for y and y . The correlation coefficient CC for the activation times
of the two models in the entire slab amounts to 0.9971 in the orthotropic case
and to 0.9997 in the axisymmetric case. Similarly, the CC for the repolarization
times on the entire slab amounts to 0.99976 and to 0.9998 in the orthotropic
and axisymmetric cases, respectively. The wave front velocity depends on the
propagating direction and the Bidomain and Monodomain models exhibit a different
anisotropic dependence. In spite of this difference, we found not only a strong
276 9 Simulation Studies of Cardiac Bioelectrical Activity

qualitative agreement but also a small quantitative difference in the activation and
repolarization sequences between the Monodomain and the Bidomain model. The
relative error RE D kTB  TM k2 =kTB k2 (with k  k2 the Euclidean norm) between
the activation times TB ; TM of the Bidomain and Monodomain models is 0.115 in
the orthotropic case and 0.063 in the axisymmetric case. The relative error between
the repolarization times of the two models is 0.018 in the orthotropic case and 0.009
in the axisymmetric case. We remark that the repolarization isochrones simulated
by both models, especially on the epicardial plane, exhibit a smoother dimple-like
inflection than that appearing in the excitation sequence.
APD distribution. Both orthotropic model simulations exhibit a common
narrow valley of low APD values (see Fig. 9.15 top and bottom), corresponding
to the pathway joining the dimple-like inflections appearing in the epicardial and
intramural excitation isochrones. These dimple-like inflections are a reflection of
the intramural pathways of excitation that, starting from the stimulation site, first
proceed toward the endocardium but subsequently, due to fiber rotation, return
toward the epicardial plane. Therefore, the narrow valley configuration in the APD
distribution is an effect of the rotational anisotropy. We found a good qualitative and
quantitative agreement between the APD related to the Bidomain and Monodomain
models, since CC and RE on the entire slab are 0.9534 and 1.882 103 in the
orthotropic case, while they are 0.987 and 8.9 104 in the axisymmetric case.
The previous comparison validates the use of the Monodomain model as a
tool for investigating the qualitative macroscopic features of the activation and
repolarization sequences, as well as the APD distribution. Therefore, the high costs
required, at present, by the Bidomain simulations can be avoided by implementing
the Monodomain model, which qualitatively preserves the anisotropic features of
the APD distribution.

9.2.1.3 Central Stimulation in Anisotropic Monodomain Models

To further analyze the factors that modulate the APD in a model with homogeneous
membrane properties, we simulate the excitation and recovery sequences elicited
by a central epicardial stimulation, in case of axisymmetric and orthotropic Mon-
odomain slabs with intramural fiber rotation (Figs. 9.16 and 9.17).
The comparison between the APD dispersion patterns should allow us to detect the
role played by:
• The different phases of the wave front sequence of excitation and recovery
(i.e. the initial and extinction wave fronts, the occurrence of front-boundary
collisions) and the wave front shape;
• The anisotropic electrical conductivity (i.e. the different electrotonic currents
flowing along and across the fiber direction);
• The intramural fiber rotation, which causes the occurrence of return pathways of
excitation and recovery toward the stimulated plane.
9.2 Anisotropic Propagation of Excitation and Recovery Fronts 277

Since similar observations apply to both cases, in the results description we refer,
preferentially, to the orthotropic case in which some features are more pronounced.
Activation and recovery sequences. Stimulation at the center of the epicardial
face produces approximately elliptical excitation and repolarization isochrone lines,
a clear sign of their anisotropic propagation. Actually, activation and repolarization
spread symmetrically with respect to the center of the epicardial face. The major
axes of the oblong excitation isochrones are nearly parallel to the epicardial fiber
direction (45ı with respect to the horizontal sides of the panels in Figs. 9.16
and 9.17) while those associated to repolarization isochrones are rotated CCW with
respect to epicardial fiber direction. The oblong epicardial isochrones are slightly
rotated and bulging CCW with respect to the fiber direction, see Figs. 9.16 and 9.17
Panels ACTI, REPO. The spread of excitation and the sequence of recovery show
an acceleration in the cross-fiber direction. Also, the excitation isochrones show
an inflection corresponding to a dimple-like inflection of the wave front. These
findings, i.e. the accelerating propagation across fibers, the bulging and the dimple-
like inflection of the isochrones, are attributed to the influence on the epicardium
of the activation and recovery processes through the deeper layers, where the fiber
direction rotates CCW relative to epicardial fibers.
Proceeding from epi- to endo-cardium, on the intramural planes parallel to the
epicardium the spacing between excitation (recovery) isochrones increases, the
wave front shapes become rounder and we observe a transmural twisting of the
isochrones, i.e. the major axis of the oblong isochrones progressively rotates CCW
with increasing depth. However, their rotation lags behind the rotation of the fiber
direction at corresponding depths (see e.g. the Panels MID of Fig. 9.17, where
the fiber direction is horizontal). It is worth mentioning that the total transmural
rotation of the recovery isochrones is slightly less than the rotation of the excitation
isochrones. The spacing between the recovery isochrones, in the sections parallel to
the epicardial face, is greater than the corresponding excitation spacing, denoting a
faster motion of the recovery fronts.
On the endocardial plane, both the excitation and recovery front-boundary collision
first occur at the center of the face. This happens because our current model does not
incorporate the epi- endocardial obliqueness of the fibers (Streeter [498], Taccardi
et al. [507]), that was included in some of our previous simulations (Colli Franzone
et al. [116, 116]). Subsequent excitation and recovery isochrones have a well
rounded, elliptical shape centered at the point of endocardial BKT of the excitation
and repolarization wave front with the endocardial plane. On the endocardium,
the large spacing between successive isochrones indicates a fast excitation and
repolarization progression with a maximum apparent speed at the BKT point where
a sudden change of the wave front curvature occurs and the high curvature values
bring about high speeds of propagation around the BKT site, see e.g. Colli Franzone
et al. [102], Keener [268], Fast et al. [169]. The results for the excitation sequence
confirm earlier findings of Colli Franzone et al. [102, 116] obtained with an eikonal
approach, and are in agreement with previous experimental data (Taccardi et al.
278 9 Simulation Studies of Cardiac Bioelectrical Activity

[507]). The results relating the repolarization sequence are the main novelty of this
work.
APD distribution. Adding the intramural fiber rotation to the axisymmetric or
orthotropic anisotropic tissue, the epicardial and intramural APD distribution exhibit
a pattern displaying elongated level lines, (see Figs. 9.16 and 9.17), surrounding the
central point of the epicardial or intramural plane, corresponding to the stimulation
site and to first point reached by the excitation wave front. At first, the pattern
exhibits a quasi-elliptical shape with the major axis correlated with that of the
excitation isochrones. This indicates that APD decreases more rapidly when moving
away from the center of the face in the cross-fiber direction than along fibers.
Subsequently, the isochrones lose the quasi-elliptical shape but are still stretched
along a preferential direction, which does not coincide with the local fiber direction,
except at the epicardial level, see Panel MID in Figs. 9.16 and 9.17. The APD
decrease shows that the velocity of the repolarization front exceeds the one of the
activation front, both along and across the preferential directions, and moreover the
difference between the repolarization and the activation velocities is greater across
this direction, as confirmed by the more rapid APD decrease.
At some distance from the center of the face, the stretched portion of the level
lines of the APD, due to boundary effects, becomes straight while the presence of
valleys of APD reduction induces hollow, dog-bone profiles, more pronounced in
the orthotropic slab (see Fig. 9.17). The total dispersion in the axisymmetric and in
the orthotropic slabs amount to 16.7 and 14.8 ms, respectively.
On the intramural sections (from subepicardial (not shown) to midwall ones), two
valleys of decreasing APD values occur. Each valley exhibits a minimum located
at the boundary of the slab. These narrow valleys of low APD values are located
in the regions where excitation isochrones exhibit a dimple-like inflections. These
depression valleys are not related to boundary collisions but are strictly correlated
with the anisotropic influence of the intramural fiber rotation.
In the endocardial plane, the APD distribution displays a saddle point at the
endocardial BKT; the APD increases reaching a maximum when moving away from
the BKT point in a direction parallel to the endocardial fibers of 45ı CW, while
in the cross-fiber direction the APD decreases toward two minima located at the
boundary and corresponding to front-boundary collision points; the axis joining the
two maxima is roughly parallel to the endocardial fiber direction of 45ı CW and the
two maxima are separated by a valley of APD reduction.
We remark that the anisotropic features appearing in the APD pattern are more
pronounced in the orthotropic case; because these emphasized effects are likely
to help in the investigation of their origin and due to the recent interest about the
influence of orthotropic myocardial structure on the cardiac electric activity (see
LeGrice et al. [292,293], Costa et al. [137] for anatomical studies and Colli Franzone
et al. [121] for simulation studies), in the following the simulations have been
performed only for an orthotropic Monodomain model with orthotropic anisotropy.
9.2 Anisotropic Propagation of Excitation and Recovery Fronts 279

Intramural excitation and repolarization sequences in the wall thickness.

Excitation and recovery isochronal profiles on the epicardial, subepicardial (not
shown) and midwall intramural planes exhibit a dimple-like inflection, which is
smoothed in the repolarization sequence, in a region where excitation wave fronts
propagate mainly across fibers, see Fig. 9.17, Panels EPI, MID. These dimple-like
inflections are an effect of the presence of excitation and recovery pathways that start
from the stimulation site, proceed first toward the endocardium and subsequently,
due to fiber rotation, return toward the epicardium; this results in accelerated
propagation in some portions of the epicardial fronts that move mainly across fibers.

9.2.2 Discussion on APD Distribution and Dispersion

Previous simulation studies considered simulations of the entire excitation and

repolarization sequences on 1D cables (Joyner [264], Steinhaus [497], Shaw and
Rudy [474], Wang and Rudy [548], Viswanathan et al. [546]), 2D sheets (Burgess
et al. [67], Cates and Pollard [80]) and 3D slabs (Efimov et al.[155], Pollard et al.
[398], Garfinkel et al. [191]). A limited number of simulations studies used the
full Bidomain Model in three dimensions. Some histological findings (see LeGrice
[292, 293]) have supported the idea that the cardiac tissue anisotropy could be
orthotropic, but most simulations have been performed for axisymmetric anisotropy,
with the exception of Colli et al. [121], Hooks et al [234] for the excitation phase
and Colli Franzone et al. [120] for both the excitation and recovery phases.
The results presented in the previous Section describe the features of the
excitation and recovery sequence, using both the Bidomain and the Monodomain
models, and show that APD patterns present a definite spatial dispersion in spite
of the assigned homogeneity of the individual cellular membrane properties. These
patterns, are qualitatively similar in the Bidomain and the Monodomain models.
The anisotropic features of the excitation and recovery times, briefly mentioned
in describing the vertex simulations, are easier to detect and to interpret when we
inspect the patterns elicited by central epicardial stimulations (Figs. 9.16 and 9.17).
We considered both axisymmetric and orthotropic anisotropy and our simulations
show that several features are more pronounced in the orthotropic case.
Now we briefly review these main anisotropic features appearing in both the
excitation and repolarization processes, elicited by an epicardial central stimulation
in an orthotropic slab. A main anisotropic feature is the presence of return excitation
pathways, i.e. pathways that, starting from the epicardial stimulation site, proceed
toward the endocardium but about midway of the wall thickness return pointing
toward the epicardial side. Signs of these returning pathways are the dimple-
like inflections appearing in the excitation wave front profiles in the epi and
midwall planes parallel to the epicardial face (see Fig. 9.17), and, in planar sections
280 9 Simulation Studies of Cardiac Bioelectrical Activity

perpendicular to the epicardium, V-shaped intramural profiles with the epi- and
endo-cardial foot of the profiles overcoming the peak of the Vs. The simulated
anisotropic behavior of the excitation sequence is in agreement with previous
experimental and simulated observations, e.g. Burgess et al. [67], Taccardi et al.
[506–508], Colli Franzone et al. [102, 114, 116, 121], Keener [268] and Henriquez
et al. [226, 344].
With regard to repolarization, the recovery isochrones on the epi, midwall and
endocardial planes elicited by an epicardial pacing exhibit a somewhat smoother
shape and faster propagation compared with the excitation sequence. In particular,
the epicardial repolarization isochrones propagate across fibers faster than the
excitation wave, yielding a progressive APD shortening in the cross-fiber direction
of propagation as shown by Figs. 9.16 and 9.17. This prediction is in agreement with
the data of Gotoh et al. [208] related to cardiac laminae. The recovery isochrones
through the thickness of the wall show the presence of repolarization return
pathways as in the excitation sequence. These pathways accelerate the repolarization
process in epicardial areas where the recovery proceeds mainly across fibers. Unlike
the excitation propagation, experimental studies of intramural repolarization is less
complete and their interpretation is still the subject of intense research.
The finger-shape valleys of APD reduction associated to the dimple-like inflec-
tions appearing in the excitation and recovery fronts are the result of the interaction
between excitation and repolarization return pathways. Since these valleys are not
present in APD distribution in a slab with parallel fibers, they can be attributed to
the rotational anisotropic structure of the ventricular wall. Moreover, these valley are
narrower in the orthotropic case. These results relating the repolarization sequence
and APD dispersion are the main novelty of this work.
Anisotropic spatial variations of the APD along and across fibers were observed
experimentally in 2D cardiac laminae by Osaka et al. [364] and Gotoh et al. [208]
and on the epicardium of dog hearts by Burgess et al. [67] and Taccardi et al. [512].
More precisely, Taccardi et al. [512] observed a typical tripolar epicardial APD
pattern, with one central maximum and two minima along a cross-fiber direction,
on the left ventricle of exposed dog hearts during left ventricular pacing. Results
were recently confirmed by optical mapping. On the other hand, Bertran et al. [48],
working with pig hearts, did not observe anisotropic APD differences only in 2D
laminae.
We remark that simulation studies and experimental data have shown that
excitation return pathways, proceedings toward the pacing level, have been observed
for pacing sites located at any intramural level, from epi- to endo-cardium. On this
ground, we expect that repolarization return pathways, here reproduced only for
epicardial pacing, will be present also for intramural pacing.
Rules for APD modulation. Even if the plane sections reported cannot com-
pletely capture the complex evolution of the excitation and repolarization surfaces
in a three-dimensional domain, the results suggest the following rules for APD
modulation on plane sections:
9.2 Anisotropic Propagation of Excitation and Recovery Fronts 281

(i) Relatively long APDs occur at points where the excitation and recovery fronts
diverge, with maxima attained at the stimulation site and at each breakthrough
point on intramural planes parallel to the epicardium with the exception of the
endocardial plane;
(ii) A relative APD shortening occurs at points where the excitation and recovery
fronts converge, with minima attained at points of front-boundary collision and
front extinction;
(iii) Moving away from the stimulation site, the APD shortening is greater in the
directions across fibers than along fibers;
(iv) Valleys of relative APD shortening are associated with areas containing
dimple-like inflection in the excitation isochrones and the level lines of the
APD display dog-bone shaped profiles.
Anisotropic curvature as a parameter affecting the APD modulation. The
motion of the excitation wave front first depends on the characteristics of the
anisotropic media, on the other hand it is well known that the shape of the wave
front is a factor which contributes to the wave propagation, see e.g. [102, 114, 116,
169, 268, 273].
In the following, we consider an excitation (or repolarization) wave front locally
convex, relative to the propagating direction, according to whether the front bulges
into the resting (excited) volume; concave when it bulges into the excited (resting)
volumes from which it is coming.
With regard to the electrotonic load, we observe that currents flowing from the
excitation layers of a locally concave front are focusing toward points ahead of
the front, whereas for locally convex front they are scattering; the former case,
producing a more rapid membrane depolarization, yields a speed up of the excitation
propagation, compared to the velocity of a local plane front, whereas a slower
spreading results in the latter case. The opposite electrotonic process takes place
when considering points ahead of repolarization fronts i.e. for locally concave
(convex) fronts we have a depletion (convergence) of repolarizing currents. In spite
of this opposite behavior, we have similar effects on the recovery propagation, i.e.
in the concave case a faster recovery while in the convex case a slowing down
repolarization.
The dependence of the conduction velocity on the wave front shape is quanti-
tatively described by means of the curvature of the wave front. It is well known
that in isotropic homogeneous media with conductivity , the velocity, along the
Euclidean normal n of the front at a point x, oriented in the
pdirection of motion, can
be approximated by .x; n/ D p .1 C K=cm /, with K D  k, where k D  div n
is twice the mean curvature of the front, see e.g [268, 273]. The velocity of a local
plane front p p also depends on the conductivity of the excitable medium, i.e. we
have p D  , with the constant  only dependent on the membrane model. Then
a locally convex front, such as an expanding circle, has negative curvature, while a
locally concave front, such as a contracting circle, has positive curvature. Hence, a
locally convex (concave) front propagates with velocity smaller (greater) than that
of a plane wave.
282 9 Simulation Studies of Cardiac Bioelectrical Activity

For an anisotropic Monodomain slab the velocity of the excitation wave front is
described by a similar law of motion .x; n/ D p .x; n/.1 C K=cm /. The principal
term p .x; n/ predicts the velocity of an idealplocally plane wave propagating along
the normal direction, given by p .x; n/ D  .x; n/, with .x; n/ D nT Dm .x/n
and Dm .x/ D Di .Di C De /1 De . The additional term is related to the anisotropic
curvature (see Bellettini et al. [37]), whose expression for the Monodomain model
by K D  div , where differently from the usual Euclidean normal, .x/ D
is given p
D.x/n= nT Dn; see Bellettini et al. [39, 102, 268].
Burgess et al. [67] observed in experiments on pulmonary cone preparations that
the repolarization pattern can be affected by the direction of the excitation wave
front and that anisotropic differences in conduction velocity induce electrotonic
effects on the cell repolarization. Their findings show that the APD is relatively long
at sites where the activation front decelerates and is relatively short at sites where
the activation front accelerates, suggesting that acceleration and deceleration of the
excitation wave front are associated with a relative APD reduction and prolongation,
respectively. They established a significant correlation between the APD distribution

.
390.0 - 395.0

a b d 385.0 - 390.0
380.0 - 385.0
375.0 - 380.0
370.0 - 375.0
365.0 - 370.0
360.0 - 365.0
355.0 - 360.0
350.0 - 355.0
345.0 - 350.0
340.0 - 345.0
335.0 - 340.0
330.0 - 335.0
325.0 - 330.0
320.0 - 325.0
315.0 - 320.0
310.0 - 315.0
305.0 - 310.0
300.0 - 305.0
295.0 - 300.0
290.0 - 295.0
285.0 - 290.0
280.0 - 285.0
275.0 - 280.0
270.0 - 275.0
265.0 - 270.0
260.0 - 265.0
255.0 - 260.0
250.0 - 255.0
245.0 - 250.0
240.0 - 245.0
235.0 - 240.0
230.0 - 235.0
225.0 - 230.0
220.0 - 225.0
215.0 - 220.0
210.0 - 215.0
205.0 - 210.0
200.0 - 205.0
195.0 - 200.0

42.4 ( 5.0 ) 119.9 0.5 ( 5.0 ) 88.0 21.8 ( 5.0 ) 129.7 190.0 - 195.0
185.0 - 190.0
180.0 - 185.0
175.0 - 180.0

f e 170.0 - 175.0
165.0 - 170.0
160.0 - 165.0
155.0 - 160.0
150.0 - 155.0
145.0 - 150.0
140.0 - 145.0
135.0 - 140.0

c
130.0 - 135.0
125.0 - 130.0
120.0 - 125.0
115.0 - 120.0
110.0 - 115.0
105.0 - 110.0
100.0 - 105.0
95.0 - 100.0
90.0 - 95.0
85.0 - 90.0
80.0 - 85.0
75.0 - 80.0
70.0 - 75.0
65.0 - 70.0
60.0 - 65.0
55.0 - 60.0
50.0 - 55.0
45.0 - 50.0
40.0 - 45.0
35.0 - 40.0
30.0 - 35.0
25.0 - 30.0
20.0 - 25.0
15.0 - 20.0
10.0 - 15.0
5.0 - 10.0

1.1 (5.0) 93.3 15.2 ( 5.0 ) 93.0 0.2 ( 5.0 ) 145.7 0.0 - 5.0
below 0.0

Fig. 9.18 Axisymmetric monodomain model with intramural fiber rotation on a truncated ellip-
soidal domain with central endocardial stimulation. Activation time isochrones on endocardial
(e), mid-myocardial (d), epicardial (a) and different transmural sections (b-c-f). All values are
expressed in ms
9.2 Anisotropic Propagation of Excitation and Recovery Fronts 283

.
390.0 - 395.0
385.0 - 390.0

a b d
380.0 - 385.0
375.0 - 380.0
370.0 - 375.0
365.0 - 370.0
360.0 - 365.0
355.0 - 360.0
350.0 - 355.0
345.0 - 350.0
340.0 - 345.0
335.0 - 340.0
330.0 - 335.0
325.0 - 330.0
320.0 - 325.0
315.0 - 320.0
310.0 - 315.0
305.0 - 310.0
300.0 - 305.0
295.0 - 300.0
290.0 - 295.0
285.0 - 290.0
280.0 - 285.0
275.0 - 280.0
270.0 - 275.0
265.0 - 270.0
260.0 - 265.0
255.0 - 260.0
250.0 - 255.0
245.0 - 250.0
240.0 - 245.0
235.0 - 240.0
230.0 - 235.0
225.0 - 230.0
220.0 - 225.0
215.0 - 220.0
210.0 - 215.0
205.0 - 210.0
200.0 - 205.0
195.0 - 200.0

289.1 ( 5.0 ) 366.2 258.6 ( 5.0 ) 333.6 273.5 ( 5.0 ) 377.2 190.0 - 195.0
185.0 - 190.0
180.0 - 185.0

f
175.0 - 180.0

e
170.0 - 175.0
165.0 - 170.0
160.0 - 165.0
155.0 - 160.0
150.0 - 155.0
145.0 - 150.0
140.0 - 145.0
135.0 - 140.0

c
130.0 - 135.0
125.0 - 130.0
120.0 - 125.0
115.0 - 120.0
110.0 - 115.0
105.0 - 110.0
100.0 - 105.0
95.0 - 100.0
90.0 - 95.0
85.0 - 90.0
80.0 - 85.0
75.0 - 80.0
70.0 - 75.0
65.0 - 70.0
60.0 - 65.0
55.0 - 60.0
50.0 - 55.0
45.0 - 50.0
40.0 - 45.0
35.0 - 40.0
30.0 - 35.0
25.0 - 30.0
20.0 - 25.0
15.0 - 20.0
10.0 - 15.0
5.0 - 10.0

258.7 (5.0) 339.0 261.5 ( 5.0 ) 389.8 258.5 ( 5.0 ) 389.8 0.0 - 5.0
below 0.0

Fig. 9.19 Axisymmetric Monodomain model with intramural fiber rotation on a truncated ellip-
soidal domain with central endocardial stimulation. Repolarization time isochrones on endocardial
(e), mid-myocardial (d), epicardial (a) and different transmural sections (b-c-f). All values are
expressed in ms

and an index related to  .x/, where is the activation time at point x. We

remark that,
p for isotropic homogeneous media, the anisotropic curvature reduces
to K D  k, with k D  div.r =kr k/ D  .x/  rkr kT r , hence the
index used in Burgess et al. [67] is only a part of the front curvature in isotropic
media. Therefore Burgess et al. [67] implicitly established a correlation between the
shape of the wave front and the APD distribution, which is here confirmed using the
more appropriate anisotropic curvature index.
Similar considerations hold for the activation, repolarization and APD isochrones
elicited by an endocardial stimulation on a truncated ellipsoidal domain, see
Figs. 9.18–9.20.
284 9 Simulation Studies of Cardiac Bioelectrical Activity

a b d

above 259.5

244.5 ( 0.5 ) 250.9 244.9 ( 0.5 ) 258.1 246.1 ( 0.5 ) 251.7 259.0 - 259.5
258.5 - 259.0
258.0 - 258.5
257.5 - 258.0

f e
257.0 - 257.5
256.5 - 257.0
256.0 - 256.5
255.5 - 256.0
255.0 - 255.5
254.5 - 255.0
254.0 - 254.5
253.5 - 254.0

c 253.0 - 253.5
252.5 - 253.0
252.0 - 252.5
251.5 - 252.0
251.0 - 251.5
250.5 - 251.0
250.0 - 250.5
249.5 - 250.0
249.0 - 249.5
248.5 - 249.0
248.0 - 248.5
247.5 - 248.0
247.0 - 247.5
246.5 - 247.0
246.0 - 246.5
245.5 - 246.0
245.0 - 245.5
244.5 - 245.0
244.0 - 244.5
243.5 - 244.0
243.0 - 243.5
242.5 - 243.0
242.0 - 242.5
241.5 - 242.0
241.0 - 241.5
240.5 - 241.0

244.6 (0.5) 257.6 244.1 ( 0.5 ) 253.5 244.1 ( 0.5 ) 258.4 240.0 - 240.5
below 240.0

Fig. 9.20 Axisymmetric monodomain model with intramural fiber rotation on a truncated ellip-
soidal domain with central endocardial stimulation. Action potential duration isochrones on
endocardial (e), mid-myocardial (d), epicardial (a) and different transmural sections (b-c-f). All
values are expressed in ms

9.3 Heterogeneous Cardiac Tissue

In this section, we extend the simulations of excitation and repolarization sequences

of the previous section to cardiac slabs with heterogeneous intrinsic properties of
the cell membrane. More complete results can be found in [122, 126].
The cardiac domain considered is a cartesian slab of dimensions fL1  L2 
L3 g D f5  5  1g cm3 modeling a portion of the left ventricle. A structured grid of
500  500  100 hexahedral isoparametric Q1 elements with size h D 0:1 mm was
used in all computations.
Parameter calibration. In a parallelepipedal geometry, modeling a portion of
the ventricular wall, we consider the following simple fiber-sheet model sharing
the previous qualitative rules. In our slab geometry using the cartesian coordinate
system fx1 ; x2 ; x3 g and associated axis unit vectors fue1 ; ue 2 ; ue3 g, the x3 coordinate
represents the transmural thickness from the endo to epicardial planes. Proceeding
9.3 Heterogeneous Cardiac Tissue 285

from endocardium to epicardium, we consider a linear transmural fiber rotation

(clockwise) for a total amount of 120ı , i.e.
˚
al .x/ D ue1 cos ˛.x3 / C ue 2 sin ˛.x3 /; ˛.r/ D 23 .1  r/ 4 ; 0r1:

The myocardial tissue is represented as a set of surfaces, called radial laminae or

sheets, that connect the epicardial with the endocardial surface having tangent plane
defined by al and by the transmural transverse fiber direction at D ue3 ; therefore,
the sheet normal is given by an .x/ D ue1 sin ˛.x3 / C ue2 cos ˛.x3 /, i.e. laminae
develops intramurally with zero sheet angle and intersects perpendicularly the epi-
and endocardial surfaces.
For our slab geometry, the conductivity tensors are Di;e Dti;e I C.li;e 
t /al aTl C .ni;e  ti;e /an aTn . In any model of the cardiac tissue, the calibration of
i;e

the parameter values is not an easy task, because the data available in the literature
refer to diverse experimental setups. The macroscopic conductivities of the media
are here based on conservative values and are determined by requiring that the
simulated excitation wavefront motion has physiologically reasonable velocities of
propagations, see [102, 123]. The conductivity coefficients, in mS cm1 , used in the
orthotropic anisotropic simulations are:

le D 2 te D 1:35 li D 3 ti D 0:465 ne D 0:5te ; ni D 0:1ti :

In the numerical tests, we have used the parameters Cm D 103 mF=cm2 ;  D

103 cm1 . The previous calibration parameters and the LR1 model, in case of
ideal plane wavefronts moving along the directions al ; at ; an , respectively, produce
conservative estimates of the propagation velocities equal to (60, 35, 10) cm s1 ,
respectively. More precisely, one dimensional simulations using as a conductivity
coefficient l;t;n D l;t;n
i e
l;t;n i
=.l;t;n e
l;t;n /, yield propagation velocities equal to
(60.6, 32.01, 9.59), respectively, using a space resolution of 0.05 mm. These
velocities agree with those measured in canine experiments along fiber and across
fiber in the transmural direction, see [401, Table 16-1, p. 148]. Since the coupling
between adjacent laminae occurs through thin, sparse, muscle bundles bridging the
clefts that separate adjacent laminae (see [293]), it is expected that the intrinsic
cross-fiber velocity should be 2 or 3 times greater in the plane of a muscle lamina
than perpendicular to it. The conductivity coefficients n1;e have been calibrated so
that the propagation velocities (n ) of an ideal planar wavefronts moving across
the fiber sheet is reduced by a factor 3.5 with respect to t . This epicardial cross-
fiber velocity is difficult to evaluate experimentally since the underlying fiber
rotation accelerates the excitation epicardial motion, as shown in our work [121],
for epicardial cross-fiber velocities in the range 14–25 cm/s.
286 9 Simulation Studies of Cardiac Bioelectrical Activity

The initial conditions for the Monodomain model with the phase I Luo-Rudy
model (LR1) [308] are chosen at the steady state, so that the rest potential is v0 D
83:84 mV. In order to elicit the excitation front, we apply a stimulus of 250 A
for 1 ms on a small volume (containing 5 mesh points in each x and ydirection
and 3 mesh points in the zdirection) at the center of the epi- or endocardial face.
Other than potentials and gating variables, at each time-step, we compute also the
activation (ACTI) and the repolarization (REPO) times, defined as the times when
the action potential crosses 60 mV during the upstroke and when it reaches 75:46
in the downstroke phase (i.e. 90 % of the resting value), respectively.
Intrinsic transmural heterogeneity. We consider three different types of
transmural distribution of the intrinsic APDs of the cells, the first homogeneous
(denoted by H-slab) and the other two heterogeneous, (denoted by 3-slab and W-
slab). We assume that the transmural intrinsic heterogeneity is the same along any
transmural epi-endocardial straight line, i.e. in any plane parallel to the epicardium
all cells have the same APD.
• H-slab: the cardiac slab has homogeneous intrinsic properties of the cellular
membrane; the IK current of the LR1 model is multiplied by 2.325, yielding
an APD of 250 ms.
• 3-slab: the cardiac slab is subdivided transmurally into three layers of 1=3 cm
thickness. More precisely, proceeding from the endocardial to the epicardial
surfaces, the values of intrinsic cellular APD amount to 235, 272, 225 ms
on the intervals Œ0; 0:33/; Œ0:33; 0:66/; Œ0:66; 1, respectively. The transmural
heterogeneity is simulated by multiplying the current IK of the LR1 model
by (2.62, 1.952, 2.88), corresponding to intrinsic APDs of (235, 272, 225)
ms, respectively. See Fig. 9.21, left panel and the dashed plots in Fig. 9.23.
Figure 9.21 left panel, displays the three different action potential waveforms
related to the three different intrinsic properties of transmural cells, i.e. they are

3−SLAB W−SLAB
50 50

0 0
AP (mV)

Mid
AP (mV)

Mid
Subendo

Endo
Endo

−50 Epi −50

Epi

−100 −100
0 100 200 300 0 100 200 300

Fig. 9.21 Intrinsic action potential waveforms assigned to the intramural heterogeneity of type
3-slab (left) and W-slab (right). Each panel shows subendocardial, midwall and subepicardial
intrinsic action potentials obtained with 0D simulations and assigned to each transmural layer
9.3 Heterogeneous Cardiac Tissue 287

obtained by 0D simulations modeling the intrinsic response of the isolated cell.

We have considered this kind of transmural heterogeneity composed of three
layers of cellular types, (i.e. epi-subepicardial, mid-myocardial and subendo-
endocardial cells), since this simplified heterogeneity model was previously used
in simulations with 1D models in [546] and in [96].
• W-slab: the cardiac slab is subdivided transmurally into four layers with different
intrinsic membrane properties in order to reproduce qualitatively the APD
transmural behavior measured in wedge preparations, see e.g. [572], Fig. 4 and
[395] Fig. 5. More precisely, proceeding from the endocardial to the epicardial
surfaces, we subdivided the slab into four layers of thickness (0.12, 0.2, 0.55,
0.15) cm, respectively, and multiplied the current IK of the LR1 model by (2.715,
1.952, 2.47, 2.88), corresponding to intrinsic APDs of (232, 272, 242, 225)
ms, respectively, right panel. Hence, we assume that sub-endocardial and mid-
myocardial layers display a longer APD than the epi- and endocardial cells. See
Fig. 9.21 and the dashed plots in Fig. 9.23.
We remark that the two transmural heterogeneity assignments are not symmetric
in the slab thickness, therefore they could yield a different APD dispersion for
an epicardial and an endocardial pacing. In this study, we consider both epi- and
endocardial stimulations for both types of transmural heterogeneities, 3-slab and
W-slab.

9.3.1 Transmural Heterogeneity in 3-D Cardiac Slabs

We now consider 3-D simulations of the excitation and repolarization processes,

elicited by an epicardial or an endocardial stimulation, in an orthotropic slab
with homogeneous (H-slab) or heterogeneous (3-slab, W-slab) intrinsic cellular
membrane properties. Our main goal is to determine how intramural heterogeneity
affects the repolarization sequence and the spatial distribution of action potential
durations.
Because the complete set of results of these 3-D simulations are too large to
report for all sections, slab types and stimulation sites, we show first the results
along simple 1-D and 2-D transmural sections of the domain and then show the
results on the whole 3-D slab for endocardial stimulation. In particular, we present
the results on:
• A 1-D transmural line passing through the stimulation site and orthogonal to both
epicardial and endocardial surfaces, for both stimulations;
• 2-D transmural planar sections passing through the center of the slab and
orthogonal to both epicardial and endocardial surfaces (see Fig. 9.22):
- for endocardial stimulation, we focus on section C for all slabs, on sections A,
B and D for the W-slab;
• The whole 3-D slab and its planar intramural sections, parallel to the epicardium,
for endocardial stimulation.
288 9 Simulation Studies of Cardiac Bioelectrical Activity

o
A (45 )

o
B (90 )

o
D (0 )
C
EPI

C (−45o)
ENDO

Fig. 9.22 Intramural sections A, B, C, and D of the orthotropic Monodomain slab of size 5 5 
1 cm3 , with isochrone contours elicited by epicardial central stimulation

We will then show some results on homogeneous and heterogeneous 3-D curved
walls.
In each case, we will examine the distributions of activation times (ACTI(x)),
repolarization times (REPO(x)) and action potential durations (APD(x) D
REPO(x)–ACTI(x)), generated by a stimulus at the center of the epicardial or
endocardial surface of each slab.
We remark that fiber direction at a depth of 0 cm from the epicardium is parallel
to section C; at 0.375, to section D; at 0.75 cm, to section A, while the fiber direction
is never parallel to section B.

9.3.1.1 1-D Transmural Line

(a) Epicardial stimulation. In the homogeneous case (H-slab), the profiles of

the activation time, displayed in Fig. 9.23 (top panel, top row left), are almost
linear. The intramural excitation wavefront reaches a quasi-stationary speed across
fibers, apart from the short acceleration in the initial phase of propagation and the
subsequent collision with the endocardial plane. On the other hand, the profile of
the repolarization time shows a less linear behavior and the recovery wavefront
does not reach a steady intramural speed, due to the interaction between the wave
length of the fast component of the repolarization process, i.e. the downstroke
phase of the action potential around the inflection point; we remark that, for an
across fiber velocity of 0.3 mm/ms, see Fig. 4 and a conservative duration of the fast
repolarization phase of the AP of 60 ms, it follows an estimate of 18 mm for the
wave length, which is greater than the wall thickness of the slab.
9.3 Heterogeneous Cardiac Tissue 289

ACTI REPO APD

40 290 260
280
H−slab 255
20 270
250
260
0 250 245
0 0.5 1 0 0.5 1 0 0.5 1
MIN = 0.13 MAX = 35.26 MIN = 257.58 MAX = 282.20 MIN = 246.93 MAX = 257.46

40 280
280
260
3−slab

20 260
240
240
0 220
0 0.5 1 0 0.5 1 0 0.5 1
MIN = 0.13 MAX = 35.26 MIN = 233.20 MAX = 279.96 MIN = 231.10 MAX = 260.09

40 280
280
30
260
W−slab

20 260
240
10
240
0 220
0 0.5 1 0 0.5 1 0 0.5 1
EPI MIN = 0.13 MAX = 35.26 MIN = 236.52 MAX = 283.31 MIN = 235.66 MAX = 255.57

ACTI REPO APD

40 290 260
280
255
H−slab

20 270
250
260
0 250 245
0 0.5 1 0 0.5 1 0 0.5 1
MIN = 0.13 MAX = 35.48 MIN = 258.42 MAX = 282.32 MIN = 246.84 MAX = 258.29

40 280 280

260
3−slab

20 260
240

0 240 220
0 0.5 1 0 0.5 1 0 0.5 1
MIN = 0.13 MAX = 35.48 MIN = 245.15 MAX = 278.65 MIN = 223.44 MAX = 260.80

40 280 280

30 270
260
W−slab

20 260
240
10 250

0 240 220
0 0.5 1 0 0.5 1 0 0.5 1
ENDO MIN = 0.13 MAX = 35.48 MIN = 252.94 MAX = 267.68 MIN = 226.79 MAX = 258.57

Fig. 9.23 Profiles along the 1-D transmural line (passing through the stimulation site and
orthogonal to both epicardial and endocardial surfaces) of the activation time (ACTI, first column),
of the repolarization time (REPO, second column), of the action potential duration (APD, third
column) for H-slab (first row), 3-slab (second row), W-slab (third row). Central stimulation:
epicardial (top panel), endocardial (lower panel). The piecewise constant dashed lines in the APD
column are the intrinsic (0D) APD of the cells. Vertical axis reports times in ms and the horizontal
axis reports distances in cm from the endocardial site
290 9 Simulation Studies of Cardiac Bioelectrical Activity

Therefore, the intramural repolarization sequence is faster than the depolarization

sequence. This brings about a progressive intramural reduction of the APD away
from the stimulation site, see Fig 9.23. The time required for activation and recovery
to reach the endocardial side is about 35.26 and 24.62 ms, respectively. Thus,
transmural repolarization moves slightly faster than activation in the homogeneous
model, see Gotoh et al. [208] and Taccardi et al. [512]. The APD dispersion amounts
to about 10.53 ms, mostly concentrated around the stimulation site.
In the heterogeneous cases 3-slab, W-slab (middle and last rows of top panel of
Fig. 9.23), as expected, the activation time profiles are essentially unchanged with
respect to the H-slab, while the repolarization times have a higher dispersion of
about 47 ms. The APD dispersion amounts to 29 and 20 ms in the 3-slab and W-slab,
respectively, i.e., it is three or two times larger than in the homogeneous case. Due
to APD modulation induced by electrotonic currents, the intrinsic APD differences
between the three or four different cell layers are strongly smoothed, as shown
by comparing the profiles with the piecewise constant dashed lines in Fig. 9.23
right column, which displays the assigned intrinsic transmural APD distribution
in the three slab types. Moreover, both the 3-slab and W-slab cases exhibit a
strongly reduced transmural APD dispersion with respect to the intrinsic dispersion,
which amounts to 47 ms. In both heterogeneous slabs, the transmural profiles of the
repolarization time exhibit a maximum located in the cell layer with the greatest
intrinsic APD; this site is also associated with the collision of the primary recovery
front, which starts at the epicardial stimulation site, with a secondary recovery wave,
which starts from the endocardium. Also, the APDs reach a maximum in the layer
with highest intrinsic APD, but the magnitude is strongly reduced, compared with
the intrinsic value. As expected, longest APD is observed near the collision of the
recovery waves.
(b) Endocardial stimulation. We now examine the results described in the lower
panel of Fig. 9.23, obtained with an endocardial stimulation at the center of the
endocardial face of the slab. In the H- slab case, apart from the change of the epi
and endocardial sides, the activation, recovery and APD profiles practically coincide
with those elicited by the epicardial stimulation. The propagation of the excitation,
following epicardial and the endocardial pacing, along the transmural vertical line,
is strictly across fibers for symmetry reasons. Note that we did not incorporate fiber
imbrication in our model. For the endocardial stimulation, the total activation and
recovery times are 35.48 and 23.90, respectively, while in the epicardial stimulation
they are 35.26 and 24.62, respectively. These small differences in the total activation
and recovery times are expected and are mainly due to differences in the initial
direction and shape of the propagating excitation wavefront, since in all our slabs,
the endocardial fiber direction is 75ı , while on the epicardial surface it is 45ı .
In 3-slab and W-slab, transmural excitation profiles coincide with those of H-slab.
The repolarization processes are completed in 33.5 and 14.74 ms, respectively. This
large difference in the dispersion of the repolarization times is associated with the
different morphology of the transmural profiles, i.e. the presence of one maximum
in 3-slab, located at the midwall level, and of two relative maxima of comparable
values in W-slab, located one at subendocardial and the other at subepicardial level.
9.3 Heterogeneous Cardiac Tissue 291

The midwall maximum in 3-slab is related to the collision of the primary recovery
front, originated at the endocardial stimulation site, with a secondary phase front,
originating at the epicardial side at a later instant. In W-slab, two collisions of
the recovery fronts are present because an additional recovery wave, starting from
the intramural site corresponding to a minimum of the repolarization time profile,
spreads toward the endocardium and the other toward the epicardium. This dynamic
behavior is confirmed by the repolarization speed profile in which, the intramural
maxima correspond to points of collision.
The APD dispersions, amounting to 37.36 (3-slab) and to 31.78 ms (W-slab), are
about four and three times greater than in the H-slab.
In comparison with the H-slab case, the endocardial stimulus yields a reduction
or increase of the recovery dispersion depending on the type of heterogeneity.
Conversely, the epicardial stimulus produces recovery dispersions, in both 3-slab
and W-slab, that are roughly twice the H-slab dispersion. For both stimulations, the
APD dispersions are higher than the H-slab APD dispersion, with greater values
in case of endocardial stimulation and 3-slab heterogeneity. Thus, the type of
intrinsic transmural heterogeneities could be inferred by comparing repolarization
profiles and APD dispersions along the 1-D transmural line, where epicardial and
endocardial stimulations produce very different values of repolarization and APD
dispersions.
Comparing the results of the two stimulations, we observe that the repolarization
profiles display greater morphological changes than the APD profiles, suggesting a
major sensitivity of the former to the distribution of transmural heterogeneity.

9.3.1.2 Transmural Sections: Endocardial Stimulation

We first describe the transmural patterns elicited by a central endocardial stimula-

tion, displayed in Fig. 9.24, on section C of the three cases H-slab, 3-slab, W-slab.
(a) Transmural excitation sequence on section C. As occurred with epicardial
stimulation, the excitation sequence is unaffected by the transmural heterogene-
ity. We note that the epicardial side of section C is parallel to the fiber direction,
while its endocardial side forms an acute angle of 60ı with fiber direction.
After epicardial breakthrough, the foot of the wavefront on the upper side of the
section propagates along fibers, while the endocardial foot propagates slowly,
mainly across fibers and is overtaken by the subepicardial portions of the front.
This generates excitation pathways that return toward the endocardium. The
latest depolarized areas are near the vertices of section C on the endocardial
side.
(b) Transmural repolarization sequence on section C. In the H-slab case,
the repolarization isochrones follow the activation sequence: return pathways
toward the endocardial plane of stimulation are present and the areas that
repolarize last coincide with those that depolarize last.
292 9 Simulation Studies of Cardiac Bioelectrical Activity

ACTI ENDO 100

80

H−slab
60
40
20
0.14 114.08 4.00

100
80
3−step

60
40
20
0.14 114.08 4.00

100
80
W−slab

60
40
20
ACTI 0.14 114.08 4.00

REPO ENDO
340
320
H−slab

300
280
258.44 358.33 4.00 260

350
3−step

300

245.17 357.33 4.00 250

340
320
W−slab

300
280
252.95 360.27 4.00 260
REPO

APD EPI
255
H−slab

250

244.25 258.30 0.50 245

250
3−slab

240

230
223.18 260.80 4.00

250
W−slab

240

226.52 258.54 4.00 230

APD

Fig. 9.24 Isochrone lines of the activation time (ACTI, top), repolarization time (REPO, middle)
and action potential duration (APD, bottom) on the transmural section C for models H-slab, 3-slab,
W-slab, with endocardial stimulation. Reported below each panel are the maximum, minimum and
step in ms of the displayed map (Reproduced with permission from [126])
9.3 Heterogeneous Cardiac Tissue 293

In the 3-slab case, recovery starts at the stimulation site. After about 13 ms,
a secondary recovery wave starts from the opposite point on the epicardial side.
After collision at midwall level, the merged repolarization front spreads laterally
with a smooth V-shaped profile. In the W-slab case, as discussed in Sect. 9.3.1.1,
the recovery process arises from three different sites and at different times: at
253 ms at the endocardial stimulation site, at 262 ms from the opposite point on the
epicardial side. Finally, at 265 ms, a new repolarization front arises at about midwall
point. After the merging of these three different recovery fronts, the transmural
isochrones exhibit V-shape profiles with some initial undulation associated with the
four different transmural layers of intrinsic heterogeneity.
The repolarization dispersion on section C amounts to (99.89, 110.84, 108.32)
ms for H-slab, 3-slab, W-slab, respectively. These values are lower then those for
activation dispersion, which amount to about 114.08 ms. This confirms that recovery
is slightly faster than excitation, especially in the homogeneous case (Table 9.3).
(c) Transmural APD distribution on section C. The APD dispersion on section
C amounts to (14.05, 37.62, 32.02) ms for H-slab, 3-slab, W-slab, respectively,
These values are comparable to the values of APD dispersion observed along
the 1-D transmural line described in Sect. 9.3.1.1. In the H-slab, the APD distri-
bution, displayed in Fig. 9.24 (bottom panel), exhibits the absolute maximum
at the stimulation site and two symmetric subepicardial minima. Shortening
of the APD and associated relative minima are present at points of front-
boundary collision of the excitation wavefront. Two maxima on the epicardial
side appear symmetrically with respect to the minimum located at the excitation
breakthrough point on the epicardium.
In both heterogeneous slabs, the APD distribution displays a striped pattern; the
stripe having the highest APD has the same location as the intramural layer with the
highest intrinsic APD.

9.3.1.3 3-D Slab: Epicardial, Endocardial and Planar Intramural Sections

Parallel to the Epicardium

We now consider the excitation, repolarization and APD values on the whole 3-D
slabs and their planar intramural sections parallel to the epicardial face, with both
endocardial and epicardial pacing.

Table 9.3 Dispersion (MAX–MIN value) of activation time (ACTI), repolarization time (REPO),
APD 90 on the transmural sections A and C
Epicardial stimulation, section A Endocardial stimulation, section C
ACTI REPO APD 90 ACTI REPO APD 90
H-slab 100.53 88.00 12.39 114.08 99.89 14.05
3-slab 100.53 107.81 38.42 114.08 110.84 37.62
W-slab 100.53 95.21 30.54 114.08 108.32 32.02
294 9 Simulation Studies of Cardiac Bioelectrical Activity

(a) H-slab. In the homogeneous case, displayed in Fig. 9.25, the spread of excita-
tion and the sequence of recovery, displayed on five planar sections parallel to
the epicardial face (located at z = 0, 0.25, 5, 0.75, 1 cm, respectively), undergo
an acceleration in endocardial areas where the fronts proceed mainly across
fibers, due to the presence of underlying intramural returning pathways of
excitation and recovery, as shown by the isochrones displayed on the transmural
section C with endocardial stimulation (Fig. 9.24). Dimple-like inflections
appear in the isochrone profiles, due to the faster propagation of the fronts in
higher layers where the fiber direction rotates CW relatively to the lower planes.
The recovery isochrones on the endo, intramural and epicardial planes exhibit a
somewhat smoother shape and faster propagation compared with the excitation
sequence.
The rotational anisotropy of the media yields APD patterns strongly correlated
with the excitation wavefront motion and the front-boundary collisions. The APD
spatial distribution shows finger-shape valleys of APD shortening associated to the
dimple-like inflections appearing in the excitation and recovery fronts. The APD
level lines display also dog-bone shaped profiles. These features can be attributed
to the rotational anisotropic structure of the ventricular wall and to the interaction
between excitation and repolarization return pathways. Moreover, APD maxima
occur near the stimulation site and at breakthrough points on each plane section
parallel to the epicardium (except on the epicardial plane), see [123, Fig. 7] for more
detailed analysis of the APD pattern.
Similar features occur in case of epicardial stimulation (not shown), if we
exchange in the description the role of endocardial and epicardial planes.
(b) 3-slab and W-slab. We now compare the H-slab patterns with those of
the heterogeneous 3-slab and W-slab, see Figs. 9.26 and 9.27. Unexpectedly,
the repolarization sequence and the APD patterns on the intramural planar
sections parallel to the epicardium exhibit the same main anisotropic spatial
features as in the homogeneous case. Therefore, the transmural heterogeneity
remains mostly confined in the transmural direction and does not diffuse to the
orthogonal planes, in spite of the combined effects of electrotonic modulation,
orthotropic conduction velocity and fiber rotation. We have computed the
correlation coefficients of the repolarization time and APD between H-slab and
3-slab and between H-slab and W-slab on each of the 101 intramural planar
sections parallel to the epicardial face (the first being the endocardium and the
last the epicardium). These coefficients show a remarkably strong correlation
between homogeneous and heterogeneous repolarization times (the coefficients
differ from 1 by less than 2  104 ) and APD (the coefficients differ from 1
by at most 5:2  102 with epicardial stimulation and at most 7:1  102 with
endocardial stimulation), despite the different magnitude of the repolarization
times and APDs.
9.3 Heterogeneous Cardiac Tissue 295

ACTI REPO APD

1 CM (EPI)

35.48 88.31 4.00 282.32 335.33 4.00 244.67 251.11 0.50

0.75 CM

27.16 91.66 4.00 278.00 339.18 4.00 246.46 250.84 0.50

0.5 CM

18.42 98.75 4.00 270.20 347.37 4.00 246.90 251.77 0.50

0.25 CM

9.46 107.07 4.00 262.70 355.00 4.00 246.31 253.23 0.50

0 CM (ENDO)

0.14 114.08 4.00 258.44 358.33 4.00 244.23 258.30 0.50

100 340
80 255
320
60
300
40 250
20 280

245
0.14 114.08 4.00 258.44 358.33 4.00 244.23 258.32 0.50

Fig. 9.25 Orthotropic Monodomain–LR1 model, homogeneous H-slab of size 5  5  1 cm3 .

Endocardial central stimulation. Isochrone lines of the depolarization time (first column ACTI),
repolarization time (second column REPO) and action potential duration (third column APD)
on five horizontal sections (z D 0, 0.25, 0.5, 0.75, 1 cm) and the whole slab. Reported below
each panel are the maximum, minimum and step in ms of the displayed map (Reproduced with
permission from [126])
296 9 Simulation Studies of Cardiac Bioelectrical Activity

ACTI REPO APD

1 CM (EPI)

35.48 88.31 4.00 258.92 311.55 4.00 221.26 227.24 0.50

0.75 CM

27.16 91.65 4.00 262.79 323.30 4.00 230.97 235.63 0.50

0.5 CM

18.42 98.75 4.00 278.61 355.46 4.00 255.27 260.22 0.50

0.25 CM

9.46 107.07 4.00 257.34 350.09 4.00 241.24 247.88 0.50

0 CM (ENDO)

0.14 114.08 4.00 245.17 345.70 4.00 231.51 245.03 0.50

350
100
80 250
60 300 240
40
230
20
250
0.14 114.08 4.00 245.17 357.33 4.00 221.26 260.22 0.50

Fig. 9.26 Orthotropic Monodomain–LR1 model, heterogeneous 3-slab of size 551 cm3 . Same
format as in Fig. 9.25 (Reproduced with permission from [126])
9.3 Heterogeneous Cardiac Tissue 297

1 CM (EPI) ACTI REPO APD

35.48 88.31 4.00 262.27 314.89 4.00 224.64 230.57 0.50

0.75 CM

27.16 91.66 4.00 266.70 327.77 4.00 235.17 239.54 0.50

0.5 CM

18.42 98.75 4.00 265.05 341.78 4.00 241.69 246.63 0.50

0.25 CM

9.46 107.07 4.00 267.53 359.62 4.00 251.02 258.07 0.50

0 CM (ENDO)

0.14 114.08 4.00 252.95 353.49 4.00 239.40 252.81 0.50

100
340
80 250
320
60
300 240
40
280
20 230
260
0.14 114.08 4.00 252.95 360.27 4.00 224.64 258.07 0.50

Fig. 9.27 Orthotropic Monodomain–LR1 model, heterogeneous W-slab of size 5  5  1 cm3 .

Same format as in Fig. 9.25 (Reproduced with permission from [126])
298 9 Simulation Studies of Cardiac Bioelectrical Activity

In the intramural sections parallel to the epicardium, the APD patterns elicited
by an endocardial (epicardial) central stimulation in the homogeneous or heteroge-
neous slabs are characterized by the following features:
(i) The APD distributions on the endocardial (epicardial) and intramural planes
exhibit a maximum located at the endocardial (epicardial) stimulation site
or at the intramural points first reached by the excitation front, respectively;
see Figs. 9.25–9.27. These maxima are surrounded by level lines elongated
along the local fiber direction. This indicates that APD decreases more rapidly
when moving away from the center of the face in the cross-fiber direction
than along fibers. This finding confirms and extends to a three-dimensional
myocardial slab the experimental findings by Gotoh et al. [208], Osaka et al.
[364], Taccardi et al. [512].
(ii) On the intramural sections (from subendocardial (subepicardial) to midwall),
two valleys of decreasing APD values occur. Each valley exhibits a minimum
located at the boundary of the slab. These narrow valleys of low APD values
are located in the regions where excitation isochrones exhibit a dimple-like
inflections.
(iii) In the epicardial (endocardial) plane, the APD distribution displays a saddle
point at the epicardial (endocardial) BKT point. The APD increases reaching
a maximum when moving away from the BKT point in a direction parallel
to the epicardial (endocardial) fibers of 45ı (75ı ), while in the cross-fiber
direction the APD decreases toward two minima located at the boundary and
corresponding to front-boundary collision points. The axis joining the two
maxima is roughly parallel to the epicardial (endocardial) fiber direction of
45ı (75ı ) and the two maxima are separated by the valley of APD reduction.
We remark that the heterogeneity we consider is only transmural, i.e. the
intrinsic membrane properties are constant in each intramural plane parallel to the
epicardium. Our results show that the presence of transmural heterogeneities cannot
be detected from the epicardial pattern of the APD distribution.
After the endocardial or epicardial breakthrough, the repolarization fronts spread
laterally in the slab (Fig. 9.24) and display similar shapes associated with the 1-D
transmural profiles of Fig. 9.23, independently of the pacing location.

9.3.2 Transmural Heterogeneity in 3-D Ellipsoids

We have extended the previous simulations to the case of a truncated ellipsoidal

domain with transmural homogeneous and heterogeneous membrane properties.
The numerical simulations are performed on a simplified left ventricular wall
shaped as an ellipsoidal volume, axisymmetric with respect to the vertical z–axis,
truncated at the base and at the apex, described by the parametric equations

x D a.r/ cos  cos ; y D a.r/ cos  sin ; z D c.r/ sin :

9.3 Heterogeneous Cardiac Tissue 299

Here r 2 Œ0; 1; 2 Œ min ; max ;  2 Œmin ; max ; a.r/ D a1 Cr.a2 a1 /; c.r/ D
c1 C r.c2  c1 / and ai ; ci ; i D 1; 2 are given coefficients determining the main
axes of the ellipsoid. We denote by r-layers the ellipsoidal surfaces described by
varying ;  for a fixed value of r; analogously, we denote by -layers the plane
surfaces described by varying r;  for a fixed value of and by -layers the surfaces
described by varying r; for a fixed value of  (see Fig. 9.28 and Table 9.4).
We consider transmural 3-wall and W-wall heterogeneous distribution of intrinsic
APD. More precisely:
W-wall. The cardiac tissue is subdivided transmurally into four layers: endo-
cardial (r 2 Œ0; 0:12/), sub-endocardial (r 2 Œ0:12; 0:32/), midmyocardial (r 2
Œ0:32; 0:85/) and epicardial (r 2 Œ0:85; 1). The IK factor assumes the following
values
8
ˆ
ˆ 2:715 r 2 Œ0; 0:12/ (endo)
<
1:952 r 2 Œ0:12; 0:32/ (sub-endo)
factW D
IK
ˆ 2:47 r 2 Œ0:32; 0:85/ (mid)
:̂
2:88 r 2 Œ0:85; 1 (epi);

θ − layer
φ−layer
(θ = π/8)

θ−layer

S
S2 1
r−layer ENDO (r=0)
P1
P2 MID (r=0.5)
P3
φ−layer
EPI (r=1) r−layers
(φ = − π/2)

Fig. 9.28 Left: ellipsoidal domain and r, ;  -layers, S1 ; S2 D endocardial and epicardial
stimulation sites, P1 ; P2 ; P3 D Purkinje ventricular junctions. Right: fiber direction on r, ;  -
layers

Table 9.4 Parameter calibration for the monodomain and bidomain models
 D 103 cm1 Cm D 103 mF=cm2
le D 2  103 1 cm1 li D 3  103 1 cm1
te D 1:3514  103 1 cm1 ti D 3:1525  104 1 cm1
ne D te =2 ni D ti =10
min D =2, max D =2 min D 3=8, max D =8
a1 D 1:5, a2 D 2:7 c1 D 4:4, c2 D 5
n D 500, n D 500, nr D 100
300 9 Simulation Studies of Cardiac Bioelectrical Activity

which yield APDs of 232 ms (endocardial), 272 ms (sub-endocardial), 242 ms

(midmyocardial) and 225 ms (epicardial). This type of transmural heterogeneity
composed of four cell layers reproduces qualitatively the APD heterogeneity
measured in wedge preparations, see e.g. [572].
3-wall. The cardiac tissue is subdivided transmurally into three layers: endocar-
dial (r 2 Œ0:0:33/), midmyocardial (r 2 Œ0:33; 0:66/) and epicardial (r 2 Œ0:66; 1).
The factor for the IK current assumes the following values
8
< 2:62 r 2 Œ0; 0:33/ (endo)
fact3IK D 1:952 r 2 Œ0:33; 0:66/ (mid)
:
2:88 r 2 Œ0:66; 1 (epi);

which yield APDs of 235 ms (endocardial), 272 ms (midmyocardial) and 225 ms

(epicardial). The three resulting action potentials waveforms are shown in Fig. 9.29.

50 280

SUB−ENDO APD (ms)

0 260
AP (mV)

MID

−50 EPI 240

ENDO

−100 220
0 100 200 300 ENDO MID EPI
50 280

MID
APD (ms)

0 260
AP (mV)

ENDO

−50 240
EPI

−100 220
0 100 200 300 ENDO MID EPI

50 270

260
APD (ms)

0
APEX
AP (mV)

250

−50 240
BASE
230
−100
0 100 200 300 APEX MID BASE

Fig. 9.29 Top: W-wall transmural intrinsic action potential profiles (left) and transmural intrinsic
APD distribution (right). Middle: 3-wall transmural intrinsic action potential profiles (left) and
transmural intrinsic APD distribution (right). Bottom: apex-to-base intrinsic action potential
profiles (left) and apex-to-base intrinsic APD distribution (right)
9.3 Heterogeneous Cardiac Tissue 301

This type of transmural heterogeneity composed of three cell layers has been
previously used in [546] for 1D simulations.
Figures 9.30–9.33 display the patterns of excitation, repolarization and APD
isochrones for the three ellipsoidal walls with endocardial stimulation. The same
qualitative features of the excitation, repolarization and APD patterns previously
described for the slab domains are also detected in these ellipsoidal walls. The visual
inspection of the repolarization patterns on r-layers of the H-wall, 3-wall and W-wall
shows that they are very similar, as confirmed by the high CC ( 0:991) reported
in Table 9.6. On the other hand, the APD correlation is lost on some r-layers. This
correlation loss of the intramural APD patterns could be attributed to the interaction
between the wall tapering and the associated curvature variation. This conclusion
differs from the one obtained for cartesian slabs, where both repolarization and
APD patters on intramural r-layers were found to be independent of the transmural
variation of the intrinsic cellular APD.

9.3.2.1 Discussion

In the present investigation we studied the combined influence of (a) the transmural
rotational anisotropy and (b) the transmural intrinsic APDs heterogeneity on the
sequence of repolarization and APD distribution. We have used the Anisotropic
Monodomain model coupled with the Luo-Rudy phase I membrane model, which
has a lower computational cost than more advanced models incorporating additional
ionic currents and a more complete intracellular calcium dynamics, see e.g. [243,
356]. We applied a single stimulus, either epicardial or endocardial, and compared
the results obtained with homogeneous membrane properties (H-slab/wall) with the
results obtained with two types of transmural heterogeneity (3-slab/wall and W-
slab/wall). By considering action potentials elicited by a single stimulus (as opposed
to a sequence of stimuli), our simulations are expected to enhance the disparity
between excitation and recovery sequences and the dispersion of APDs, since it is
well known that periodic stimulations, with relatively short cycle lengths, produce
relatively short APDs with a reduced dispersion, see [479, Fig. 3].
In case of homogeneous membrane (H-slab/wall), the repolarization wave
elicited by epi- or endocardial stimulation spreads anisotropically, like the
corresponding activation pattern. Repolarization exhibits a somewhat faster motion,
with intramural anisotropic patterns of APD distribution in sections parallel to
the epicardial face and on transmural sections A, B, C, and D. Although the
tissue is composed of cells with the same intrinsic properties, the electrotonic
effects generate anisotropic spatial variations (epicardial and transmural) of the
APD. The maximum of APD over the entire volume is consistently located at
the stimulation site. Anisotropic spatial variations of the APD along and across
fibers were observed experimentally in 2-D myocardial laminae by Osaka et al.
[364], Gotoh et al. [208] and on the epicardium of dog hearts by Burgess et al.
[67] and Taccardi et al. [512]. On the other hand, Bertran et al. [48], working with
pig hearts, did not observe anisotropic APD differences in 2-D laminae from the
302 9 Simulation Studies of Cardiac Bioelectrical Activity

ACTI REPO APD

42.39 119.91 3.00 289.05 366.29 3.00 244.41 250.90 1.00

21.78 129.70 3.00 273.50 377.24 3.00 246.13 251.72 1.00

0.12 145.71 3.00 258.52 389.76 3.00 258.52 389.76 3.00

144 388

258
93 343

253

45 298
248

0 259 244
0.12 145.71 3.00 258.52 389.76 3.00 244.05 258.42 1.00

Fig. 9.30 Orthotropic Monodomain–LR1 model, homogeneous truncated ellipsoidal domain,

endocardial central stimulation. Isochrone lines of the depolarization time (first column ACTI),
repolarization time (second column REPO) and action potential duration (third column APD) on
epicardial (first row), mid-myocardial (second row) and endocardial (third row) sections and on the
whole domain (fourth row). Reported below each panel are the maximum, minimum and step in
ms of the displayed map
9.3 Heterogeneous Cardiac Tissue 303

ACTI REPO APD

144
392
256

93 344
244

45 296 233

0 251 223

Fig. 9.31 Orthotropic Monodomain–LR1 model, W-wall with endocardial central stimulation.
Isochrone lines of the depolarization time (first column ACTI), repolarization time (second column
REPO) and action potential duration (third column APD) on epicardial (first row), mid-myocardial
(second row) and endocardial (third row) sections and on the whole domain (fourth row). Reported
below each panel are the maximum, minimum and step in ms of the displayed map
304 9 Simulation Studies of Cardiac Bioelectrical Activity

ACTI REPO APD

EPI

42.39 119.91 3.00 265.11 342.12 3.00 220.51 226.58 1.00

MID

21.78 129.69 3.00 284.34 386.87 3.00 247.51 263.09 1.00

ENDO

0.13 145.71 3.00 244.88 376.83 3.00 230.96 244.78 1.00

144 389 261

93 338 247

45 290 233

0.13 145.71 3.00 244.88 389.66 3.00 220.51 261.01 1.00

Fig. 9.32 Orthotropic Monodomain–LR1 model, 3-wall with endocardial central stimulation.
Isochrone lines of the depolarization time (first column ACTI), repolarization time (second column
REPO) and action potential duration (third column APD) on epicardial (first row), mid-myocardial
(second row) and endocardial (third row) sections and on the whole domain (fourth row). Reported
below each panel are the maximum, minimum and step in ms of the displayed map
9.3 Heterogeneous Cardiac Tissue 305

above 263.5
263.0 - 263.5

H-wall 3-wall W-wall 262.5 -

262.0 -
263.0
262.5
261.5 - 262.0
261.0 - 261.5
260.5 - 261.0
260.0 - 260.5
259.5 - 260.0
E 259.0 - 259.5
258.5 - 259.0
P 258.0 - 258.5
257.5 - 258.0
I 257.0 - 257.5
256.5 - 257.0
256.0 - 256.5
255.5 - 256.0
255.0 - 255.5
254.5 - 255.0
254.0 - 254.5
253.5 - 254.0
253.0 - 253.5
252.5 - 253.0
252.0 - 252.5

244.5 (0.5) 250.9 220.5 (0.5) 226.6 223.2 (0.5) 229.9 251.5 - 252.0
251.0 - 251.5
250.5 - 251.0
250.0 - 250.5
249.5 - 250.0
249.0 - 249.5
248.5 - 249.0
M 248.0 - 248.5
247.5 - 248.0
I 247.0 - 247.5
246.5 - 247.0
D 246.0 - 246.5
245.5 - 246.0
245.0 - 245.5
244.5 - 245.0
244.0 - 244.5
243.5 - 244.0
243.0 - 243.5
242.5 - 243.0
242.0 - 242.5
241.5 - 242.0
241.0 - 241.5

246.1 (0.5) 251.7 247.5 (0.5) 263.1 239.9 (0.5) 246.5 240.5 -
240.0 -
241.0
240.5
239.5 - 240.0
239.0 - 239.5

E 238.5 - 239.0
238.0 - 238.5
237.5 - 238.0
N 237.0 - 237.5
236.5 - 237.0
D 236.0 -
235.5 -
236.5
236.0
O 235.0 - 235.5
234.5 - 235.0
234.0 - 234.5
233.5 - 234.0
233.0 - 233.5
232.5 - 233.0
232.0 - 232.5
231.5 - 232.0
231.0 - 231.5
230.5 - 231.0
230.0 - 230.5
229.5 - 230.0
244.1 (0.5) 258.4 231.0 (0.5) 244.8 237.9 (0.5) 251.3 229.0 - 229.5
228.5 - 229.0
228.0 - 228.5
227.5 - 228.0
227.0 - 227.5
226.5 - 227.0
226.0 - 226.5
225.5 - 226.0
225.0 - 225.5
224.5 - 225.0
224.0 - 224.5
223.5 - 224.0
223.0 - 223.5
222.5 - 223.0
222.0 - 222.5
221.5 - 222.0
221.0 - 221.5
220.5 - 221.0
220.0 - 220.5
below 220.0

244.1 (0.5) 258.4 222.2 (2.0) 263.1 224.9 (2.0) 259.6

Fig. 9.33 Orthotropic Monodomain–LR1 model, homogeneous and heterogeneous truncated

ellipsoidal domains, endocardial central stimulation. Isochrone lines of the action potential
duration on epicardial (first row), mid-myocardial (second row) and endocardial (third row)
sections and on a transmural section (fourth row). Reported below each panel are the maximum,
minimum and step in ms of the displayed map
306 9 Simulation Studies of Cardiac Bioelectrical Activity

left ventricle. In vitro experimental studies have demonstrated the heterogeneity of

the action potential morphology and duration within the ventricular wall, see e.g.
[7, 572] and the survey [8]. The intrinsic APD heterogeneity in the ventricular wall
is not uniform, depending on the part of the ventricular wall where the myocytes
or the transmural wedges are isolated [7] and also on the artificial environment
of the preparation [533]. Therefore, in our simulations we have considered, in
addition to homogeneous intrinsic membrane properties, also two different types
of transmural heterogeneity, labeled 3-slab/wall and W-slab/wall in the previous
sections, obtained by subdividing the cardiac domain into layers parallel to the
epicardium, and assigning different cellular APD to each layer. In the 3-slab/wall,
we considered three cell layers (subendo-, mid- and subepi-cardial), as used before
in 1D simulation studies [96, 467, 546]. In the W-slab/wall, we incorporated four
cell layers, in order to approximate the APD transmural distributions detected in
experiments with wedge preparations (see [395, 396, 572]).
We have examined the effects of reversing the transmural activation from
epicardial to endocardial central stimulation. In the heterogeneous cases, the
distributions of both the transmural intrinsic APD and the transmural rotation of
the fiber direction are not symmetric with respect to the intramural midwall plane,
since the fiber direction rotates CCW from 45ı at the epicardium to 75ı at the
endocardium. Thus, the repolarization sequence and the APD distribution obtained
with an endocardial stimulation display some differences with respect to the same
features produced by an epicardial stimulation, and cannot be obtained from the
latter by simply exchanging epicardium with endocardium.
Firstly, we have extracted from our 3D simulations the 1D profiles of the
activation and repolarization times and of the APD elicited by a local epicardial or
endocardial stimulation. These simulated profiles are qualitatively similar to those
obtained by using homogeneous and heterogeneous 1D models of the ventricular
wall, see [96, 452, 467, 496, 546], but they do not coincide due to the presence of
anisotropic curvature effects. Only the case of a plane front obtained by stimulating
the whole epicardial or endocardial face of our cardiac slab could be reduced to a
1D propagation, but this is not our case.
In our simulations on heterogeneous slabs, when we switched the pacing site
from epicardium to endocardium, the dispersion of repolarization times along the
vertical line passing through the pacing site decreased from 46 to 33 ms for the
3-slab and from 47 to 14 ms for the W-slab (Table 9.5); this is consistent with
the simulations and measurement in Fish et al. [176]. Also, the profile of the
repolarization time displayed very different shapes in the two pacing protocols
(see Figs. 3 and 5). On the other hand, unlike Fish et al. [176] findings, the APD
dispersion in the two pacing protocols increased, when we moved from epi to
endo pacing while the APD profiles showed very similar shapes. These large
differences between the epicardial and endocardial stimulations are not present in
the full domain of the slab: for example, epicardial pacing for the W-slab yields
a total dispersion of 103 ms for the repolarization time and 31 ms for the APD,
while endocardial pacing for the same slab yields dispersions of 108 ms for the
repolarization time and 34 ms for the APD.
9.3 Heterogeneous Cardiac Tissue 307

Table 9.5 Dispersion (MAX–MIN value) of activation time (ACTI), repolarization time (REPO),
APD 90 on the 1-D transmural line passing through the stimulation site and orthogonal to both
epicardial and endocardial surfaces
Epicardial stimulation Endocardial stimulation
ACTI REPO APD 90 ACTI REPO APD 90
H-slab 35.26 24.62 10.53 35.48 23.90 11.45
3-slab 35.26 46.76 28.99 35.48 33.50 37.36
W-slab 35.26 46.79 19.91 35.48 14.74 31.78

In our heterogeneous models, the amount of assigned transmural intrinsic

heterogeneity yields a dispersion of APDs which is lower than that observed in
experimental wedge preparations, but is sufficient to produce a transmural change
of the repolarization sequence, compared with the homogeneous model. This
sequence is partially independent of the activation process and the lateral spread
of repolarization is driven by the transmural intrinsic heterogeneity of APDs. The
APD distribution on transmural sections exhibits a striped structure with elongated
equilevel lines, which are mostly flat and parallel to the epi- and endocardial
surfaces.
In sections parallel to the epi- and endocardial faces the repolarization sequences
and APD patterns are not affected by the transmural APD heterogeneity, except for
the absolute values of the two variables. More precisely, the distributions of the
repolarization times and APDs on sections parallel to the epicardium are strongly
correlated in the three different slabs/walls.
On intramural sections parallel to the epicardial surface, in the 3-slab and W-slab,
the quasi-elliptical pattern of repolarization isochrones is qualitatively similar to
that observed in the homogeneous slab, showing that the spread of repolarization in
these sections is primarily determined by the activation sequence and the rotational
anisotropy. This confirms that the repolarization sequence is not associated with a
propagating wavefront, but it is a phase wave.
Thus, in spite of the coupling effects due to diffusion, reaction and modulation,
the transmural heterogeneity does not influence the patterns on intramural sections
parallel to the epicardial face, but produces effects that remain confined in the
transmural direction. The finding that the effects of the heterogeneity and the
rotational anisotropy are so clearly decoupled in the heterogeneous slabs was
unexpected. Clearly, the former effects prevail in the transmural direction and the
latter in the sections parallel to the epicardial surface. This complex interactions
between orthotropic anisotropy, fiber rotation and cell heterogeneities determine
complex excitation and repolarization wavefronts that cannot be understood by
using simple 1D simulations.
We remark that in vivo experimental studies, see e.g. in intact canine [9,161,515]
or human [517] hearts, the observed APD dispersions are lower than the dispersions
observed in isolated cells and in wedge preparations. We refer also to [9, 70] for a
discussion about the difference between in vivo and in vitro measurements. Recent
experimental studies [515] in exposed and isolated dog hearts have shown that the
308 9 Simulation Studies of Cardiac Bioelectrical Activity

transmural dispersion of APD in the left ventricular wall is on the order of 15–20 ms
during ventricular pacing with cycle length of 350 or 400 ms. In these preparations,
the repolarization sequence is often qualitatively similar to the activation sequence.
When the pacing site was epicardial, both the excitation and the repolarization
fronts returned toward the epicardium in a transmural plane perpendicular to the
epicardial fiber direction near the pacing site. However, confirmatory studies are
needed in order to determine whether these findings occur consistently in varying
experimental conditions. Moreover, Taccardi et al. [512] observed a typical tripolar
epicardial APD pattern, with one central maximum and two minima in the cross-
fiber direction, on the left ventricular epicardium of exposed dog hearts during left
ventricular pacing. Poelzing et al. [395, 396] found that in a narrow subepicardial
layer the intercellular conductance is lower than in deeper layers, due to a local
shortage of gap junctions. This factor contributes to the transmural inhomogeneity
of excitation velocity and recovery durations.
As mentioned before, the data on transmural dispersion of APD currently
available in the literature are not consistent, since there is a disparity between the
transmural dispersion of the APD observed in in vivo hearts, see e.g. [9, 258], and
in vitro experiments in wedge preparations, see e.g. [8, 572]. The transmural APD
dispersion reported by the former authors is less than the one reported by the latter.
In this simulation study on the electrotonic modulation in a three dimensional
structure, we assumed that: (i) the heterogeneity of the intrinsic APDs is transmural,
i.e. the APDs values are constant in each section parallel to the epicardial face;
(ii) the fibers have a constant orientation in each section parallel to the epicardial
face; (iii) an idealized sheet-fiber model (iv) a normal cell coupling (i.e. normal
conductivity coefficients). The simulations show that the electrotonic modulation
is not strong enough to level the intrinsic transmural APD differences, but these
strong differences are not revealed at the epicardial level. Hence, our results do not
explain why in intact hearts several investigators, see e.g. [9, 258], have observed
a strongly reduced dispersion of APD across the ventricular wall as compared with
the dispersion observed in wedge preparations.
In conclusion, the complex 3-D paths of electrotonic currents during the repo-
larization phase in our orthotropic medium do not seem to extend the effects of
transmural inhomogeneity to the sections parallel to the epicardium. These results
suggest the presence of two separate and competing mechanisms that modulate the
repolarization sequence and APD patterns, one due to the rotational anisotropy and
the other to the transmural heterogeneity of intrinsic APD. While the first seems to
prevail in the sections parallel to the epicardium, the second is the major determinant
of the transmural shape of repolarization fronts and APD patterns outside a volume
surrounding the transmural line issuing from the stimulation site.
9.3 Heterogeneous Cardiac Tissue 309

9.3.3 Transmural and Apex-Base Heterogeneity in 3-D

Ellipsoids

We now extend the previous simulations in ellipsoidal cardiac domains to the case
of apex-to-base and mixed transmural APD heterogeneities; for more results and
details we refer to [130]. In particular, we consider the following heterogeneous
ellipsoidal domains:
Apex-base-wall. The intrinsic APDs of cells vary in this case with respect to the
-direction. We define the intrinsic APD along the apex-to-base direction decreasing
monotonically from apex to base, by introducing a scaling factor factIK of the IK
current depending linearly on  according to the following rule

2:88  1:952
factIAB ./ D .  min / C 1:952;
K
max  min

for  2 Œmin ; max . The resulting dependence of APD on  is reported in Fig. 9.29,
showing an almost linear variation from 272 ms at the apex to 225 ms at the base,
with an intrinsic APD dispersion of 47 ms.
Mixed-wall. Both transmural 3-wall and apex-to-base-wall heterogeneities are
introduced in the cardiac tissue by setting factIK D fact3IK C factAB
IK ./, i.e.

8
< 2:62 C factAB ./  2 Œmin ; max ; r 2 Œ0; 0:33/ (endo)
factIK D 1:952 C factAB ./  2 Œmin ; max ; r 2 Œ0:33; 0:66/ (mid)
:
2:88 C factAB ./  2 Œmin ; max ; r 2 Œ0:66; 1 (epi):

We investigate the interaction between the two intrinsic APD heterogeneities

considered, one in the transmural direction (3-wall) and the other in the intramural
apex-to-base direction (apex-base-wall). Unexpectedly, the repolarization and APD
patterns on r-layers of the apex-base-wall (Fig. 9.34) and the mixed-wall (Fig. 9.35)
are very similar apart from their numerical values. On the other hand, the patterns
displayed on the transmural diagonal section are very dissimilar due to the
transmural heterogeneity in the mixed-wall. As in the case without apex-to-base
heterogeneity (Fig. 9.36), the transmural heterogeneity has only a weak influence on
the repolarization patterns on the r-layers, apart from a shift toward the ventricular
base of the site that repolarize first and an increase of the spatial dispersion.
In order to quantitatively study the influence of the two heterogeneities, we
computed the correlation coefficient (CC) on the intramural r-layers and transmural
- and -layers, for the pair 3-wall and mixed-wall, and for the pair apex-base-
wall and mixed-wall. The average CCs on each layer are reported in Table 9.6.
The results confirm that the two heterogeneities are somewhat decoupled, i.e. (a)
the high CC values on -layers show that the repolarization and APD patterns
are independent of the apex-to-base heterogeneity and mostly determined by
the transmural heterogeneity; (b) the high CC values on r-layers show that the
repolarization and APD patterns are independent of the transmural heterogeneity
310 9 Simulation Studies of Cardiac Bioelectrical Activity

ACTI REPO APD

E 395

P 390 - 395
385 - 390
380 - 385
I 375 - 380
370 - 375
365 - 370
360 - 365
355 - 360
350 - 355
345 - 350
340 - 345
335 - 340
330 - 335
325 - 330
320 - 325
42.6 ( 5.0) 119.9 315 - 320 279.9 (5.0) 371.7 220.5 ( 1.0) 266.4
310 - 315
305 - 310
300 - 305
295 - 300 277
290 - 295 276 - 277
285 - 290 275 - 276
M 280 - 285
275 - 280
274 - 275
273 - 274
I 270 - 275
265 - 270
272 - 273
271 - 272

D 260 - 265
255 - 260
270 - 271
269 - 270
250 - 255 268 - 269
245 - 250 267 - 268
240 - 245 266 - 267
235 - 240 265 - 266
230 - 235 264 - 265
225 - 230 263 - 264
220 - 225 262 - 263
215 - 220 261 - 262
210 - 215 260 - 261
205 - 210 259 - 260

21.9 ( 5.0) 130.0 200 - 205

195 - 200
264.1 (5.0) 373.1 222.8 ( 1.0) 271.5 258 - 259
257 - 258
190 - 195 256 - 257
185 - 190 255 - 256
180 - 185 254 - 255
E 175 - 180
170 - 175
253 - 254
252 - 253
N 165 - 170
160 - 165
251 - 252
250 - 251

D 155 - 160
150 - 155
249 - 250
248 - 249

O 145 - 150
140 - 145
247 - 248
246 - 247
135 - 140 245 - 246
130 - 135 244 - 245
125 - 130 243 - 244
120 - 125 242 - 243
115 - 120 241 - 242
110 - 115 240 - 241
105 - 110 239 - 240
100 - 105 238 - 239
95 - 100 237 - 238
90 - 95 236 - 237
85 - 90 235 - 236
0.2 ( 5.0) 146.0 80 - 85 249.2 (5.0) 378.2 219.7 ( 1.0) 268.5 234 - 235
75 - 80 233 - 234
70 - 75 232 - 233
65 - 70 231 - 232
60 - 65 230 - 231
55 - 60 229 - 230
50 - 55 228 - 229
45 - 50 227 - 228
40 - 45 226 - 227
35 - 40 225 - 226
30 - 35 224 - 225
25 - 30 223 - 224
20 - 25 222 - 223
15 - 20 221 - 222
10 - 15 220 - 221
5 - 10 219 - 220
0- 5 218 - 219
0 218

0.2 ( 5.0) 146.0 249.3 (5.0) 378.2 219.7 ( 1.0) 268.6

Fig. 9.34 Apex-base-wall with central endocardial stimulation. Same format as Fig. 9.36 (Repro-
duced with permission from [130])
9.3 Heterogeneous Cardiac Tissue 311

ACTI REPO APD

E 400 279

P 395 - 400
390 - 395
278 - 279
277 - 278
385 - 390 276 - 277
I 380 - 385
375 - 380
275 - 276
274 - 275
370 - 375 273 - 274
365 - 370 272 - 273
360 - 365 271 - 272
355 - 360 270 - 271
350 - 355 269 - 270
345 - 350 268 - 269
340 - 345 267 - 268
335 - 340 266 - 267
330 - 335 265 - 266
325 - 330 264 - 265
42.6 ( 5.0) 119.9 320 - 325 264.5 (5.0) 349.9 205.2 ( 1.0) 244.7 263 - 264
315 - 320 262 - 263
310 - 315 261 - 262
305 - 310 260 - 261
300 - 305 259 - 260
295 - 300 258 - 259
290 - 295 257 - 258
M 285 - 290
280 - 285
256 - 257
255 - 256
I 275 - 280
270 - 275
254 - 255
253 - 254

D 265 - 270
260 - 265
252 - 253
251 - 252
255 - 260 250 - 251
250 - 255 249 - 250
245 - 250 248 - 249
240 - 245 247 - 248
235 - 240 246 - 247
230 - 235 245 - 246
225 - 230 244 - 245
220 - 225 243 - 244
215 - 220 242 - 243
210 - 215 241 - 242

21.9 ( 5.0) 130.0 205 - 210

200 - 205
283.0 (5.0) 381.5 235.1 ( 1.0) 280.1 240 - 241
239 - 240
195 - 200 238 - 239
190 - 195 237 - 238
185 - 190 236 - 237
E 180 - 185
175 - 180
235 - 236
234 - 235
N 170 - 175
165 - 170
233 - 234
232 - 233

D 160 - 165
155 - 160
231 - 232
230 - 231

O 150 - 155
145 - 150
229 - 230
228 - 229
140 - 145 227 - 228
135 - 140 226 - 227
130 - 135 225 - 226
125 - 130 224 - 225
120 - 125 223 - 224
115 - 120 222 - 223
110 - 115 221 - 222
105 - 110 220 - 221
100 - 105 219 - 220
95 - 100 218 - 219
90 - 95 217 - 218
0.2 ( 5.0) 146.0 85 - 90 244.0 (5.0) 372.7 213.1 ( 1.0) 261.1 216 - 217
80 - 85 215 - 216
75 - 80 214 - 215
70 - 75 213 - 214
65 - 70 212 - 213
60 - 65 211 - 212
55 - 60 210 - 211
50 - 55 209 - 210
45 - 50 208 - 209
40 - 45 207 - 208
35 - 40 206 - 207
30 - 35 205 - 206
25 - 30 204 - 205
20 - 25 203 - 204
15 - 20 202 - 203
10 - 15 201 - 202
5 - 10 200 - 201
5 200

0.2 ( 5.0) 146.0 244.1 (5.0) 384.0 206.7 ( 1.0) 278.2

Fig. 9.35 Mixed-wall with central endocardial stimulation. Same format as Fig. 9.36 (Reproduced
with permission from [130])
312 9 Simulation Studies of Cardiac Bioelectrical Activity

ACTI REPO APD

E 395 280

P 390 - 395
385 - 390
279 - 280
278 - 279
380 - 385 278 - 278
I 375 - 380
370 - 375
278 - 278
277 - 278
365 - 370 276 - 277
360 - 365 276 - 276
355 - 360 276 - 276
350 - 355 275 - 276
345 - 350 274 - 275
340 - 345 274 - 274
335 - 340 274 - 274
330 - 335 273 - 274
325 - 330 272 - 273
320 - 325 272 - 272
42.4 ( 5.0) 119.9 315 - 320 289.1 (5.0) 366.2 244.5 ( 0.5) 250.9 272 - 272
310 - 315 271 - 272
305 - 310 270 - 271
300 - 305 270 - 270
295 - 300 270 - 270
290 - 295 269 - 270
285 - 290 268 - 269
M 280 - 285
275 - 280
268 - 268
268 - 268
I 270 - 275
265 - 270
267 - 268
266 - 267

D 260 - 265
255 - 260
266 - 266
266 - 266
250 - 255 265 - 266
245 - 250 264 - 265
240 - 245 264 - 264
235 - 240 264 - 264
230 - 235 263 - 264
225 - 230 262 - 263
220 - 225 262 - 262
215 - 220 262 - 262
210 - 215 261 - 262
205 - 210 260 - 261

21.8 ( 5.0) 129.7 200 - 205

195 - 200
273.5 (5.0) 377.2 246.1 ( 0.5) 251.7 260 - 260
260 - 260
190 - 195 259 - 260
185 - 190 258 - 259
180 - 185 258 - 258
E 175 - 180
170 - 175
258 - 258
257 - 258
N 165 - 170
160 - 165
256 - 257
256 - 256

D 155 - 160
150 - 155
256 - 256
255 - 256

O 145 - 150
140 - 145
254 - 255
254 - 254
135 - 140 254 - 254
130 - 135 253 - 254
125 - 130 252 - 253
120 - 125 252 - 252
115 - 120 252 - 252
110 - 115 251 - 252
105 - 110 250 - 251
100 - 105 250 - 250
95 - 100 250 - 250
90 - 95 249 - 250
85 - 90 248 - 249
0.2 ( 5.0) 145.7 80 - 85 258.5 (5.0) 389.8 244.1 ( 0.5) 258.4 248 - 248
75 - 80 248 - 248
70 - 75 247 - 248
65 - 70 246 - 247
60 - 65 246 - 246
55 - 60 246 - 246
50 - 55 245 - 246
45 - 50 244 - 245
40 - 45 244 - 244
35 - 40 244 - 244
30 - 35 243 - 244
25 - 30 242 - 243
20 - 25 242 - 242
15 - 20 242 - 242
10 - 15 241 - 242
5 - 10 240 - 241
0- 5 240 - 240
0 240

0.2 ( 5.0) 145.7 258.5 (5.0) 389.8 244.1 ( 0.5) 258.4

Fig. 9.36 H-wall with central endocardial stimulation. Activation (ACTI, first column), repolar-
ization (REPO, second column), APD (third column) patterns on the epicardium (EPI, first row),
midwall (MID, second row), endocardium (ENDO, third row), diagonal section (fourth row). Below
each panel are reported the minimum, stepsize (in brackets), and maximum value of the displayed
map (Reproduced with permission from [130])
9.3 Heterogeneous Cardiac Tissue 313

Table 9.6 Average correlation coefficients of repolarization times and APD on r, ;  -layers
Endo stim. Epi stim.
Average CC on Average CC on
r-layer -layer  -layer r-layer -layer  -layer
H-wall,3-wall REPO 0.997 0.838 0.841 0.998 0.842 0.565
APD 0.590 0.460 0.442 0.699 0.386 0.366
H-wall,W-wall REPO 0.999 0.913 0.911 0.999 0.914 0.657
APD 0.830 0.273 0.275 0.873 0.373 0.376
3-wall,mixed-wall REPO 0.853 0.890 0.997 0.882 0.887 0.997
APD 0.146 0.673 0.999 0.150 0.668 0.999
Apex-base-wall, REPO 0.998 0.854 0.829 0.999 0.904 0.562
mixed-wall APD 0.993 0.745 0.455 0.992 0.739 0.375

and mostly determined by the apex-to-base heterogeneity. We remark that due

to the transmural fiber rotation and the presence of the fiber imbrication angle,
the complex three-dimensional anisotropic current density vectors cross each r
and layers transversally. Nevertheless, the repolarization and APD patterns
on layers unmask the transmural heterogeneity and musk the apex-to-base
heterogeneity and opposite holds for the patterns on the r-layers.
We also consider a case with multiple stimulations in order to mimic the
presence of a few Purkinje ventricular junctions. More precisely, we consider on
the endocardium three stimulation sites near the apex, with stimuli delivered at
0, 4, 8 ms, see e.g. Fig. 9.37 for the apex-base-wall results. Similar results and
observations described above apply to this strongly different case with multiple
activation and repolarization wavefronts.
Our simulations have shown that in cardiac walls with constant curvature, such
as a slab or a spherical wall geometry, the repolarization sequence and APD patterns
on the wall surfaces parallel to the epicardium (r-layers) do not reveal the type of
underlying transmural APD heterogeneity.
In walls with variable curvature and tapering, the repolarization sequences on
all intramural r-layers (apart from their numerical values) again appear to be
independent of the underlying transmural heterogeneity, while only the epicardial
and endocardial APD patterns appear to be independent of the APD transmural
heterogeneity.
As a corollary, the epicardial repolarization and APD patterns are dominated
by the influence of the apex-to-base heterogeneity, whereas the same patterns on
the transmural layers are dominated by the transmural heterogeneity. Hence, the
epicardial features of the repolarization and APD patterns do not allow us to make
inferences about the variation of the underlying transmural heterogeneities, while
they allow us to detect the presence of apex-to-base heterogeneity. These findings
are independent of the stimulus location (epicardial, endocardial) and of Purkinje
involvement.
314 9 Simulation Studies of Cardiac Bioelectrical Activity

ACTI REPO APD

E 395 279

P 390 - 395
385 - 390
278 - 279
277 - 278
380 - 385 276 - 277
I 375 - 380
370 - 375
275 - 276
274 - 275
365 - 370 273 - 274
360 - 365 272 - 273
355 - 360 271 - 272
350 - 355 270 - 271
345 - 350 269 - 270
340 - 345 268 - 269
335 - 340 267 - 268
330 - 335 266 - 267
325 - 330 265 - 266
320 - 325 264 - 265
35.9 ( 5.0) 159.1 315 - 320 294.5 (5.0) 385.8 224.9 ( 1.0) 274.0 263 - 264
310 - 315 262 - 263
305 - 310 261 - 262
300 - 305 260 - 261
295 - 300 259 - 260
290 - 295 258 - 259
285 - 290 257 - 258
M 280 - 285
275 - 280
256 - 257
255 - 256
I 270 - 275
265 - 270
254 - 255
253 - 254

D 260 - 265
255 - 260
252 - 253
251 - 252
250 - 255 250 - 251
245 - 250 249 - 250
240 - 245 248 - 249
235 - 240 247 - 248
230 - 235 246 - 247
225 - 230 245 - 246
220 - 225 244 - 245
215 - 220 243 - 244
210 - 215 242 - 243
205 - 210 241 - 242

18.7 ( 5.0) 165.2 200 - 205

195 - 200
281.7 (5.0) 390.0 224.9 ( 1.0) 278.9 240 - 241
239 - 240
190 - 195 238 - 239
185 - 190 237 - 238
180 - 185 236 - 237
E 175 - 180
170 - 175
235 - 236
234 - 235
N 165 - 170
160 - 165
233 - 234
232 - 233

D 155 - 160
150 - 155
231 - 232
230 - 231

O 145 - 150
140 - 145
229 - 230
228 - 229
135 - 140 227 - 228
130 - 135 226 - 227
125 - 130 225 - 226
120 - 125 224 - 225
115 - 120 223 - 224
110 - 115 222 - 223
105 - 110 221 - 222
100 - 105 220 - 221
95 - 100 219 - 220
90 - 95 218 - 219
85 - 90 217 - 218
0.1 ( 5.0) 161.9 80 - 85 270.0 (5.0) 388.3 226.4 ( 1.0) 278.3 216 - 217
75 - 80 215 - 216
70 - 75 214 - 215
65 - 70 213 - 214
60 - 65 212 - 213
55 - 60 211 - 212
50 - 55 210 - 211
45 - 50 209 - 210
40 - 45 208 - 209
35 - 40 207 - 208
30 - 35 206 - 207
25 - 30 205 - 206
20 - 25 204 - 205
15 - 20 203 - 204
10 - 15 202 - 203
5 - 10 201 - 202
0- 5 200 - 201
0 200

0.1 ( 5.0) 166.2 270.0 (5.0) 390.9 224.6 ( 1.0) 278.3

Fig. 9.37 Apex-base-wall, Purkinje stimulation. Same format as Fig. 9.36 (Reproduced with
permission from [130])
9.4 QRS Complex and T Wave Morphology in Electrograms 315

Our main finding is that for cardiac tissue with normal cell coupling the
anisotropic diffusion current (electrotonic modulation) is not sufficient to mask the
intrinsic repolarization differences (both transmural and apex-to-base) between each
myocyte and its neighboring cells. Thus, in our orthotropic ellipsoidal wall, the
complex 3D electrotonic modulation of APDs does not fully mix the effects of
the transmural and apex-to-base heterogeneity. The intrinsic spatial heterogeneity
of the APDs is unmasked in the modulated APD patterns only in the appropriate
transmural or intramural sections.

9.4 QRS Complex and T Wave Morphology in Electrograms

The factors determining the shape and polarity of the T wave in electrocardiograms
(ECGs) are still a matter of debate. According to the classical theory based on
the double layer source model (see e.g. [538] and the classical references therein),
an activation sequence starting from the endocardium and proceeding towards the
epicardium generates ECG waveforms at leads facing the epicardium with an R
wave. By applying the heart surface model (see e.g. [483, 538]), it is generally
believed that if the repolarization sequence follows the activation one, then a
negative T wave is expected, due to opposite direction of the transmembrane
potential gradients during the downstroke phase of the action potential. On the
contrary, measured human Einthoven leads II exhibit QRS complex and T wave with
the same positive polarity. In order to explain this discrepancy, a different transmural
repolarization sequence within the ventricular wall has been proposed, where the
epicardium recovers before the endocardium based on the assumption that the action
potential duration (APD) of epicardial cells is considerably shorter than the APD of
endocardial cells. Experimental evidence of transmural APD heterogeneity and of
transmural repolarization gradients has been established in [571] for isolated cells
and in vitro wedge preparations of canine left ventricles. Computational studies
on isotropic [95, 148, 176, 205] or anisotropic cardiac tissue [58], with transmural
APD cell heterogeneity, have been able to reproduce concordant R and T waves.
Nevertheless, a repolarization sequence directed from epicardium to endocardium
has never been observed in in vivo animal models; see [133, 183, 258, 517]. In
alternative to transmural repolarization gradients, apico-basal APD heterogeneities
have been proposed as a determinant of the T wave polarity. The aim of this section
is to provide some insight on how the tissue anisotropy, cellular APD heterogeneities
and shape of excitation wavefronts influence the polarity of the T wave in unipolar
and bipolar ECGs. The study is based on three-dimensional simulations of the
entire depolarization and repolarization phases of the propagating action potential,
using the insulated anisotropic Monodomain model in a slab of cardiac tissue. Then
the primary source is used to compute the extracardiac potential in an unbounded
homogeneous conducting medium.
316 9 Simulation Studies of Cardiac Bioelectrical Activity

9.4.1 Methods and Parameter Calibration

The anisotropic Monodomain model. The evolution of the transmembrane poten-

tial v.x; t/, extracellular/extracardiac potential u.x; t/, gating variables w.x; t/ and
ionic concentrations c.x; t/, is described by the reaction-diffusion system
8
ˆ
ˆ cm @t v  div.Dm rv/ C iion .v; w; c/ D iapp in ˝H ; nT Dm rv D 0 on @˝H
<
@t w  R.v; w/ D 0; @t c  S.v; w; c/ D 0 in ˝H
 
ˆ div Di rv in ˝H Di CDe in ˝H
:̂  div Dru D where DD
0 in R3 n ˝H 0 I in R3 n ˝H
(9.3)
with appropriate initial conditions. Here ˝H is the cardiac volume, cm and iion
denote the capacitance and the ionic current of the membrane per unit volume, iapp
represents the applied current per unit volume, and 0 is the conductivity coefficient
of the extracardiac medium.
The conductivity tensors Di .x/ and De .x/ at any point x 2 ˝H are defined as
in (3.41). The Monodomain conductivity tensor Dm is given by Dm D De .Di C
De /1 Di , see Chap. 4.
The fibers rotate intramurally linearly with the depth for a total amount of
120ı and when the point of view is from the epicardial side, the rotation is
counterclockwise proceeding from the epicardium to endocardium. More precisely,
in the canonic reference system .e1 ; e2 ; e3 /, the fiber direction al .x/ and the other
two principal axes at a point x are given by

al .x/Due1 cos ˛.r/Cue2 sin ˛.r/; at .x/Due3 ; an .x/Due1 sin ˛.r/  ue 2 cos ˛.r/
˛.r/D 23 .1  r/ 4 ; with r 2 Œ0; 1:

The conductivity tensors are assumed orthotropic with conductivity coefficients

along the principal axis given in the Table 9.7 (right) below.

Table 9.7 Parameter calibration for modeling the transmural heterogeneities (left) and conductiv-
ity coefficients in mS cm1 (right)
Slab type H-slab 3-slab Conductivity coefficients
Number of layers 1 3 Intra Extra
Layer thickness (cm) Endo Mid Epi l 3 2
1 0.33 0.34 0.33 t 0:3152 1:3514
fact_IK 1 2.62 1.95 2.88 n 0:0315 0:6757
APD (ms) 266 235 272 225 0 D 6
9.4 QRS Complex and T Wave Morphology in Electrograms 317

Transmural heterogeneity. The membrane model employed is the Luo-Rudy

phase I (LR1) ionic membrane model (see [308]), requiring six gating variables and
one calcium concentration variable. We consider two different types of transmural
distribution of the intrinsic APDs of the cells, one homogeneous (H-slab) and one
heterogeneous (3-slab). The transmural intrinsic APD heterogeneity is assumed
to be the same along any transmural epi-endocardial straight line, i.e. in any
plane parallel to the epicardium all cells have the same intrinsic APD. In the
heterogeneous slab, the intrinsic APD of the cells is obtained by multiplying the
potassium current IK in the LR1 model by a factor factIK . This modulation factor
is chosen in order to introduce a transmural APD profile with M-cell layers; see
Table 9.7, Fig. 9.38-right and [126, 205] for more details.
Computation of unipolar and bipolar electrograms. Unipolar ECGs u are
computed at an array of 3  3 points on planes located 2 cm (1 cm, not shown)
above the epicardium (plane L1 ) and 2 cm (1 cm, not shown) below the endocardium
(plane L2 ), see Fig. 9.38-left. In order to investigate the qualitative behavior of the
T wave of ECGs for sites at some distance from the active tissue, we disregard
the anisotropic conduction effects of the bulk tissue, but we take into account the
anisotropic cardiac source structure. Thus, the ECGs far away from the tissue slab
are computed by assuming Di C De  0 I , hence according to the formula (see
e.g. [117, 119, 295, 483])
Z  
1 1
u.x; t/ D  Di .y/rv.y; t/  r d y; (9.4)
40 ˝H jy  xj

where Di .y/rv.y; t/ represents the anisotropic cardiac electric sources generating

the extracardiac potential field.

PLANE L1
50
TRANSMEMBRANE POTENTIAL (mV)

2 cm
0

EPI
M
1 cm

−50
5 cm ENDO EPI
2 cm

ENDO

−100
PLANE L 0 50 100 150 200 250 300
2
TIME (ms)

Fig. 9.38 Left: cardiac slab and location of the 3  3 transmural ECG electrode array 2 cm above
the epicardial surface (plane L1 ) and 2 cm below the endocardial surface (plane L2 ). Right:
Epicardial, Midmyocardial and Endocardial action potentials in the transmural heterogeneous
3-slab calibration
318 9 Simulation Studies of Cardiac Bioelectrical Activity

Decomposition of the ECG waveform. In order to study the contribution of

the anisotropic cardiac sources to the T wave morphology, we introduce an additive
splitting of the intracellular conductivity tensor Di D DHS C DAS given by

DHS D ti I and DAS D .li  ti /al .x/aTl .x/ C .ni  ti /an .x/aTn .x/:

This splitting of Di induces an analogous splitting of the cardiac sources into the
so called double layer or heart surface (HS) source model (see [483, 538]) and into
a residual anisotropic source (AS). Consequently, the source term is decomposed
as Di rv D DHS rv C DAS rv and since u depends linearly on Di rv in Eq. (9.4),
the full unipolar ECG is decomposed as the sum of an isotropic (heart surface)
and anisotropic components u D uHS C uAS . Given two points xepi 2 cm above
the epicardium (plane L1 ) and xendo 2 cm below the endocardium (plane L2 ), we
introduce also the transmural bipolar ECG given by .xepi ; xendo ; t/ D u.xepi ; t/
u.xendo ; t/: Using the same splitting as before, we also decompose the full bipolar
ECG into the HS component HS .xepi ; xendo ; t/ and the anisotropic component
AS .xepi ; xendo ; t/.
Numerical methods. The cardiac domain ˝H considered is a cartesian slab of
dimensions 5  5  1 cm3 modeling a portion of the left ventricle; see Fig. 9.38-left.
In all computations, a structured grid of 500  500  100 hexahedral isoparametric Q1
elements of size h D 0:1 mm is used in space. The time discretization is based on
an Euler Imex method.

9.4.2 Unipolar and Bipolar ECG Simulations

We consider two endocardial stimulation protocols: a single (ectopic) site stimula-

tion (S1 ), eliciting a twisted semi-ellipsoidal excitation front, or multiple (12) sites
stimulation (S2 ), modeling a simplified Purkinje network, eliciting a large excitation
front spreading transmurally mainly across fiber. These two protocols are applied to
both H-slab and 3-slab, for a total of four simulations.
S1 : unipolar ECGs. The activation (ACTI) and repolarization (REPO90)
isochrones on a transmural section of the tissue are reported in Fig. 9.39a. The
HS components of all unipolar ECGs on plane L2 (2 cm below the endocardium)
for H-slab (see Fig. 9.40a) display a Q wave and a positive T wave, as expected. The
same behavior is also observed in 3-slab case (not shown), where the intramural
repolarization sequence displays a primary and a secondary fronts coming from
the endocardial and epicardial surfaces, respectively. Instead, the HS components
on plane L1 (2 cm above the epicardium) display a QRS complex with R and S
waves and, unexpectedly for a tissue without transmural heterogeneities, a positive
T wave, see Fig. 9.41a, c. The AS components of unipolar ECGs on both planes L1
and L2 display a Q wave and a positive T wave. The same morphology is shared
by the full unipolar ECGs on both planes L1 and L2 (Figs. 9.40a and 9.41a, c).
9.4 QRS Complex and T Wave Morphology in Electrograms 319

S1 S2
a TRANSMURAL SECTION b TRANSMURAL SECTION
80 60
ACTI

ACTI
60
40
40
20 20

0.11 85.62 4.00 0.11 70.88 4.00

H−slab H−slab
REPO90

REPO90
300 300

250 250
258.50 330.79 4.00 258.24 315.85 4.00

3−slab 3−slab
REPO90

REPO90
300 300

250 250
244.66 326.50 4.00 244.49 312.16 4.00

Fig. 9.39 (a): Single site stimulation (S1 ), isochrones of activation time ACTI (first row) and
repolarization time REPO90 (second and third row) on a transmural diagonal section of H-slab,
3-slab. (b): Multiple sites stimulation (S2 ), same format as in (a)

a S1, zoom at site 5 b S2 , zoom at site 5

0.5 0.5

0 0
mV

mV

−0.5 −0.5

−1 −1

−1.5 −1.5
0 100 200 300 400 0 100 200 300 400
ms ms

Fig. 9.40 H-slab: Unipolar ECGs at site 5 on plane L2 (site 5 is the central one in the 3  3 array,
see Fig. 9.41). (a): Single site stimulation (S1 ). (b): Multiple sites stimulation (S2 ). Continuous
line: full unipolar ECG. Dashed line: HS component. Dashed-Dotted line: AS component

This fact indicates that the full unipolar ECGs are dominated by their anisotropic
AS components. The HS components have little influence on the full ECGs, since
their positive T waves reinforce those of the full ECGs but their R waves at almost
all sites on plane L1 are absent in the QRS complex of the full ECGs. Analogous
morphological considerations hold for the 3-slab (Fig. 9.42), although with slight
differences in magnitude.
S2 : unipolar ECGs. The isochrones of ACTI and REPO90 on a transmural
section of the tissue are reported in Fig. 9.39b. The excitation wavefronts are larger
than those elicited by the ectopic stimulation. On plane L2 , the morphology of the
unipolar ECGs is the same as in the ectopic case described before; see Fig. 9.40b. On
plane L1 , all the AS components have Q waves and positive T waves, while the HS
320 9 Simulation Studies of Cardiac Bioelectrical Activity

a S1 b S2
7 8 9 7 8 9
0.5 0.5
mV

mV
−1 −0.5

4 5 6 4 5 6
0.5 0.5
mV

mV
−1 −0.5

1 2 3 1 2 3
0.5 0.5
mV

mV
−1 −0.5
0 400 0 400 0 400 0 400 0 400 0 400
ms ms ms ms ms ms

c S1 , zoom at site 5 d S2 , zoom at site 5

0.5 0.5

0 0
mV

mV

−0.5 −0.5

−1 −1

−1.5 −1.5
0 100 200 300 400 0 100 200 300 400
ms ms

Fig. 9.41 H-slab: Unipolar ECGs on plane L1 . Continuous line: full unipolar ECG. Dashed line:
HS component. Dashed-Dotted line: AS component (only in zoom). (a): Single site stimulation
(S1 ). (b): Multiple sites stimulation (S2 ). (c, d): zoom at site 5

components exhibit R waves and negative (H-slab, Fig. 9.41b, d) or positive (3-slab,
Fig. 9.42b, d) T waves. Nevertheless, the larger magnitude of the AS components
prevail in the full unipolar ECGs and they still exhibit a positive T wave and QRS
complex entirely negative or with RS configuration (small R and large S wave) and
positive T wave. Therefore, the full unipolar ECGs are still dominated by their AS
components.
S1 : bipolar ECGs. Unlike unipolar waveforms, the bipolar ECGs (Figs. 9.43
and 9.44a, c) appear to be dominated by their HS components, all with R and S
waves and biphasic (for H-slab) or positive (for 3-slab) T waves. Since the AS
components of the unipolar ECGs on planes L1 and L2 are almost identical, the
bipolar AS component has a very small magnitude in comparison with the bipolar
HS component, hence they have a negligible influence on the full bipolar ECGs.
The negative minimum of the biphasic T wave appearing in the bipolar ECGs in the
H-slab case is due to the T peak of the unipolar HS components at sites on plane
L2 preceding the T peak of the symmetric sites on plane L1 .
9.4 QRS Complex and T Wave Morphology in Electrograms 321

S1 S2
a 7 8 9
b 7 8 9
0.5 0.5
mV

mV
−1 −0.5

4 5 6 4 5 6
0.5 0.5
mV

mV
−1 −0.5

1 2 3 1 2 3
0.5 0.5
mV

mV
−1 −0.5
0 400 0 400 0 400 0 400 0 400 0 400
ms ms ms ms ms ms

c S1 , zoom at site 5 d S2 , zoom at site 5

0.5 0.5

0 0
mV

mV

−0.5 −0.5

−1 −1

−1.5 −1.5
0 100 200 300 400 0 100 200 300 400
ms ms

Fig. 9.42 3-slab: Unipolar ECGs on plane L1 . Same format as Fig. 9.41. (a) S1 . (b) S2 . (c) S1 ,
zoom at site 5. (d) S2 , zoom at site 5

S2 : bipolar ECGs. Bipolar ECGs now display a QRS complex with a large
R wave and negative T wave (Figs. 9.43 and 9.44b, d), similarly to their HS
components. The bipolar AS components have little influence because of their much
smaller magnitude and therefore, as in the ectopic case, the full bipolar ECGs are
dominated by their HS components.
Discussion of the results. By means of three-dimensional simulations based
on the anisotropic Monodomain model, we have studied the influence of tissue
anisotropy, cellular APD heterogeneities and the shape of the excitation wavefront
on the T wave polarity. The simulation results have shown that: (i) unipolar ECGs
exhibit a positive T wave mainly determined by the anisotropy of the cardiac
tissue, irrespective of cellular APD heterogeneities and the shape of the excitation
wavefront; (ii) on the other hand, bipolar ECGs are mainly determined by their
isotropic component and their T wave turns out to be positive only for a single
stimulation in presence of transmural APD heterogeneity, while it becomes always
negative in case of multiple stimulation generating large activation wavefronts,
regardless of the considered cellular APD heterogeneities. We remark that in our
ECG computation, we have used an idealized cardiac geometry taking into account
322 9 Simulation Studies of Cardiac Bioelectrical Activity

S1 S2
a 7 8 9 b 7 8 9
0.5 1.5
mV

mV
−0.5 −0.5

4 5 6 4 5 6
0.5 1.5
mV

mV
−0.5 −0.5

1 2 3 1 2 3
0.5 1.5
mV

mV
−0.5 −0.5
0 400 0 400 0 400 0 400 0 400 0 400
ms ms ms ms ms ms

S1 , zoom at site 5 S2 , zoom at site 5

c 1.5
d 1.5

1 1
mV

mV

0.5 0.5

0 0

−0.5 −0.5
0 100 200 300 400 0 100 200 300 400
ms ms

Fig. 9.43 H-slab: Bipolar ECGs on plane L1 . Continuous line: full unipolar ECG. Dashed line:
HS component. Dashed-Dotted line: AS component (only in zoom). (a): Single site stimulation
(S1 ). (b): Multiple sites stimulation (S2 ). (c, d): zoom at site 5

anisotropic cardiac sources but considering the widely used assumptions of the same
isotropic properties for bulk tissue and for the surrounding infinite extracardiac
medium. Previous studies [117, 119] have shown that including anisotropic bulk
tissue affects the ECGs magnitude far away from the cardiac surface but it does not
alter their QRS morphology. The extension of our conclusions on T wave polarity
to real 12-lead ECGs will require considering realistic geometries for both cardiac
tissue and torso, as well as the anisotropy of the bulk tissue. Nevertheless, the real
12-lead ECGs are far-field signals and we expect that the full ECG and the isotropic/
anisotropic ECG components will be smoothed out due to the greater distance from
the active tissue.
9.5 Extracellular Markers of Excitation and Repolarization Times 323

a S1 b S2
7 8 9 7 8 9
0.5 1.5
mV

mV
−0.5 −0.5

4 5 6 4 5 6
0.5 1.5
mV

mV
−0.5 −0.5

1 2 3 1 2 3
0.5 1.5
mV

mV
−0.5 −0.5
0 400 0 400 0 400 0 400 0 400 0 400
ms ms ms ms ms ms

c S1 , zoom at site 5 d S2 , zoom at site 5

1.5 1.5

1 1
mV

mV

0.5 0.5

0 0

−0.5 −0.5
0 100 200 300 400 0 100 200 300 400
ms ms

Fig. 9.44 3-slab: Bipolar ECGs. Same format as Fig. 9.43

9.5 Extracellular Markers of Excitation and Repolarization

Times

The time evolution of the cardiac bioelectric activity is described by the time
and spatial distribution of the intracellular and extracellular potential fields
ui .x; t/; ue .x; t/, both dependent on a chosen reference potential. By taking the
difference between the intra- and extracellular potentials, we obtain the distribution
of the transmembrane potential v.x; t/ D ui .x; t/  ue .x; t/; that is independent of
the reference potential. During a heart beat, the transmembrane potential of each
myocardial cell undergoes a time variation called transmembrane action potential
(TAP), characterized by a fast upstroke of about 1 ms and a subsequent slower
downstroke of about 50 ms, associated with the activation and recovery phases,
respectively, also called depolarization and repolarization phases.
Maps of the activation and recovery sequences provide important information for
identifying normal heart activity and cardiac arrhythmias that are often associated
with abnormal recovery times and action potential durations. Methods for determin-
ing activation and recovery times on the epicardial and endocardial surfaces, as well
as transmurally (i.e. in the thickness of the ventricular wall), are essential tools for
understanding the recovery process in normal and pathological conditions at both
experimental and clinical levels.
324 9 Simulation Studies of Cardiac Bioelectrical Activity

While methods for determining activation sequences from direct leads (recorded
directly from the heart) are well established, the assessment of local recovery times
is more difficult, due to the larger time and space scales involved in the recovery
process. The activation time is generally defined as the time instant when the TAP
exhibits the fastest upstroke and the unipolar electrogram (EG) shows the fastest
downstroke (i.e. the intrinsic deflection) during the QRS complex; see e.g. [408]
for a recent update. The assessment of local recovery times is based on indexes
associated with the TAP downstroke phase and with the upstroke of the T wave in
the EG, see Fig. 9.45.
Widely used TAP recovery markers are the time of fastest repolarization RTtap ,
defined as the instant of TAP fastest downstroke during the recovery phase, and the
late recovery time RT90tap , defined as the instant when the TAP reaches 90 % of its
resting value during the downstroke phase.
The potential drop across the membrane can be measured only from a few sites
in a given preparation by microelectrodes (micropipette). Therefore, transmembrane
potentials cannot be simultaneously recorded from hundreds of sites in vivo. Optical

50 50
∂ v
t
0 0
v (mV)

−50 −50
APD APD90
AT RT ATtap RT90
tap
tap tap
−100 −100
0 100 200 300 0 100 200 300
40 40
∂ u ∂ttue
t e
ue (mV)

20 20
ARI
eg ARI90eg
0 0
AT RT AT RT90
eg eg eg eg
−20 −20
0 100 200 300 0 100 200 300

15 15
∂thmap
5 5
hmap (mV)

−5 −5
ARI ARI90
−15 hmap −15 hmap
AT RT AThmap RT90
hmap
hmap hmap
−25 −25
0 100 200 300 0 100 200 300
time (ms) time (ms)

Fig. 9.45 Transmembrane action potential (v), unipolar electrogram (ue ) and hybrid monophasic
action potential (hmap) waveforms and the associated activation and recovery time markers
(Reproduced with permission from [461])
9.5 Extracellular Markers of Excitation and Repolarization Times 325

techniques can be used to detect the transmembrane potential on exposed tissue

surface, such as the epicardial surface, but cannot be used for intramural recordings.
Transmural and intramural explorations across the ventricular wall can only be
afforded with extracellular multi-electrode arrays. These recording techniques are
applicable especially when exploring large regions of a beating heart. Therefore, it is
very important to derive from the analysis of the extracellular potential time course
some recovery time markers that accurately estimate the recovery TAP markers.
The most widely accepted EG recovery time marker for the fastest recovery time
RTtap is the time RTeg of maximum upslope during the T wave, see e.g. [221, 566].
The RTeg marker has been widely employed in many experimental studies on
animals [9, 135, 160, 194, 363, 512, 515] and humans [84, 87, 194, 517, 567, 580]; see
also the survey paper [133]. Nevertheless, its reliability in estimating the RTtap has
been assessed experimentally in only a few papers, i.e. [160, 221], by evaluating the
correlation between the activation-recovery interval (ARI, defined as the difference
between RTeg and the activation time) and the action potential duration (APD,
defined as the difference between RTtap and the activation time), under a variety of
physiological conditions. A simulation study of the reliability of RTeg versus RTtap
has been carried out in [496], using a one-dimensional model of a cardiac strand.
Recently in [127], an EG marker for late recovery time based on the time RT90eg
of minimum second derivative of the EG waveform during the T wave has been
proposed.
Another extracellular technique for determining recovery times is based on
bipolar signals recorded by taking the potential difference between a site inside a
permanently depolarized area (obtained e.g. by pressure, suction or KCl injection)
and an exploring site. When the exploring and permanently depolarized sites are
very close to each other, we obtain the classical monophasic action potential (MAP),
see e.g. [181]. Instead, when we record the potential difference between a fixed
extracellular recording site inside a permanently depolarized region, and a generic
exploring site, we obtain the hybrid monophasic action potentials (HMAP), see
[553]. The TAP and MAP techniques cannot be implemented extensively in vivo.
Conversely, the EG and HMAP extracellular recording techniques are applicable in
studying the activation and recovery sequences in large regions of a beating heart
in vivo. Since both TAP and HMAP waveforms exhibit monophasic downstroke
phases, we can define the HMAP recovery markers RThmap and RT90hmap in the
same way as the TAP markers RTtap and RT90tap , respectively. Due to the close
resemblance of TAP and MAP waveforms at the same exploring site, there is a
general agreement on the reliability of MAP recovery times. On the other hand, the
use of a fixed extracellular recording site inside a permanently depolarized region
and the more varied morphology of the HMAP signal has made controversial the
information content of the HMAP, see e.g. [128, 135].
It is important to validate the previously described methods for assessing the local
recovery time from unipolar or bipolar recordings by comparing these markers with
the TAP markers, which are widely considered to be the best standard. The extension
of the pioneering one-dimensional investigation of [496] to three dimensions is
a demanding computational task, since it requires efficient numerical solvers for
326 9 Simulation Studies of Cardiac Bioelectrical Activity

the Bidomain model describing the cardiac bioelectric activity at a macroscopic

tissue level. In the last years, the development of powerful computing platforms and
efficient numerical parallel solvers have made possible large scale three-dimensional
Bidomain simulations, see e.g. [104, 458, 545], allowing the investigators to plan
and accomplish reliability studies of EG recovery time markers in 3D cardiac tissue
models.
Recently, using parallel Bidomain solvers [399, 458], the reliability of RTeg
versus RTtap markers has been investigated in normal [400, 460] and pathological
[461] cardiac tissue. In the following sections, we will review some of the results
obtained in [460, 461].

9.5.1 Waveform Postprocessing and Repolarization Time

Markers

In addition to the intracellular, extracellular and transmembrane potential fields

ui .x; t/; ue .x; t/; v.x; t/ D ui .x; t/  ue .x; t/ defined before, we will also consider
the hybrid monophasic action potential (HMAP)

hmap.x; t/ D ue .xPD ; t/  ue .x; t/;

where xPD is a point in a permanently depolarized region of ˝H , i.e. v.xPD ; t/ D

vPD is constant in time. We now define some markers of activation time (AT) and
repolarization time (RT) derived from these potentials.
Transmembrane action potential (TAP) waveforms and markers. We recall
that the ventricular transmembrane action potential (TAP) given by the time course
of v.x; t/ at a given point x 2 ˝H (see e.g. Fig. 9.45 top left panel) displays
mainly five phases, having different time scales. The first (phase 0) is related to the
excitation process, also called depolarization, where the TAP undergoes an abrupt
temporal change lasting about 2 ms, followed by a fast initial repolarization (phase
1) characterized by a fast exponential decay toward a plateau value. The plateau
phase (phase 2) lasts about 250–300 ms during normal heartbeats (depending on
the heart rate) or about 50–100 ms during tachyarrhythmias. In this phase, the
potential varies very slowly in comparison with the other phases. The plateau phase
is followed by the faster terminal repolarization (phase 3), where the TAP returns
to the resting value and after which the tissue becomes normally excitable again
(phase 4).
The excitation phase of the TAP is characterized by a fast upstroke with a well-
defined activation time, given by the instant ATtap of the TAP maximum derivative.
On the other hand, local repolarization time can be assessed by different markers
associated with the TAP downstroke phase. Widely used repolarization markers are
the time RTtap of the TAP minimum derivative and the time RT90tap when the TAP
reaches 90 % of the resting value during the downstroke phase (phase 3). The former
9.5 Extracellular Markers of Excitation and Repolarization Times 327

is related to the moment of fastest repolarization, while the latter indicates the
ending phase of repolarization. The TAP activation and repolarization time markers
(sketched graphically in Fig. 9.45) are defined as follows:
• ATtap .x/ D argmax f@t v.x; t/; t 2 upstrokeg,
i.e. the instant of maximum time derivative of v.x; t/ during the upstroke;
• RTtap .x/ D arg min f@t v.x; t/; t 2 downstrokeg,
i.e. the instant ˚of minimum time derivative of v.x; t/ during the downstroke;
• RT90tap .x/ D t W v.x; t/ D 0:9vr during downstroke ,
i.e. the instant when v.x; t/ reaches 90 % of its resting value vr during down-
stroke.
The TAP activation and repolarization markers ATtap , RTtap and RT90tap are the gold
standards for determining cardiac activation and repolarization times.
Unipolar electrograms (EG) waveforms and markers. A typical morphology
of an electrogram (EG), given by the time course of ue .x; t/ at a given point x 2 ˝H ,
is displayed in the mid panel of Fig. 9.45. The excitation phase is related to the
QRS complex, while the faster terminal repolarization occurs during the T wave.
This wave has positive and negative polarity for sites that repolarize early and late,
respectively, while it has a biphasic waveform for intermediate repolarization states.
A firmly established marker for the activation time from the EGs is the instant ATeg
of minimum derivative during the QRS complex; see e.g. [408, 493]. The most
widely accepted EG repolarization marker is the time RTeg of maximum upslope
during the T wave, see e.g. [221, 566]. Recently, we have proposed an EG marker
for late repolarization based on the time RT90eg of minimum second derivative of
the EG waveform during the T wave, see [460]. Note that the second time derivative
of electrograms has been used in [187, 188, 311] in order to define the end of the T
wave. The EG activation and repolarization time markers (sketched graphically in
Fig. 9.45) are defined as follows:
• ATeg .x/ D arg min f@t ue .x; t/; t 2 QRS complexg,
i.e. the instant of minimum time derivative of ue .x; t/ during the QRS complex;
• RTeg .x/ D argmax f@t ue .x; t/; t 2 T waveg,
i.e. the instant of maximum time derivative of ue .x; t/ during the T wave;
• RT90eg .x/ D arg min f@t t ue .x; t/; t 2 T waveg,
i.e. the instant of minimum second time derivative of ue .x; t/ during the T wave.
Hybrid monophasic action potential (HMAP) waveforms and markers. The
extracellular waveforms, recorded by Weissenburger et al. [553], are the potential
difference between a fixed extracellular recording within a permanently depolarized
region and a generic exploring site. These signals can be obtained directly from
the heart or computed as the difference between the unipolar EG from a fixed
permanently depolarized site and the unipolar EG from a generic exploring site.
These computed signals are obviously independent of the (identical) reference
potential used for the two unipolar EGs. We call these extracellular signals, hybrid
monophasic action potentials (HMAPs) to distinguish them from the classical close
bipolar monophasic action potentials (MAP) of Franz [181], where the exploring
328 9 Simulation Studies of Cardiac Bioelectrical Activity

and permanently depolarized sites are always very close to each other. HMAPs
are contaminated by far-field effects, while MAPs, being close bipolar signals,
are relatively insensitive to far-field effects. A typical morphology of an HMAP
waveform is displayed in the bottom panel of Fig. 9.45. The HMAP signal exhibits
multiple upstroke phases followed by the appearance of a monophasic component.
The fastest upstroke is associated with the activation of the exploring site and the
others with the depolarization of sites around the boundary of the PD volume. The
HMAP activation and recovery time markers are defined in the same way as the
TAP markers:
• AThmap .x/ D argmax f@t hmap.x; t/; t 2 upstrokeg,
i.e. the instant of maximum time derivative of hmap.x; t/ during the upstroke;
• RThmap .x/ D arg min f@t hmap.x; t/; t 2 downstrokeg,
i.e. the instant ˚of minimum time derivative of hmap.x; t/ during the downstroke;
• RT90hmap .x D t W hmap.x; t/ D 0:9vr during downstroke ,
i.e. the instant when hmap.x; t/ reaches 90 % of its resting value during the
downstroke.
We recall that the information content of the HMAP waveform is controversial,
see e.g. [182, 265, 285, 347, 542, 553]. In particular, the reliability of recovery time
markers derived from HMAP signals has been questioned, see [128, 135, 560].
As a byproduct of the AT and RT markers defined above, the following action
potential duration (APD) and Activation–Recovery Interval (ARI) markers are
defined.
APD markers from TAP waveforms.

APD.x/ D RTtap .x/  ATtap .x/; APD90.x/ D RT90tap .x/  ATtap .x/:

ARI markers from EG waveforms.

ARIeg .x/ D RTeg .x/  ATeg .x/; ARIhmap .x/ D RThmap .x/  AThmap .x/I

ARI90eg .x/ D RT90eg .x/ATeg .x/; ARI90hmap .x/DRT90hmap .x/AThmap .x/:

ARIeg has been widely used in estimating the APD in experiments on animals
and humans, see e.g. [135, 221, 328] and [84, 194, 517], respectively.
We recall that the excitation process is a self-sustained propagating phenomenon
and the isochrones of the activation time define a propagating excitation front.
Conversely, the repolarization process is not a matter of conduction but it is
a synchronization phenomenon, see e.g. [133]. Nevertheless, the isochrones of
successive repolarization times define a sequence of repolarization fronts and even
if these fronts are not propagating in a classical sense, for convenience of notation
we will use in the following the terms acceleration, deceleration, collision regarding
the sequence of repolarization fronts sweeping the myocardial volume.
9.5 Extracellular Markers of Excitation and Repolarization Times 329

9.5.2 Parameter Calibrations for the Model Simulations

In all computations, a structured grid of 192  192  48 hexahedral isoparametric Q1

finite elements of size h D 0:1 mm is used in space. All simulations have been run
on a Linux Cluster with 56 Opteron AMD processors and Infiniband network. Each
simulation required about 8 h on 36 processors.
Multi-electrode array. The cardiac domain ˝H considered in this study is
a cartesian slab of dimensions 1:92  1:92  0:48 cm3 , modeling a portion of
the left ventricular wall. In this slab, we consider a matrix of 12  12 exploring
multielectrode needles spaced 1.6 mm from each other and 0.8 mm from the slab
boundary, as shown in Fig. 9.46. Each needle carries 13 recording sites, spaced
0.4 mm along the shank. We then have 12  12 sites on each of the 13 intramural
planes, for a total of 12  12  13 D 1 872 recording sites in the slab, each recording
the intra and extracellular potentials. We will indicate each needle location by its
column and row indexes, shown in the left panel of Fig. 9.46. This multi-electrode
array mimics the experimental setup used in intramural mapping of the extracellular
potential, see e.g. [162, 184].
Transmural heterogeneity. Three different types of transmural distribution of
the intrinsic APDs of the cells are considered, one homogeneous (H-slab) and
the other two heterogeneous (3-slab and W-slab). The transmural intrinsic APD
heterogeneity is assumed to be the same along any transmural epi-endocardial
straight line, i.e. in any plane parallel to the epicardium all cells have the same
intrinsic APD. In the heterogeneous slabs, the intrinsic APD of the cells is obtained
by multiplying the potassium current IK in the LR1 model [308] by a factor factIK ,
as detailed in Table 9.8. This modulation factor is chosen in order to introduce a
transmural APD profile with M-cell layers as in [546] (3-slab) or as in [572, Fig. 4],
[395, Fig. 5] (W-slab), to mimic their experimental transmural APD profile; see
[126, 460] for more details.

PD3 PD2 PD1

12
11
10
EPI
FIBER 9
DIR.
o
(−45 )
8
7
6
5
ISCHEMIC REGION 4
3
ENDO 2
FIBER A
DIR. TRANSMURAL NEEDLES 1
o
(+45 ) 1 2 3 4 5 6 7 8 9 10 11 12

Fig. 9.46 Left: cardiac slab ˝H , PD sites, ischemic region, transmural needles. Right: needle
locations on the epicardial plane and their row and column indexes (Reproduced with permission
from [461])
330 9 Simulation Studies of Cardiac Bioelectrical Activity

Table 9.8 Parameter calibration for modeling the transmural heterogeneities in the three cardiac
slabs H-slab, 3-slab, W-slab
H-slab 3-slab W-slab
# of layers 1 3 4
Endo Mid Epi Endo Sub-endo Mid Epi
Thickness (cm) 0.48 0.16 0.16 0.16 0.058 0.096 0.254 0.072
fact_IK 1 2.62 1.95 2.88 2.71 1.95 2.47 2.88
APD (ms) 266 235 272 225 232 272 242 225

Subendocardial ischemia. Two simulations with subendocardial moderate (MI-

slab) and severe (SI-slab) ischemic regions (defined below) are performed. The
ischemic region has dimensions 0:4  0:4  0:16 cm3 and is located as shown in
Fig. 9.46. In the LR1 model, the current IK is scaled by a factor 2.325, yielding TAPs
with APD90 D 250 ms. Inside the ischemic region, the extracellular potassium
concentration ŒKo is increased from 5:4 mM (control) to 10:5 mM (MI-slab) or
18 mM (SI-slab); for more details see [129].
Permanently depolarized (PD) volume. A permanently depolarized (PD) site
is obtained experimentally by contact pressure or by a KCl injection in a region
HD , holding the TAP in such a region to some fixed depolarized value vD , i.e. for
x 2 HD ; v.x; t/ D vD . In our simulation study, such a PD site is obtained by
assigning the extracellular potassium concentration equal to the intracellular one,
i.e. IK1 is zero in the small PD volume. The PD site is labeled PD in Fig. 9.46 and it
has dimensions 0:8  0:8  0:8 mm3.
Stimulation site. Inside the PD volume the transmembrane potential values
are above threshold thus generating a first excitation-recovery TAP that sweeps
the cardiac slab ˝H . We wait for 500 ms and take the steady state reached by
the Bidomain system as the initial condition for our simulations. An extracellular
e
stimulus (iapp D 250 mA=cm3 for 1 ms) is then applied in a small volume (3 mesh
points in each direction) at the locations A in Fig. 9.46 and an intracellular stimulus
i
iapp D iapp
e
is also applied in order to satisfy the compatibility condition for the
solvability of the Bidomain system (3.42).
Figure 9.47 displays the isochrones of ATtap , RT90tap and APD90 on epicardial,
midmyocardial and endocardial intramural sections and on a transmural diagonal
section perpendicular to the epicardium in the PD3 case. The excitation wave front
starts at the stimulated endocardial corner A and proceeds faster along the fiber
direction than in the orthogonal direction. In fact, in Fig. 9.47 the endocardial
and transmural activation isochrones present a quasi-elliptical shape with major
axis parallel to the fibers. Excitation reaches the epicardial breakthrough about
20 ms after the stimulus, and then in about 40 more ms the whole epicardium
is depolarized. Two distinct repolarization sequences start, almost simultaneously,
on the endocardial and epicardial surfaces, and subsequently collide and merge
transmurally in the middle region of the cardiac wall. This produces an APD90
9.5 Extracellular Markers of Excitation and Repolarization Times 331

ACTI REPO APD

60 230 ACTI 60
300
40 220 40
250
EPI

20 200 210 20

150 200 0.10 62.94 2.00

0.05 62.94 2.00 −300.00 320.81 2.00 −357.95 266.94 0.50
60 230 REPO
50 250
250 220
MID

40
30 200 210 200
20
150 200 219.95 280.83 2.00 150
10.75 64.61 2.00 233.08 296.08 2.00 221.15 232.79 0.50

60 220 APD
250 215
ENDO

220
40
210
200
20 205 210

150 200 200.25 229.44 2.00

0.10 71.56 2.00 219.95 290.46 2.00 211.77 222.40 0.50 200

Fig. 9.47 3-slab, PD3. Activation (ACTI), repolarization (REPO) and action potential duration
(APD) isochrone lines, computed with space resolution h D 0:1 mm, related to the transmembrane
markers ATtap .x/, RT90tap .x/, APD90.x/. Left panel: intramural sections (epicardium, midwall,
endocardium). Right panel: transmural diagonal section A perpendicular to the epicardium (see
Fig. 9.46). Maximum, minimum and step of the displayed map are reported below each panel
(Reproduced with permission from [461])

transmural pattern with a maximum in the midmyocardial regions, where cells with
a longer intrinsic APD are located.
Postprocessing. We saved the extracellular and the intracellular potential wave-
forms ue .x; t/ and ui .x; t/ at the 12  12  13 locations of the multi-electrode array
described above. These waveforms for ue and for the byproduct v D ui  ue are then
post-processed by computing the additional activation and repolarization markers
defined previously. The TAP markers are assumed to be the reference markers.
The EG and HMAP markers are compared with the reference TAP markers, i.e.:
ATeg and AThmap vs ATtap ; RTeg and RThmap vs RTtap ; RT90eg and RT90hmap vs
RT90tap ; ARIeg and ARIhmap vs APD; ARI90eg and ARI90hmap vs APD90.
In order to compare these large data sets, we then compute, for vectors X; Y of
length n D 1 872, the following quantities, reported in the Tables of the Results:
P
• mean.X / D n1 n1 X pi D the average of X ;
• std.X / D kX0 k= n D the standard deviation of X , where X0 D X 
mean.X /1;
• corr.X; Y / D .X0 =kX0 k; Y0 =kY0 k/ D the correlation coefficient between X
and Y ;
• mrd D mean.jX  Y j/=mean.jY j/ D relative mean error of X with respect to Y ,
where k  k; .; / denote the Euclidean norm and scalar product.
332 9 Simulation Studies of Cardiac Bioelectrical Activity

9.5.3 Global Quantitative Analysis of RT Markers

Effects of the PD location on the extracellular potential field. Table 9.9 reports
the comparison between the EG and HMAP repolarization markers and the refer-
ence TAP markers (RTeg vs RTtap , RThmap vs RTtap , RT90eg vs RT90tap , RT90hmap
vs RT90tap ). The sites with discrepancies larger than 30 ms are excluded but, in
each case, the points excluded are less than 3. All the extracellular estimates
show a high reliability as confirmed by a correlation coefficient always greater
than 0.95. The RT90eg marker exhibits the best performance, followed by RTeg
and then by RThmap , with average absolute discrepancies ranging between 1:5 and
3:2 ms. Despite the high correlation coefficient, the RT90hmap estimate exhibits
a larger average discrepancy ranging between 3:1 and 11:3 ms depending on the
location of the PD volume. The relevant increase of both average discrepancies and
standard deviations for all markers with respect to the case involving only isotropic
waveforms (see [461]) indicates that the anisotropic component of the extracellular
potential ue is the major determinant of the discrepancies between EG and TAP
repolarization markers.
The EG markers are weakly influenced by the location of the PD volume.
The global performance of the RThmap marker exhibits a weak dependence on the
position of the PD volume. Conversely, the performance of the RT90hmap marker,
in terms of average discrepancy not of correlation, is strongly dependent on the
PD volume location. This strong performance dependence on the location of the
PD volumes may be related to the different morphology of the EGs in the PD

Table 9.9 RT markers discrepancies, 3-slab, PD volume in PD1, PD2 and PD3. EG and HMAP
markers computed from extracellular waveforms (ue and hmap) versus TAP markers. meanjX 
Y j D average of jX  Y j; meanjY j D average of jY j; mrd D meanjX  Y j/meanjY j D relative
mean discrepancy; std D standard deviation of jX  Y j; corr.X; Y / D correlation coefficient
between X and Y
Mean Mean Corr
X Y mrd jX  Y j jY j Std .X; Y /
PD1 RTeg RTtap 0.76e2 1.86 244.41 1.72 0.99
RThmap RTtap 1.30e2 3.17 244.43 3.50 0.96
RT90eg RT90tap 0.63e2 1.59 252.94 1.79 0.99
RT90hmap RT90tap 4.46e2 11.29 252.92 6.28 0.97
PD2 RTeg RTtap 0.77e2 1.91 246.87 1.63 0.98
RThmap RTtap 1.23e2 3.04 246.78 3.54 0.95
RT90eg RT90tap 0.64e2 1.64 255.58 1.95 0.99
RT90hmap RT90tap 1.21e2 3.08 255.26 2.51 0.96
PD3 RTeg RTtap 0.78e2 1.92 245.40 1.65 0.99
RThmap RTtap 0.97e2 2.39 245.42 2.23 0.98
RT90eg RT90tap 0.64e2 1.63 254.01 1.75 0.99
RT90hmap RT90tap 2.07e2 5.26 254.02 2.45 0.98
9.5 Extracellular Markers of Excitation and Repolarization Times 333

volumes. The EGs in the PD volumes exhibit an upstroke phase contaminated by

some spikes followed by a downstroke phase. The EGs in PD1 and PD3 present also
an undershooting behavior at the end of the downstroke phase, which is absent in
the EGs morphology for sites inside the PD2 volume, see Fig. 9.48. In the following
simulations, we fix the location of the PD volume in the PD3 region.
Transmural heterogeneity and subendocardial ischemia. We now simulate
the excitation and repolarization processes in tissue slabs with different heteroge-
neous cellular membrane properties, i.e. in H-slab, 3-slab, W-slab, MI-slab, SI-slab
with PD volume located in PD3, by applying a local stimulus at the endocardial
vertex A in Fig. 9.46. We evaluate the global performance of the extracellular
markers for each slab type and for the entire collection of the five simulations, by
considering marker values with discrepancies less than 30 ms. In the multielectrode
array of 1 872, the discarded marker values are at most 6 and mostly related to sites
within the inexcitable regions, i.e. the PD volume and the ischemic region in the
SI-slab type.
The comparison of the activation markers shows that both the extracellular
markers ATeg and AThmap are very accurate estimates of the reference marker ATtap ,
with absolute discrepancies of the order of 0.05 ms. This shows that the instant of
maximum time derivative of the HMAP matches very well the instant ATtap in spite
of the possible presence of multiple upstrokes in the HMAP signal.
The results reported in Table 9.10 show that all the extracellular recovery
markers provide very reliable estimates of the reference transmembrane markers,
with correlation coefficients always greater than 0:98. This good match between
the EG repolarization markers and the reference TAP markers is also confirmed by

40

35

30

25

20 PD3
mV

PD2
15

10
PD1
5

−5
0 50 100 150 200 250 300 350
ms

Fig. 9.48 Shifted extracellular waveforms ue .xPDR; t / in the PD volume for xPD 2
PD1; PD2; PD3 (see Fig. 9.46) and CR.t / D j˝1H j ˝H v.x; t / dx (dashed line) for the 3-slab
(Reproduced with permission from [461])
334 9 Simulation Studies of Cardiac Bioelectrical Activity

Table 9.10 Recovery time (RT) markers discrepancies for H-slab, 3-slab, W-slab, MI-slab and
SI-slab with PD volume in PD3. Comparison between the RT markers (RTeg vs RTtap , RThmap vs
RTtap , RT90eg vs RT90tap , RT90hmap vs RT90tap ): meanjX Y j D average of jX Y j; meanjY j D
average of jY j; mrd D meanjX  Y j/meanjY j D relative mean discrepancy; std D standard
deviation of jX  Y j; corr.X; Y / D correlation coefficient between X and Y
Mean Mean Corr
X Y mrd jX  Y j jY j Std .X; Y /
H-slab RTeg RTtap 0.84e2 2.14 253.22 1.98 0.99
RThmap RTtap 0.72e2 1.83 253.15 1.60 0.99
RT90eg RT90tap 0.36e2 0.95 260.32 0.86 0.99
RT90hmap RT90tap 1.17e2 3.04 260.38 2.15 0.98
3-slab RTeg RTtap 0.78e2 1.92 245.40 1.65 0.99
RThmap RTtap 0.97e2 2.39 245.42 2.23 0.98
RT90eg RT90tap 0.64e2 1.63 254.01 1.75 0.99
RT90hmap RT90tap 2.07e2 5.26 254.02 2.45 0.98
W-slab RTeg RTtap 0.56e2 1.39 248.25 1.19 0.99
RThmap RTtap 0.84e2 2.09 247.52 2.18 0.98
RT90eg RT90tap 0.41e2 1.04 255.66 1.09 0.99
RT90hmap RT90tap 2.00e2 5.10 255.79 1.99 0.99
MI-slab RTeg RTtap 0.86e2 2.16 250.76 2.23 0.99
RThmap RTtap 0.77e2 1.94 250.63 2.22 0.99
RT90eg RT90tap 0.51e2 1.31 258.92 1.91 0.99
RT90hmap RT90tap 1.19e2 3.07 258.95 2.24 0.98
SI-slab RTeg RTtap 0.79e2 2.01 252.75 1.67 0.99
RThmap RTtap 0.75e2 1.90 252.24 1.80 0.99
RT90eg RT90tap 0.52e2 1.34 258.86 2.19 0.99
RT90hmap RT90tap 1.14e2 2.95 258.94 2.13 0.98

Fig. 9.49, reporting the regression plots of RTeg vs RTtap , RThmap vs RTtap , RT90eg vs
RT90tap and RT90hmap vs RT90tap on all five simulations. In terms of accuracy, the
RT90eg marker gives the best performance, with a relative average discrepancy of at
most 0.64e2 and an average absolute discrepancy of at most 1:63 ms. The RTeg and
RThmap , as estimates of RTtap , show comparable global performance, because both
the absolute and relative averages of jRTeg –RTtap j and jRThmap –RTtap j are about 2 ms
and 0.8e2 with a range 0.56e2  0.97e2. As previously observed only for the
3-slab, despite a global correlation coefficient of 0:98, the RT90hmap marker exhibits
larger absolute (relative) average discrepancies, ranging between 3:04 (1.17e2)
and 5:3 ms (2.1e2), than the other EG markers.
These global performance findings are also confirmed by the distributions of
the discrepancies RTeg –RTtap , RThmap –RT90tap , RT90eg –RT90tap and RT90hmap –
RT90tap , reported in Fig. 9.50. These distributions clearly show that almost all
discrepancies range between 10 and 10 ms and most of them are confined within
5 and 5. Figure 9.51 reports the discrepancies with respect to the transmural
9.5 Extracellular Markers of Excitation and Repolarization Times 335

y = −13.5 + 1.1*x, corr = 0.98 y = 0.5 + 1.0*x, corr = 0.99

300 310

RT90eg
RTeg

160 170
160 300 170 310
RTtap RT90tap

y = 12.1 + 0.9*x, corr = 0.98 y = 2.5 + 1.0*x, corr = 0.98

300 310
hmap
hmap

RT90
RT

160 170
160 300 170 310
RT RT90
tap tap

Fig. 9.49 Regression lines of RTeg vs RTtap (first column), RT90eg vs RT90tap (second column),
RThmap vs RTtap (third column) and RT90hmap vs RT90tap (fourth column) for all the five slab types
H-slab, 3-slab, W-slab, MI-slab and SI-slab (Reproduced with permission from [461])

position of the electrodes for the 3-slab, i.e. the discrepancy variation over an
intramural plane.
For 3-slab, we found that the largest discrepancies are located in the midmyocar-
dial regions, where the two repolarization fronts, that started from the endocardium
and the epicardium, merge. An analogous consideration apply to the subendocardial
regions in W-slab. We refer to our recent paper [460] for a discussion of the locations
and origin of the RTeg and RT90eg markers largest discrepancies, as well as for a
study of the associated artifacts appearing in the repolarization sequences. Signs
of the presence of these discrepancies are the correlation coefficients of the APD
markers, reported in Table 9.11, which are always lower than those of the associated
RT markers, indicating a loss of precision of the extracellular RT markers. We
remark that in all simulations the global performance indexes (mrd, mean, std)
336 9 Simulation Studies of Cardiac Bioelectrical Activity

3000 3000
NUMBER OF SITES

2000 2000

1000 1000

0 0
−20 −10 0 10 20 −20 −10 0 10 20
RTeg − RTtap RT90eg − RT90tap

3000 3000

2000 2000

1000 1000

0 0
−20 −10 0 10 20 −20 −10 0 10 20
RThmap − RTtap RT90 − RT90
hmap tap

Fig. 9.50 Histograms of discrepancies RTeg RTtap (first column), RT90eg RT90tap (second
column), RThmap RTtap (third column) and RT90hmap RT90tap (fourth column) with 1 ms bins,
for all the five slab types H-slab, 3-slab, W-slab, MI-slab and SI-slab (Reproduced with permission
from [461])

reported in Table 9.10, except the correlation coefficients (corr), are the same for
the pairs ARIeg (ARIhmap ) vs APD and RTeg (RThmap ) vs RTtap , as well as for the
pairs ARI90eg (ARI90hmap ) vs APD90 and RT90eg (RT90hmap ) vs RT90tap , because
the relative and absolute discrepancies of jATeg ATtap j and jAThmap ATtap j are of
the order of 1.e4 and of 0.05 ms, respectively.
When considering exploring sites located a few mm from the PD volume, the
bipolar signal HMAP simulates the condition of the classical monophasic action
potential (MAP), see Franz [181], because the exploring and the PD reference
electrode are very close. In order to evaluate the markers performance at these
particular sites, we now consider in each simulation the six electrodes close to the
PD volume (within 2 mm distance), yielding a total amount of 30 electrodes, and
we compare the RT markers at these sites. The average indexes (mean, std, corr) are
(1.06, 0.60, 0.99) for jRThmap –RTtap j and (1.99, 0.56, 0.96) for jRT90hmap –RT90tap j,
to be compared with (3.84, 1.24, 0.99) for jRTeg –RTtap j and with (2.04, 1.70, 0.95)
for jRT90eg –RT90tap j, respectively.
9.5 Extracellular Markers of Excitation and Repolarization Times 337

20 20

10 10

RT90eg − RT90tap
− RTtap

0 0
eg
RT

−10 −10

−20 −20
ENDO MID EPI ENDO MID EPI

20 20

− RT90tap
10 10
tap
RThmap − RT

0 0
hmap
RT90

−10 −10

−20 −20
ENDO MID EPI ENDO MID EPI

Fig. 9.51 3-slab, PD3. Discrepancies RTeg RTtap (first column), RT90eg RT90tap (second col-
umn), RThmap RTtap (third column) and RT90hmap RT90tap (fourth column) (Reproduced with
permission from [461])

Table 9.11 Correlations ARIeg ARIhmap ARI90eg ARI90hmap

between action potential
vs vs vs vs
duration markers. ARIeg vs
APD, ARIhmap vs APD, APD APD APD90 APD90
ARI90eg vs APD90, H-slab 0.84 0.77 0.91 0.72
ARI90hmap vs APD90 3-slab 0.95 0.94 0.94 0.92
W-slab 0.96 0.92 0.98 0.95
MI-slab 0.91 0.93 0.95 0.90
SI-slab 0.90 0.84 0.90 0.63

These results show that near the PD volume RThmap (RT90hmap ) is a better
estimate of RTtap (RT90tap ) than RTeg (RT90eg) and consequently confirms that the
downstroke phase of the HMAP yields a more accurate approximation of the TAP
downstroke phase. This fact is also confirmed by the comparison of a suitable scaled
and shifted TAP waveform with the HMAP waveform displayed in Fig. 9.52-top for
a site close to the PD volume. Figure 9.52-bottom shows the plots of TAP, EG and
HMAP waveforms at three endocardial sites, for the 3-slab in the PD3 case. The EGs
exhibit T waves with different polarity, i.e. positive, biphasic and negative. We have
338 9 Simulation Studies of Cardiac Bioelectrical Activity

50 50

0 0

−50 −50

−100 −100
0 100 200 300 0 100 200 300

40 40

20 20

0 0

−20 −20

−40 −40
0 100 200 300 0 100 200 300
RTeg − RT tap = 3.77 RT90eg − RT90 tap = −2.74

40 40

20 20

0 0

−20 −20

−40 −40
0 100 200 300 0 100 200 300

RThmap − RT tap = 0.86 RT90 hmap − RT90 tap = −1.71

50 50

0 0

−50 −50

−100 −100
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350

40 40

20 20

0 0

−20 −20

−40 −40
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
RT − RT = 0.17 (A) 1.03 (B) 2.23 (C) RT90 − RT90 = 2.23 (A) 0.51 (B) 1.03 (C)
eg tap eg tap

40 40

20 20

0 0

−20 −20

−40 −40
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350

RThmap − RTtap= 1.71 (A) 1.37 (B) 0.86 (C) RT90 − RT90 = −5.48 (A) −3.77 (B) −3.60 (C)
hmap tap

Fig. 9.52 Top panel. 3-slab, PD3: TAP (first row), EG (second row) and HMAP (third row)
waveforms in the epicardial site located in the needle (2,12), see Fig. 9.46. The TAP waveform,
plotted in dashed line in the third row superimposed to the HMAP, is shifted in order to have
the same resting value of the HMAP waveform and it is scaled by a factor ˛. The vertical lines
indicate RTtap (first row, left), RTeg (second row, left) and RThmap (third row), RT90tap (first row,
right), RT90eg (second row, right) and RT90hmap (third row, right). Bottom panel. 3-slab, PD3: same
format as in the top panel for the waveforms on the three endocardial sites located in the needles
(2,2) (A), (6,6) (B) and (11,11) (C), see Fig. 9.46 (Reproduced with permission from [461])

not observed a significant dependence of the performance of the EG markers on the

T wave polarity, in agreement with [127,135] and in contrast with the controversial
experimental findings of [87].
9.6 Subendocardial Ischemia, ST Depression and Elevation 339

9.6 Subendocardial Ischemia, ST Depression and Elevation

Myocardial ischemia is a reduction of blood supply to a region of the myocardium,

normally due to the occlusion of one or more coronary arteries. The ischemic
region is classified either as subendocardial if it begins at the endocardial surface
and extends partially through the heart wall, or as transmural, if it reaches the
epicardium. Understanding the consequences of subendocardial ischemia on the
epicardial surface has an important clinical interest. One way to locate ischemia in
clinical electrocardiography is the so-called ST segment elevation in the electrocar-
diogram (ECG), but both the choice of this criterion and the mechanism underlying
ST shifts are still controversial.
Several experimental works have studied these phenomena: Samson et al.[453],
Prinzmetal et al.[403], Kjekshus et al.[286], and Kleber et al.[279], suggested
that subendocardial ischemia produces ST elevation at the epicardial level, while
Wolferth et al.[565] and Li et al.[298] found ST depression. MacLeod et al.[314]
observed depression only under conditions of combined coronary occlusion and
elevated heart rate. Many simulation studies (e.g. [235, 262, 263, 313]) investigated
the mechanisms behind the ST segment shifts found in experiments as a marker
for predicting and locating the myocardial injury within the first 15–20 min of an
acute ischemic episode. These studies are mainly based on a time independent
Bidomain model assuming a stationary transmembrane potential distribution with a
sharp variation across the ischemic boundary during the diastolic or systolic phases
of the heartbeat. It is shown in [235, 313] that for a subendocardial ischemic region
the epicardial ST elevation is associated with the transmural ischemic boundary,
while ST depression is associated with the lateral ischemic boundaries.
In this section, we simulate the entire QRST complex of the heartbeat by integrat-
ing the Bidomain model, a degenerate system of two parabolic reaction-diffusion
equations, coupled with Luo-Rudy type models for the ionic currents, described
by a stiff systems of ordinary differential equations. We study the influence of an
ischemic subendocardial region on the excitation and the recovery sequences, the
distribution of the action potential duration (APD) and the extracellular potential
patterns at intramural and epicardial levels. In order to better understand the
mechanism of ischemic ST segment shifts, we decompose the cardiac current
sources into conormal, axial and orthogonal components and investigate which
component is dominant during the ST interval. Other recent simulation studies of
ischemic regions can be found in [353, 354, 456].

9.6.1 Mechanisms for the ST Segment Potential Patterns

We briefly recall the well-established mechanism at a cellular level (see [564] Ch. II
Fig. 2.13) of the potential changes during ST interval, evidenced first by Samson
and Sher[453], Prinzmetal et al.[403] and later confirmed by Kleber et al.[279]; see
340 9 Simulation Studies of Cardiac Bioelectrical Activity

also the updated surveys[257, 276] and the references therein. Due to the reduction
of APD, the transmembrane potential in the ischemic cells during the ST interval is
lower than in non-ischemic cells. This transmembrane potential difference produces
a local intracellular current of injury flowing from non-ischemic to ischemic cells,
associated with a local extracellular current flowing in opposite direction. This latter
current is the cause of an elevation of the extracellular potential (ST elevation)
recorded from the ischemic zone and a depression detected in the healthy part.
Thus, during the ST interval, at the tissue level we have a distribution of local
injury currents sources around the ischemic boundary. According to Holland and
Brooks[231], one way to extend the cellular mechanism to the ventricular wall
is based on assuming that the elementary injury sources can be model as: (i) a
dipole layer on the ischemic boundary, with dipole direction normal to this boundary
and with moment proportional to the transmembrane potential jump; (ii) the dipole
layer is embedded into an isotropic homogeneous and infinite conducting medium.
This modeling of the injury sources, known as solid angle theory, predicts at
the epicardial side an ST depression associated with the subendocardial ischemic
volume. This extension of the cellular mechanisms to a tissue level is questionable,
since the cardiac tissue exhibits a strong anisotropic structure and the injury currents
sources are embedded in rotational anisotropic intra and extracellular media.
We apply now the source split of the extracellular potential field described in
Chap. 5 to the particular case of the ST segment in the presence of an ischemic
subendocardial region. Denoting by ˝i the ischemic tissue, define i D @˝i ,
˙i D i \ ˙, and l D i n ˙i , where ˙ is the non-insulated boundary of
˝H , see Fig. 5.1. In the ST segment, the distribution of the transmembrane results
can be practically approximated by a piecewise constant distribution. Denoting
by vIr , vN
r the values of the transmembrane potential in the ischemic and normal
tissue, respectively, then v.x; t/ D vN r C .vr  vr /˝i .x/, where ˝i denotes the
I N

characteristic function of the set ˝i (i.e. ˝i .x/ D 1 if x 2 ˝i or 0 if x 62 ˝i ).

The injury current sources are concentrated on the ischemic boundary i , hence
rv.x; t/ D .vIr  vN r /r ˝i .x/ D .vr  vr /ni ıi .x/, where n denotes the normal
I N

vector to i pointing inside the ischemic domain ˝i and ıi .x/ is a Dirac measure
on i . Therefore, the injury current can be represented as an oblique dipolar layer
with moment given by

Jv D .vIr  vN
r / Di .x/n.x/ ıi .x/:

Applying the split of the tensor Di given in (5.40), the oblique dipole layer Jv .x/ can
be viewed as the superposition of a conormal dipole with direction Dn, of a axial
dipole layer having the same direction of the fiber al , and of a cross-axial dipole
layer, since the matrix I  al aTl is the projection matrix on the plane orthogonal to
al , i.e.

Jv D Jc C Ja C Jo :
9.6 Subendocardial Ischemia, ST Depression and Elevation 341

Here
   
Jc D ˛Dn
O ıi Ja D ˇO aTl n al ıi ; Jo D  O I  al aTl n ıi ;

with ˛O D .vIr  vN O
r /.t  n /=.t  n /, ˇ D .vr  vr /.l  ˛l / and O D
i i I N i

.vr  vr /.t  ˛t / D .vr  vr /.n  ˛n /. We remark that since vIr > vN
I N i I N i
r ,
ti > ni and te > ne , it follows that ˛O > 0, ˇO > 0 and O < 0.
The conormal, axial and orthogonal current densities Jc ; Ja ; Jo define three
distributions of dipolar current sources lying on the ischemic boundary with dipole
axes parallel to the conormal direction Dn, to the fiber direction al and orthogonal
to the fiber direction, respectively. Since only rv appears in the current density, we
shift the distribution v by introducing the discrepancy with respect to the normal
resting value vN O D v  vN r D .vr  vr /˝i .x/.
I N
r , i.e. v
The transmembrane component uvO is given by
 
j˝i j
uvO .x/ D ˛.v
O Ir  vN / ˝i .x/  :
r
j˝H j

In the case of the endocardial surface in contact with the cavitary blood, we have

uc .x; t/ D uvO C uhs .x; t/;

where uhs is the heart surface component, see (5.41) and (5.42).
The full potential distribution u is the superposition of its field components

u.x; t/ D uvO .x; t/ C uhs .x; t/ C ua .x; t/ C uo .x; t/:

This expression leads to the following remarks.

(a) The transmembrane component uvO is a jump component, because its effect on
the full potential u reduces to a shift proportional to the jump vIr  vNr of the
transmembrane potential from the ischemic to the normal resting values through
the ischemic boundary. Therefore, uvO does not contribute to extracellular
currents and potential pattern.
(b) Since vO .x; t/ D 0, when x 2 ˙i D ˙ \ i , we have ŒŒ u.x; t/ ˙i D ˛Ov.x; t/ on
˙i and ŒŒ u.x; t/ ˙i D 0 on ˙ n ˙i . Therefore, the heart surface component
uhs depends on the ischemic resting value and on the area of boundary of the
ischemic region in contact with the cavitary blood. This field is the potential
generated by a dipole layer on ˙i with moment ˛vIr and direction orthogonal
to ˙i and pointing toward the blood cavity. Thus it is similar to the classical
342 9 Simulation Studies of Cardiac Bioelectrical Activity

uniform dipole layer source on ˙i , but embedded in the anisotropic bulk

medium. We remark that for a fully insulated slab, the heart surface contribution
disappears.
(c) The axial and orthogonal potentials ua ; uo are generated by surface sources
on i represented by a current dipole density parallel and orthogonal to
the local fiber direction al on the ischemic boundary, respectively; since the
fiber architecture defines the direction of the current source flow we expect
that structural properties like epi-endocardial fiber rotation on the ischemic
boundary will be the major determinants affecting the axial and orthogonal
potential patterns.
An alternative and equivalent way to relate the potential field to its current source
density is to use an integral representation that turns out to be more efficient than
the previous differential representation for e.g. the computation of electrograms at a
limited number of sites (see e.g[118, 483]).
Using as a reference potential the average potential on the cardiac volume ˝H ,
Z
1
w.x; t/ D u.x; t/  u.; t/ d :
j˝H j ˝H

For x 2 ˝H [ ˝b , i.e. within the cardiac wall or in the extracardiac medium, the
full potential can be expressed by means of the following integral over the cardiac
domain
Z Z
w.x; t/ D JTv r  .; x/ d  D  .rv.; t//T Di ./r  .; x/ d ;
˝H ˝H
(9.5)
where  .; x/ is the solution of the following lead field problem
8
ˆ
ˆ O   D  1  ./ C ı.  x/  2 ˝H [ ˝b
ˆ
ˆ div Dr
ˆ
< j˝H j ˝H

ˆ O   ˙ D 0
ŒŒ  ˙ D 0; ŒŒ nT Dr 2˙
ˆ
ˆ
ˆ
:̂ nT Dr
O  D 0 on ;

with ˝H ./ the characteristic function of ˝H , and ı. x/ denotes the Dirac delta
function at point x. Note that the boundary condition imposed on the “lead field” 
actually reflects the property that w.x; t/, defined by (9.5), has zero average in the
heart domain.
9.6 Subendocardial Ischemia, ST Depression and Elevation 343

Using again as a reference the average potential on ˝H , the split field com-
ponents wa , wo , wc , associate with the sources Jc , Ja and Jo , respectively, can be
represented by means of the following surface integral on the ischemic boundary
Z
wc .x; t/ D ˛O D./n./  r  .; x/ d ;
i

Z
wa .x; t/ D ˇO .aTl ./n.//al ./  r  .; x/ d ;
i

Z
 
wo .x; t/ D O I  al ./aTl ./ n./  r  .; x/ d :
i

In the following, we will denote by full extracellular waveform w, by isotropic (IS)

extracellular waveform wIS D wc and by anisotropic (AS) extracellular waveform
the sum wAS D wa C wo , and we will refer to the splitting

w D wIS C wAS

as the decomposition of the full extracellular waveform into its isotropic and
anisotropic components.

9.6.2 Ischemic Simulations

We investigate the effects induced by the presence of an ischemic subendocardial

region on the extracellular potential distributions, the activation-repolarization
sequences and the action potential duration (APD).
A simplified left ventricular geometry is considered, defined as a set of packed
ellipsoidal surfaces truncated at the base and the apex, according to the parametric
equations

x D a.r/ cos  cos ; y D a.r/ cos  sin ; z D c.r/ sin :

Here r 2 Œ0; 1; 2 Œ min ; max ;  2 Œmin ; max ; a.r/ D a1 Cr.a2 a1 /; c.r/ D
c1 Cr.c2 c1 / and ai ; ci ; i D 1; 2 are given coefficients determining the main axes of
the ellipsoid. The intracavitary blood and the extracardiac bath are also modeled as
a set of packed ellipsoidal surfaces in contact with the endocardial and the epicardial
surface, respectively.
We have assumed homogeneous cellular membrane properties, i.e. all individual
cells have the same intrinsic transmembrane action potential, except in the ischemic
region. The three main pathophysiological conditions of myocardial ischemia are
344 9 Simulation Studies of Cardiac Bioelectrical Activity

elevated extracellular potassium, acidosis and anoxia, whose effects on transmem-

brane potential are elevation of resting potential and reduction of upstroke velocity
and APD. After the occlusion of a coronary artery, we can distinguish three stages
of ischemia: moderate, corresponding to the first 5–7 min, with an elevation of
extracellular potassium ŒK C o to about 10 mM; early severe, after 10–12 min, with
a second increase of ŒK C o ; and severe, characterized by the occurrence of gap
junction uncoupling and irreversible cell damage. In our study, we consider the first
and the second stages, and therefore we do not alter the conductance coefficients.
In the remainder of the section, we will refer to early severe ischemia as severe (or
acute) ischemia. In the LR1 model, we increase the value of extracellular potassium
ŒK C o from 5.4 mM (control) to 10.5 (moderate) and 18 mM (severe); see [569].
The resting transmembrane potentials are 84 mV (control), 70 mV (moderate)
and 55 mV (severe), while the APDs are 260 ms (control), 170 ms (moderate) and
70 ms (severe), see Fig. 9.53. For economy of space, we will report the details and
figures for the case of acute ischemia and only comment on the differences in the
case of moderate ischemia.
The fibers rotate intramurally linearly with the depth for a total amount of
120ı and when the point of view is from the epicardial side, the rotation is
counterclockwise (CCW) proceeding from the epicardium to endocardium. More
precisely, in the canonic reference system .ue 1 ; ue2 ; ue 3 /, the fiber direction al .x/
and the other two principal axes at a point x are given by

al .x/ D ue 1 cos ˛.r/ C ue 2 sin ˛.r/; at .x/ D ue 3 ; an .x/ D ue 1 sin ˛.r/  ue 2 cos ˛.r/
2 
˛.r/ D .1  r/  ; with r D x3 =c:
3 4
(9.6)

LR1
50
CONTROL
MODERATE
SEVERE

0
mV

−50

−100
0 50 100 150 200 250 300 350 400
ms

Fig. 9.53 Control, moderate and severe ischemic action potentials generated by 0-dimensional
LR1 model
9.6 Subendocardial Ischemia, ST Depression and Elevation 345

The conductivity tensors are assumed orthotropic with values

li D 3  103 ; ti D 3:1525  104 ; ni D 3:1525  105 ;

le D 2  103 ; te D 1:3514  103 ; ne D 6:757  104 .1 cm1 /:

We will also consider an isotropic conducting media modeling the cavitary blood
with conductivity coefficient b D 6  103 1 cm1 .
The macroscopic features of the excitation and subsequent repolarization process
are described by extracting from the spatio-temporal transmembrane potential
the sequence of the propagating excitation and repolarization wave fronts. In
particular, we define the excitation time te .x/ at a given point x as the unique time
when v.x; te .x// D 40 mV during the excitation phase. Analogously, during the
repolarization phase we define the recovery time as the unique time instant tr .x/
when v.x; tr .x// D 50 mV. In Figs. 9.54 and 9.55, we report the isochrone lines

ACTI REPO APD

120 350 240
100
80 235
EPI

300
60
40 230
20 250
36.76 105.22 264.54 332.39 225.88 231.10

120 350 240

100
80 235
MID

300
60
40 230
20
250
19.09 111.58 250.93 339.78 227.15 232.29

120 350 240

100
80
ENDO

235
300
60
40 230
20
250
0.50 124.54 238.34 351.09 225.73 240.92

Fig. 9.54 Contour plots of depolarization time (first column ACTI), repolarization time (second
column REPO) and action potential duration (third column APD) on epi, midwall and endocardial
surfaces in case of healthy tissue with multiple endocardial stimulation. Reported below each panel
are the maximum, minimum and step in ms of the displayed map
346 9 Simulation Studies of Cardiac Bioelectrical Activity

ACTI REPO APD

80 300
200
60
200
EPI

150
40
100
20 100
50
38.08 86.85 267.92 314.93 228.06 230.78

80 300
200
60
MID

200 150
40
100
20 100
50
17.35 75.98 187.37 305.17 166.83 235.00

80 300
200
60
ENDO

200 150
40
100
20 100
50
0.27 68.28 105.06 297.00 102.39 242.32

Fig. 9.55 Contour plots of depolarization time (first column ACTI), repolarization time (second
column REPO) and action potential duration (third column APD) on epi, midwall and endocardial
surfaces in case of an acute subendocardial ischemic region. Reported below each panel are the
maximum, minimum and step in ms of the displayed map

of the activation time te .x/ (first column, ACTI), repolarization time tr .x/ (second
column, REPO) and APD = tr .x/  te .x/ (third column) on the epicardium (first
row), midwall (second row) and endocardium (third row), for a healthy tissue with
multiple endocardial stimulation and for a tissue with a subendordial ischemic
region and central endocardial stimulation, respectively.
Extracellular potential distributions and electrograms: healthy tissue. We
report in Fig. 9.56 the decomposition of four epicardial electrograms and in Fig. 9.57
the decomposition of 16 electrocardiograms recorded 2 cm far away from the
epicardium, in case of normal tissue. We observe that all the isotropic components
(in blue) present a mainly positive QRS complex, while the anisotropic components
(in red) present a negative QRS complex. Due to the larger magnitude of the
anisotropic components, the resulting full extracellular/extracardiac waveforms (in
black) have a mainly negative biphasic QRS complex.
9.6 Subendocardial Ischemia, ST Depression and Elevation 347

5
mV

−5
5
mV

−5
0 400 0 400
ms ms

Fig. 9.56 Decomposition of four epicardial electrograms in case of normal tissue. Full extracellu-
lar waveform (black), isotropic component (blue) and anisotropic component (red)

Extracellular potential distributions and electrograms: ischemic tissue.

Figure 9.58 reports the decomposition of the epicardial extracellular potential
during the ST interval (140 ms) for different percentages (10, 25, 50 %) of the
ischemic region thickness with respect to the total transmural wall thickness. For
thin ischemic regions (10 and 25 % of the whole transmural thickness), the isotropic
component is dominant and the full extracellular potential field presents a minimum
(ST depression) on the epicardial projection of the subendocardial ischemic region.
Instead, in the case of thick ischemia (50 % and more), the anisotropic component
is dominant, because the lateral boundaries of the ischemic region are large, thus
the full extracellular potential presents a maximum (ST elevation) on the epicardial
projection of the ischemic region.
Figure 9.59 displays the epicardial electrograms calculated from a grid of 4  4
electrodes. Figure 9.60 shows the decomposition of the epicardial electrogram
above the subendocardial ischemic region for different percentages of the ischemic
region thickness. Again, if the subendocardial ischemic region is thicker, then the
anisotropic component is more dominant, yielding the ST elevation in the full
extracellular waveform.
348 9 Simulation Studies of Cardiac Bioelectrical Activity

3
mV

−3
3
mV

−3
3
mV

−3
3
mV

−3
0 400 0 400 0 400 0 400
ms ms ms ms

Fig. 9.57 Decomposition of 16 electrocardiograms recorded 2 cm far away from the epicardium
in case of normal tissue. Full extracardiac waveform (black), isotropic component (blue) and
anisotropic component (red)

9.7 Reentry Phenomena

Atrial or ventricular reentry is a type of arrhythmia where the electric signal is not
completing the normal circuit, but it rather follows an alternative circuit looping
back upon itself and developing a self-perpetuating rapid and abnormal activation,
also called circus movement. Necessary conditions for the onset of reentry include
a combination of unidirectional block and slowed conduction. Reentry is divided
into two major types: anatomical and functional reentry. The circus movement can
occur around an anatomical (e.g. an infarct region) or functional core. Either of two
types may occur alone, or together. In the following, we will focus on functional
reentry, and in particular on the causes of its induction.
A widely accepted theory for the induction of functional reentry in cardiac tissue
is based on a hypothesis proposed by Winfree in [561], known as the critical
point hypothesis, see also the review [431]. This hypothesis can be illustrated by
considering the so-called pinwheel experiment. Suppose that an action potential
wavefront, elicited by a brief electrical stimulus (S1), propagates from left to right
in a two-dimensional sheet of cardiac tissue, as shown in Fig. 9.61, Panels a–c.
9.7 Reentry Phenomena 349

FULL IS AS

1.5 1.5 1.5

1 1 1
50%

0.5 0.5 0.5

0 0 0
−0.5 −0.5 −0.5

−0.96 1.51 0.10 −0.44 0.60 0.10 −0.85 1.94 0.10

1.5 1.5 1.5

1 1 1
25%

0.5 0.5 0.5

0 0 0
−0.5 −0.5 −0.5

−0.44 0.92 0.10 −0.42 0.58 0.10 −0.45 0.35 0.10

1.5 1.5 1.5

1 1 1
10%

0.5 0.5 0.5

0 0 0
−0.5 −0.5 −0.5

−0.36 0.88 0.10 −0.36 0.56 0.10 −0.35 0.33 0.10

Fig. 9.58 Contour plots of the epicardial extracellular potential decomposition during the ST
interval (140 ms) for different percentages (10, 25, 50 %) of the subendocardial ischemic region
thickness with respect to the total transmural wall thickness. Reported below each panel are the
maximum, minimum and step in mV of the displayed map

During the plateau and repolarization phases of the action potential, the tissue is
refractory. The critical phase of the action potential is the time when the application
of a stimulus of appropriate strength induces reentry and it occurs during the so-
called vulnerable window, near the end of the refractory period.
Suppose now that a second brief point stimulus (S2) is applied at the center of
the sheet. If the S2 stimulus is applied early (see Fig. 9.61, Panel d), the tissue
under the stimulation electrode is still refractory and the stimulus does not elicit
a wavefront (see Fig. 9.61, Panel g). On the other hand, if the S2 stimulus is applied
late (see Fig. 9.61, Panel f), the tissue under the electrode has already recovered from
refractoriness and a closed wavefront propagating outward is initiated (see Fig. 9.61,
Panel i). However, if the S2 stimulus is applied at an intermediate time (see Fig. 9.61,
Panel e), when the critical phase is occurring in the tissue under the electrode, the
portion of tissue on the left of the electrode is excited, while that on the right is still
too refractory to be excited. As a result, a wavefront is propagating outward, but it is
not closed (see Fig. 9.61, Panel h). It ends at two points where the contours of critical
350 9 Simulation Studies of Cardiac Bioelectrical Activity

10 10 10 10

0 0 0 0

−10 −10 −10 −10

0 400 0 400 0 400 0 400

10 10 10 10

0 0 0 0

−10 −10 −10 −10

0 400 0 400 0 400 0 400

10 10 10 10

0 0 0 0

−10 −10 −10 −10

0 400 0 400 0 400 0 400

10 10 10 10

0 0 0 0

−10 −10 −10 −10

0 400 0 400 0 400 0 400

Fig. 9.59 Epicardial electrograms calculated from a grid of 4 4 electrodes in the case of an acute
subendocardial ischemic region 50 % thick

10% 25% 50%

5 5 5
mV

0 0 0

−5 −5 −5
0 200 400 0 200 400 0 200 400
ms ms ms

Fig. 9.60 Decomposition of the epicardial electrogram above the subendocardial ischemic region
for different percentages (10, 25, 50 %) of the ischemic region thickness with respect to the total
transmural wall thickness. Full extracellular waveform (black), isotropic component (blue) and
anisotropic component (red)
9.7 Reentry Phenomena 351

a t = 20 ms
50
b t = 40 ms
50
c t = 60 ms
50

0 0 0

−50 −50 −50

−100 −100 −100

−81.72 13.45 5.00 −81.74 13.18 5.00 −81.73 12.93 5.00

d t = 265 ms
50
e t = 285 ms
50
f t = 305 ms
50

0 0 0

−50 −50 −50

−100 −100 −100

−81.02 197.86 5.00 −81.54 197.90 5.00 −81.62 197.92 5.00

g t = 320 ms h t = 320 ms i t = 320 ms

50 50 50

0 0 0

−50 −50 −50

−100 −100 −100

−81.65 −76.46 5.00 −81.65 10.62 5.00 −81.65 10.20 5.00

Fig. 9.61 The pinwheel experiment on a two-dimensional sheet of cardiac tissue. Panels a-b-c:
transmembrane potential patterns at three instants during the propagation of the planar excitation
wavefront elicited by the S1 stimulation. Panels d-e-f: transmembrane potential patterns during an
early (d), intermediate (e) and late (f) S2 central point stimulation. Panels g-h-i: transmembrane
potential patterns some instants after the onset of the early (g), intermediate (h) and late (i) S2
stimulations. Below each panel are reported the minimum, maximum and step in mV of the
displayed map and the colorbar denotes the range of values of the displayed equipotential lines

phase and critical depolarization intersect. Frazier et al. in [185] refer to these points
as critical points. As time proceeds further, the two critical points become the pivots
of two spiral excitation waves. In the cardiac electrophysiology literature, spiral
waves are also called rotors and the particular type of reentry described above is
called figure of eight reentry.
The critical point hypothesis has been verified experimentally by Shibata et al.
[476], who performed the pinwheel experiment in a dog heart. The critical point
hypothesis also inspired an experimental method for the induction of a single rotor in
the heart, known as cross-field stimulation. In this method, two electric stimulations,
S1 and S2, are delivered with an appropriate delay through two lines of electrodes
352 9 Simulation Studies of Cardiac Bioelectrical Activity

aligned perpendicular to each other, see e.g. [185], where cross-field stimulation
was used to initiate rotors in a dog heart.
The pinwheel experiments [476] and the cross-field stimulation experiments
[185] validated strongly the critical point hypothesis, which now, thanks to this
experimental evidence, is usually referred to as critical point theory.
In the pinwheel experiment, a pair of rotors is induced by applying two
successive stimuli, S1 and S2, at different locations. Suppose now that both S1 and
S2 are applied at the same site, so that the S1 critical phase contour and the S2
stimulus contour are both circles centered at the same point. Because two concentric
circles never intersect, the critical point theory predicts that reentry cannot occur.
However, Matta et al. in [322] performed this experiment and initiated reentry.
A possible explanation of this paradox is based on the anisotropic Bidomain
model, and in particular on the virtual electrode polarization pattern due to the
unequal anisotropy rations of the intra- and extracellular media. In [429], Roth
showed by Bidomain numerical simulations that an S1-S2 stimulation protocol
through the same unipolar electrode might induce reentry.
Three-dimensional reentry simulations. We now turn to the simulation of
reentry phenomena in three dimensions, where the possible configurations of
reentrant fronts are much more complex and less understood than in two dimensions
(three-dimensional rotors are often called scroll waves). There is a vast literature on
cardiac reentry simulations and experiments; among the several simulation works,
we refer the interested reader to e.g. [49, 98, 170, 171, 191, 287, 345, 368, 411, 412,
431, 450, 488, 519, 562, 569], Part VII of [587, 588], and the references therein. Due
to the high computational complexity of large scale simulations, virtually all works
in the vast existing literature on cardiac reentry simulations employ some model
simplifications in order to obtain a tractable discrete problem see e.g. [278], [589,
Part V and VII], [369]. The most used simplifications are in the cardiac tissue model
(Monodomain instead of Bidomain, fiber structure without intramural rotation or
full anisotropy), in the ionic model (FHN or intermediate variations instead of
LR models), in the domain (two instead of three dimensions), in the numerical
methods (operator splitting instead of the fully coupled Bidomain system, coarse
time and/or space mesh sizes compared to the length scale of the problem), etc. On
the contrary, we retain here the full complexity of the coupled Bidomain–LR1 model
with intramural fiber rotation and orthotropic anisotropy in three dimensions and we
compare the resulting scroll waves with the one obtained with the Monodomain–
LR1 model. With the latter, we then simulate scroll wave breakups on larger
domains and counterrotating scroll waves on half ventricle.

9.7.1 Stable Scroll Waves

We start with the simulation of a stable scroll wave using both the anisotropic
Bidomain–LR1 and the Monodomain–LR1 models. The conductivity tensors are
assumed axisymmetric with values
9.7 Reentry Phenomena 353

li D 3  103 ; ti D 3:1525  104; le D 2  103 ; te D 1:3514  103 .1 cm1 /

in the Bidomain case and

l D 1:2  103 ; t D 2:5  104 .1 cm1 /

in the Monodomain case.

The original LR1 model is modified in order to shorten the APD according to
Garfinkel et al. [191], setting GNa D 16; GK D 0:432; Gsi D 0. We have assumed
throughout the slab homogeneous intrinsic cellular membrane properties. Due to
the high computational costs of the Bidomain model, we limit our simulation to a
cartesian slab of dimensions 2  2  0:5 cm3 , discretized with 200  200  50 finite
elements.
The intramural fibers rotate linearly with depth for a total amount of 90ı ,
i.e. 18ı =mm, starting from 90ı (0ı ) on the lower-endocardial (upper-epicardial)
surface of the slab with respect to a side on the slab.
Reentry is initiated with a cross gradient procedure, applying at first an impulse
of 200 A=cm3 for 1 ms along one of the main intramural sides of the slab,
generating initially a plane wave, and then eliciting an orthogonal front by applying
a second impulse at a proper time (in this case t D 68:4 ms) in the bottom-left
quarter volume of the domain. This cross gradient stimulation elicites a vortex-like
pattern, usually called scroll wave, rotating around a tube-like filament which is a
3D analog of the core of a spiral wave . The simulation was run for 200 ms at a fixed
time step size of 0:05 ms.
The time evolution at a given point of the transmembrane potential v and of the
LR1 variables during a reentry elicited by a stable scroll wave is shown in Fig. 9.62.
The contour plots of the resulting scroll waves are shown in Fig. 9.63 in the
Bidomain case and in Fig. 9.64 in the Monodomain case, from t D 90 to t D
200 ms, every 10 ms. The colormap of the transmembrane potential distribution
ranges from blue (resting values around 84 mV) to red (excitation front around
10 mV). The effect of the anisotropy appears stronger in the Bidomain model, as
shown by the more elongated spirals on the epicardium and by the more twisted
scroll waves in the intramural sections. Moreover, the meandering of the epicardial
spiral tip follows a more regular trajectory in the Monodomain case than in the
Bidomain case. Nevertheless, the plots in Fig. 9.65 of the transmembrane potential
v in the two models, at a given point as a function of time, show very similar
propagation speed and frequency, as the curves appear to be just shifted in time.

9.7.2 Scroll Waves Breakup

The breakup of spiral and scroll waves has been intensely studied in recent
years because it could underlie ventricular fibrillation according to some authors
354 9 Simulation Studies of Cardiac Bioelectrical Activity

50
1

m (w1)
0
0.5
V

−50
0
0 200 400 600 800 0 200 400 600 800

1 1
h (w2)

j (w3)
0.5 0.5

0 0
0 200 400 600 800 0 200 400 600 800
0.1 1
d (w4)

f (w5)
0.05 0.5

0 0
0 200 400 600 800 0 200 400 600 800
−5
x 10
0.2
10
Cai (w7)
X (w )
6

0.1 8

6
0
4
0 200 400 600 800 0 200 400 600 800

Fig. 9.62 Time course of the transmembrane potential v, of the LR1 gating variables
m; h; j; d; f; X and of the Ca2C concentration during reentry determined by a stable scroll wave
(800 ms)

and it still remains a complex phenomenon with multiple causes not yet fully
understood, see e.g. Fenton et al. [171], Garfinkel et al. [191]. Among the possible
factors contributing to the breakup of the excitation front and to the instability of
reentry circuits, many researchers indicated steep APD restitution curve, intramural
rotational anisotropy, tissue thickness, filament twist, etc.
We simulated a spiral and scroll wave breakup using the axisymmetric
anisotropic Monodomain with the conductivity coefficients

l D 1:2  103 ; t D 2:5  104 .1 cm1 /;

and the LR1 model with GNa D 16; GK D 0:432; as before, but setting
the conductance of the slow-inward current Gsi D 0:056. It has been shown
that increasing this parameter increases the meandering of the scroll filament and
eventually leads to a breakup regime if the cardiac tissue is thick enough; see
Garfinkel et al. [191].
9.7 Reentry Phenomena 355

Fig. 9.63 Stable scroll wave with Bidomain–LR1 model. Domain: 2  2  0:5 cm3 ; 4x D
0:01; Gsi D 0, t D 90, 100, 110, 120, 130, 140, 150, 160, 170, 180, 190, 200 ms. The colormap
ranges from blue (resting values around 84 mV) to red (excitation front around 10 mV)

Next, we consider scroll waves breakup in three dimensions. In order to contain

the increased meandering of the filament inside the computational domain, we
consider a three-dimensional domain of size 6  6  0:6 cm3 and consider a high
degree of intramural fiber rotation, for a total amount of 120ı. The spatial mesh
consists of 400  400  40 finite elements and the simulation is run up to 1 300 ms
(26 000 time steps with t D 0:05 ms), starting with the same cross gradient
stimulation as before. The results displayed in Fig. 9.66 focus on three moments
356 9 Simulation Studies of Cardiac Bioelectrical Activity

Fig. 9.64 Stable scroll wave with Monodomain–LR1 model. Domain: 2  2  0:5 cm3 ; 4x D
0:01; Gsi D 0, t D 90, 100, 110, 120, 130, 140, 150, 160, 170, 180, 190, 200 ms. The colormap
ranges from blue (resting values around 84 mV) to red (excitation front around 10 mV)

of the front evolution at differently spaced time intervals: (a) just after the first
breakup (first rows, t D 500; 505; 510 ms, second row, t D 520; 530; 540 ms) due
to a head-to-tail collision and generating two new spiral tips, one drifting outside
the domain and the other, near the center of the slab, generating a new scroll wave;
(b) a subsequent breakup (third row, t D 705; 715; 725 ms) generating additional
scroll waves and spiral tips; (c) a later time with many broken fronts (last row,
t D 1 250; 1 275; 1 300ms) displaying a spatio-temporal chaotic configuration of
multiple wave reentry. Figure 9.67 shows the approximated scroll filaments at an
early time after reentry initiation (t D 265 ms, left panel) when only one filament
9.7 Reentry Phenomena 357

a b
10 0

0 −10

−10 −20

−20
−30
−30
−40
−40
−50
−50
−60
−60

−70 −70

−80 −80

50 100 150 200 50 100 150 200

Fig. 9.65 Transmembrane potential at a given point as a function of time for the two tests of
Fig. 9.63 ((a), Bidomain model) and Fig. 9.64 ((b), Monodomain model).

is present and almost at the end of our simulation (t D 1 207 ms) when many
filaments have appeared (for clarity, the slab height has been magnified). Figure 9.68
compares the time course of the transmembrane potential v of the Monodomain–
LR1 model for the stable scroll wave of Fig. 9.64 (left) and for the unstable scroll
wave with breakups of Fig. 9.66 (right).

9.7.3 Scroll Waves in Ellipsoidal Geometry

We now extend our Monodomain–LR1 simulations to the more realistic geometry

for half a ventricle modeled by a truncated half ellipsoid (with parameters described
at the beginning of this section), discretized with 500  500  80 isoparametric
finite elements. The original LR1 model is modified as described before in order
to shorten the APD (GNa D 16; GK D 0:432; Gsi D 0); in addition, we set
GK1 D 0:6047 and scale the time constants d and f by a factor 10. In order to see
more clearly the effect of the curved geometry, we considered in this case parallel
fibers (i.e. no fiber rotation). Reentry is initiated with a broken wave procedure,
where at time t D 0 we set v D 10 mV on a vertical intramural section running
from epi to endocardium and from the bottom to about 3=4 of the ventricle height
(see the first panel of Fig. 9.69); moreover, on another vertical section on the right
side of the previous one, we set at t D 0 the gating variables w to their steady state
corresponding to the fixed value v D 10 mV, which is in the refractory phase of the
action potential. In this way, an excitation front starts from the vertical section where
v D 10 mV, but on the right side the front is blocked by the other section where the
gating variables are inhibiting propagation because they are in their refractory phase.
Therefore, the front curls around the upper end of the sections and originates a scroll
wave. Unexpectedly, the front also curls around the bottom end of the sections,
possibly because of the high curvature of the domain geometry there, resulting in
358 9 Simulation Studies of Cardiac Bioelectrical Activity

Fig. 9.66 Scroll wave breakup with Monodomain–LR1 model, Gsi D 0:056, domain: 6  6 
0:6 cm3 . First row: t D 500, 505, 510; second row: t D 520, 530, 540; third row: t D 705, 715,
725; fourth row: t D 1 250, 1 275, 1 300 ms. The colormap ranges from blue (resting values around
84 mV) to red (excitation front around 10 mV)

a second counterrotating scroll wave. The evolution of these two scroll waves is
displayed in Fig. 9.69, where after an initial adjustment, the two scroll waves seem
to reach a stable counterrotating configuration (last panel, t D 500 ms), also known
as figure-8 or double loop reentry, see e.g. Wit and Janse [564].
9.7 Reentry Phenomena 359

a b

6 6

0.6 0.6

6 6

0 0

Fig. 9.67 Scroll wave filaments for the breakup test of Fig. 9.66 near reentry initiation ((a), t D
265 ms) and during breakup regime ((b), t D 1 297 ms). Monodomain–LR1 model with Gsi D
0:056. Domain: 6  6  0:6 cm3 (height of the slab has been magnified)

10
0
0

−10
−10

−20
−20

−30
−30

−40
−40

−50
−50

−60 −60

−70 −70

−80 −80

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
3 3
x 10 x 10

Fig. 9.68 Comparison of the transmembrane potential v at a point as a function of time (0  t 

1 000 ms) for the Monodomain–LR1 model. Left: stable scroll wave on a domain 2  2  0:5 cm3 .
Right: scroll wave breakup on a domain 6  6  0:6 cm3
360 9 Simulation Studies of Cardiac Bioelectrical Activity

T = 25
T = 0.25

13.55 22.366

−20.0931 −14.3799

−53.7362 −51.1257

−84.015 −84.197

T = 50 T = 100

23.403 21.216

−13.6991 −15.1299

−50.8011 −51.4757

−84.193 −84.187

T = 300 T = 500

19.886 27.703

−15.9061 −10.6991

−51.6981 −49.1011

−83.911 −83.663

Fig. 9.69 Counterrotating double scroll wave with the Monodomain–LR1 model on half ventricle
(500  500  80 elements). Distribution of v on the endocardium and intramural sections at times
t D 0:25, 25, 50, 100, 300, 500 ms. Viewpoint is inside the ventricle
Appendix A
Cardiac Simulation Projects, Software,
Libraries

We collect in this Appendix some of the available cardiac simulation projects,

software, and libraries that we encountered while writing this book. This is only
a partial list, due to both our involuntary omissions and to the vast and rapidly
changing field we tried to survey. We apologize for the fact that some links, sites
and references provided might change with time and become outdated.

A.1 IUPS Physiome Project

The Physiome Project (http://www.physiome.org.nz/) is a worldwide public domain

effort to provide a computational framework for understanding human and other
eukaryotic physiology. It aims to develop integrative models at all levels of
biological organization, from genes to the whole organism via gene regulatory
networks, protein pathways, integrative cell function, and tissue and whole organ
structure/function relations. Current projects include the development of:
• Ontologies to organize biological knowledge and access to databases;
• Markup languages to encode models of biological structure and function in a
standard format for sharing between different application programs and for re-
use as components of more comprehensive models;
• Databases of structure at the cell, tissue and organ levels;
• Software to render computational models of cell function such as ion channel
electrophysiology, cell signalling and metabolic pathways, transport, motility, the
cell cycle, etc. in 2 and 3D graphical form;
• Software for displaying and interacting with the organ models which will allow
the user to move across all spatial scales.
Under the wider umbrella program of the Physiome Project, there is also The
Wellcome Trust Heart Physiome Project (http://www.physiome.ox.ac.uk/), a 5-year

© Springer International Publishing Switzerland 2014 361

P. Colli Franzone et al., Mathematical Cardiac Electrophysiology, MS&A 13,
DOI 10.1007/978-3-319-04801-7
362 A Cardiac Simulation Projects, Software, Libraries

international collaborative effort between the universities of Auckland and Oxford

to develop a multi-scale modelling framework for the heart that can be used for
addressing a wide range of scientific and clinical questions.

A.2 Virtual Physiological Human (VPH)

The VPH Initiative (VPH-I), http://www.vph-noe.eu/, was a project funded by

the European Framework 7 (FP7) in 2008, aimed to support and foster European
research in biomedical modeling and simulation of the human body. It has lead to the
creation of the VPH Network of Excellence (VPH NoE), three integrated projects
(IPs), nine specific targeted research projects (STREPs) and two cooperative actions
(CAs). The VPH Network of Excellence (VPH NoE) was designed to foster,
harmonise and integrate pan-European research in the field of (i) patient-specific
computer models for personalised and predictive healthcare and (ii) ICT-based tools
for modeling and simulation of human physiology and disease-related processes.
In 2011, the Virtual Physiological Human Institute for Integrative Biomed-
ical Research, in short VPH Institute (http://www.vph-institute.org/what-is-vph-
institute.html) was funded and started operating. The VPH Institute is an inter-
national non-profit organisation, whose mission is to ensure that the Virtual
Physiological Human is fully realised, universally adopted, and effectively used both
in research and clinic. VPH activities also include support for integrative research (in
particular development of the VPH ToolKit http://toolkit.vph-noe.eu/), training and
dissemination activities, networking activities (in particular the VPH Conferences);
see also the review papers by Hunter et al. [247, 248].

A.3 NSR Physiome

NSR, the National Simulation Resource (http://www.physiome.org/) at the Univer-

sity of Washington, Department of Bioengineering, provides the following tools in
support of the Physiome Project:
• The JSim Modeling System and Related Tools (http://www.physiome.org/jsim/).
This is a Java-based simulation system for building quantitative numeric models
and analyzing them with respect to experimental reference data. With primary
focus in physiology and biomedicine, JSim computational engine may intermix
ODEs, PDEs, implicit equations, integrals, summations, discrete events and
procedural code.
• The NSR Physiome Model Repository (http://www.physiome.org/Models/), with
Tutorials, Standards and a Consolidated Model Database.
• The Virtual Physiological Rat Project (http://virtualrat.org/).
• The Systems Biology Workbench (SBW) (http://sbw.sourceforge.net/).
A.4 Other Simulation Software and Modeling Environments 363

A.4 Other Simulation Software and Modeling Environments

• Berkeley Madonna (www.berkeleymadonna.com/), developed at the University

of California, Berkeley, CA, USA.
• Biomed Town (www.biomedtown.org) is an on-line community open and free
to anyone has a professional or educational interest in biomedical research
and practice. Started in 2006 as support of the coordination action STEP, it
slowly became the de facto home for various biomedical research initiatives, in
particular those related to the so-called integrative research, an umbrella concept
that includes physiome, system biology, and multiscale modeling. etc.
• BIONT: Modeling Program for the Complex Biochemical Systems
(members.tripod.com/mitoart/permtran/biont.htm).
• BioSpice – Arkin Lab (biospice.lbl.gov/), developed at the Lawrence Berkeley
National Laboratory, University of California, Berkeley, USA.
• BISEN: The Biochemical Simulation Environment
(june.phys.mcw.edu/BioWiki/index.php/BISEN), developed at the BioTech Cen-
ter, Medical College of Wisconsin, Milwaukee, WI, USA.
• Chaste: Cancer, Heart and Soft Tissue Environment (www.cs.ox.ac.uk/chaste/)
(see also [331]), is a general purpose simulation package aimed at multi-scale,
computationally demanding problems arising in biology and physiology. Current
functionality includes tissue and cell level electrophysiology, discrete tissue
modeling, and soft tissue modeling. The package is being developed by a team
mainly based in the Computational Biology Group at the Department of Com-
puter Science, University of Oxford, and development draws on expertise from
software engineering, high performance computing, mathematical modeling and
scientific computing.
• CMISS (www.cmiss.org/), developed at the University of Auckland, Auckland,
NZ.
• Continuity 6 (cmrg.ucsd.edu/Continuity), developed at the University of Califor-
nia, San Diego, CA, USA.
• E-Cell System (www.e-cell.org/software/), developed at the Keio University,
Tokyo, Japan.
• GENESIS: GEneral NEural SImulation System (genesis-sim.org/), is a general
purpose simulation platform that was developed to support the simulation of
neural systems ranging from subcellular components and biochemical reactions
to complex models of single neurons, simulations of large networks, and systems-
level models.
• Gepasi: computer simulation of biochemistry (www.gepasi.org/).
• LifeV (http://www.lifev.org/) is a finite element (FE) library providing imple-
mentations of state of the art mathematical and numerical methods for biomedical
research and production simulations, such as fluid structure interaction and mass
transport. LifeV is the joint collaboration between EPFL (CMCS), Politecnico di
Milano (MOX), INRIA (REO, ESTIME) and Emory University.
364 A Cardiac Simulation Projects, Software, Libraries

• SCIRun: Problem solving environment for modeling, simulation and visu-

alization of scientific problems (www.sci.utah.edu/cibc/software/index.html),
developed at the NCRR Center for Integrative Biomedical Computing, University
of Utah, USA.
• SimTK simulation and modeling resources (simtk.org/xml/index.xml) is a soft-
ware framework initiated and developed by Simbios, the National NIH Center
for Biomedical Computing focusing on Physics-based Simulation of Biological
Structures.
• Simbiome simulation resources (simbiome.org/): Resources for physics-based
simulation of biomedical structures.
• The Virtual Heart (thevirtualheart.org/), with many interactive java applets and
movies, mantained and developed by Flavio Fenton and Elizabeth Cherry.
• VHM: Penn Virtual Heart Model (www.seas.upenn.edu/~zhihaoj/VHM.html).
• Virtual Cell Modeling and Analysis Software (vcell.org/index.html), developed
by the National Resource for Cell Analysis and Modeling (NRCAM) University
of Connecticut Health Center.
• XPPAUT (www.math.pitt.edu/~bard/xpp/xpp.html): a software tool for simulat-
ing, animating, and analyzing dynamical systems based on differential equations,
as well as difference, delay, functional, and stochastic equations.

A.5 Some Related Monographs

[47] Bers, D.M.: Excitation-Contraction Coupling and Cardiac Contractile Force,

2nd edn. Kluwer, Dordrecht (2001)
[56] Beuter, A., Glass, L., Mackey, M.C., Titcombe, M.S.: Nonlinear Dynamics
in Physiology and Medicine. Springer, New York (2003)
[63] Britton, N.F.: Reaction – Diffusion Equations and Their Applications to
Biology. Academic, London (1986)
[71] Cabo, C., Rosenbaum, D.: Quantitative Cardiac Electrophysiology. Marcel
Dekker, New York (2002)
[78] Carmeliet, E., Vereecke, J.: Cardiac Cellular Electrophysiology. Kluwer,
Dordrecht (2002)
[150] Doi, S., Inoue, J., Pan, Z., Tsumoto, K.: Computational Electrophysiology.
Springer, Tokyo (2010)
[158] Efimov, I.E., Kroll, M.W., Tcho, P.J. (eds.): Cardiac Bioelectric Therapy.
Springer, New York (2009)
[168] Fall, C.P., Marland, E.S., Wagner, J.M., Tyson, J.J.: Computational Cell
Biology, 3rd edn. Springer, New York (2005)
[215] Gulrajani, R.M.: Bioelectricity and Biomagnetism. Wiley, New York
(1998)
[250] Jack, J.J.B., Noble, D., Tsien, R.W.: Electric Current Flow in Excitable
Cells. Oxford University Press, Oxford (1975)
A.5 Some Related Monographs 365

[261] Johnston, P.: Computational Inverse Problems in Electrocardiography.

WIT, Southampton (2001)
[273] Keener, J.P., Sneyd, J.: Mathematical Physiology, 2nd edn. Springer,
New York (2008)
[274] Kerckhoffs, R.C.P. (ed.): Patient-Specific Modeling of the Cardiovascular
System: Technology-Driven Personalized Medicine. Springer, New York (2010)
[283] Kogan, B.Y.: Introduction to Computational Cardiology. Springer,
New York (2010)
[284] Kohl, P., Sachs, F., Franz, M.R.: Cardiac Mechano-Electric Coupling and
Arrhythmias. Oxford University Press, Oxford (2011)
[312] Macfarlane, P.W., van Oosterom, A., Janse, M., Kligfield, P., Camm,
J., Pahlm, O. (eds.): Basic Electrocardiology. Cardiac Electrophysiology, ECG
Systems and Mathematical Modeling. Springer, New York (2012)
[317] Malmivuo, J., Plonsey, R.: Bioelectromagnetism. Oxford University Press,
Oxford (1995)
[369] Panfilov, A.V., Holden, A.V.: Computational Biology of the Heart. Wiley,
New York (1997)
[383] Peskin, C.S.: Mathematical Aspects of Heart Physiology. Lecture Notes of
the Courant Institute of Mathematical Sciences, New York University, New York
(1975)
[393] Plonsey, R., Barr, R.C.: Bioelectricity: A Quantitative Approach. Springer,
New York (2007)
[406] Pullan, A.J., Buist, M.L., Cheng, L.K.: Mathematically Modelling the
Electrical Activity of the Heart. World Scientific, Singapore (2005)
[446] Sachse, F.B.: Computational Cardiology. Modeling of Anatomy, Electro-
physiology, and Mechanics. LNCS, vol. 2966. Springer, Berlin (2004)
[481] Sigg, D.C., Iaizzo, P.A., Xiao, Y.-F., He, B.: Cardiac Electrophysiology
Methods and Models. Springer, New York (2010)
[500] Sundnes, J., Lines, G.T., Cai, X., Nielsen, B.F., Mardal, K.-A., Tveito, A.:
Computing the Electrical Activity of the Heart. Springer, Berlin (2006)
[562] Winfree, A.T.: The Geometry of Biological Time, 2nd edn. Springer,
New York (2001)
[586] Zipes, D., Jalife, J.: Cardiac Electrophysiology, 2nd edn. W. B. Saunders,
Philadelphia (1995)
[587] Zipes, D., Jalife, J.: Cardiac Electrophysiology, 3rd edn. W. B. Saunders,
Philadelphia (2000)
[588] Zipes, D., Jalife, J.: Cardiac Electrophysiology: From Cell to Bedside, 4th
edn. W. B. Saunders, Philadelphia (2004)
[589] Zipes, D., Jalife, J.: Cardiac Electrophysiology, 5th edn. W. B. Saunders,
Philadelphia (2009)
[590] Zipes, D., Jalife, J.: Cardiac Electrophysiology: From Cell to Bedside, 6th
edn. W. B. Saunders, Philadelphia (2013)
366 A Cardiac Simulation Projects, Software, Libraries

A.6 Physical Units and Constants

Table A.1 Physical quantities and related unit measures

Quantity Unit measure SI unit symbol Equivalent unit measure
Length Meter m
Mass Kilogram kg
Time Second s
Electric current Ampere A
Temperature Kelvin K
Amount of substance Mole mol
Frequency Hertz Hz s1
Force Newton N kg m s2
Pressure Pascal Pa N m2
Electric charge Coulomb C As
Energy Joule J Nm
Electric potential Volt V J C1
Electric resistance Ohm  V A1
Electric conductance Siemens S 1
Electric capacitance Farad F C V1
Electric resistivity m
Electric conductivity ( m)1
Electric current density A m2
Electric charge density C m3
Concentration of substance Molar M 103 mol m3

Table A.2 Physical constants

Name Symbol Value Dimension
Avogadro’s number NA 6:022  1023 mol1
Faraday’s constant F 9:64853  104 C mol1
Gas constant R 8:31446 J K1 mol1
Boltzmann’s constant kB 1:3806488  1023 J K1
Elementary charge e 1:602176  1019 C
Permittivity in vacuum 0 8:8542  1012 A s V1 m1
References

1. Aliev, R.R., Panfilov, A.V.: A simple two-variable model of cardiac excitation. Chaos Solitons
Fract. 7, 293–301 (1996)
2. Alistair, Y., Frangi, A.: Computational cardiac atlases: from patient to population and back.
Exp. Physiol. 94(5), 578–596 (2009)
3. Amato, S., Bellettini, G., Paolini, M.: The nonlinear multidomain model: a new formal
asymptotic analysis. In: Chambolle, A., Novaga, M., Valdinoci, E. (eds.) Proceedings of
Geometric Partial Differential Equations, pp. 33–74. Edizioni della Normale, Pisa (2013)
4. Ambrosio, L., Colli Franzone, P., Savaré, G.: On the asymptotic behaviour of anisotropic
energies arising in the cardiac bidomain model. Interfaces Free Bound. 2(3), 213–266 (2000)
5. Ambrosio, L., Fusco, N., Pallara, D.: Functions of Bounded Variation and Free Discontinuity
Problems. Oxford University Press, Oxford (2000)
6. Ambrosio, L., Gigli, N., Savaré, G.: Gradient flows in metric spaces and in the space of
probability measures. Lectures in Mathematicsi, ETH Zuerich, 2nd edn. Birkhäuser, Basel
(2008)
7. Antzelevitch, C., Fish, J.: Electrical heterogeneity within the ventricular wall. Basic Res.
Cardiol. 96, 517–527 (2001)
8. Antzelevitch, C., Sicouri, S., Lukas, A., Nesterenko, V.V., Lu, D.-W., Di Diego, J.M.:
Regional differences in the electrophysiology of ventricular cells: physiological and clinical
implications. In: Zipes, D., Jalife, J. (eds.) Cardiac Electrophysiology, chap. 23, pp. 228–245.
W. B. Sauders, Philadelphia (1995)
9. Anyukhovsky, E.P., Sosunov, E.A., Rosen, M.R.: Regional differences in electrophysiological
properties of epicardium, midmyocardium and endocardium. Circulation 94, 1981–1988
(1996)
10. Anyukhovsky, E.P., Sosunov, E.A., Gainullin, R.Z., Rosen, M.R.: The controversial M cell. J.
Cardiovasc. Electrophysiol. 10, 244–260 (1999)
11. Arevalo, H., Rodriguez, B., Trayanova, N.A.: Arrhythmogenesis in the heart: multiscale mod-
eling of the effects of defibrillation shocks and the role of electrophysiological heterogeneity.
Chaos 17(1), 015103 (2007)
12. Arisi, G., Macchi, E., Corradi, C., Lux, R.L., Taccardi, B.: Epicardial excitation during
ventricular pacing. Relative independence of breakthrough sites from excitation sequence in
canine right ventricle. Circ. Res. 71, 840–849 (1992)
13. Arthurs, C.J., Bishop, M.J., Kay, D.: Efficient simulation of cardiac electrical propagation
using high order finite elements. J. Comput. Phys. 231(10), 3946–3962 (2012)
14. Arthurs, C.J., Bishop, M.J., Kay, D.: Efficient simulation of cardiac electrical propagation
using high order finite elements. II: adaptive p-version. J. Comput. Phys. 253, 443–470 (2013)

© Springer International Publishing Switzerland 2014 367

P. Colli Franzone et al., Mathematical Cardiac Electrophysiology, MS&A 13,
DOI 10.1007/978-3-319-04801-7
368 References

15. Asanov, G.S.: Finsler Geometry, Relativity and Gauge Theories. D. Reidel, Dordrecht (1985)
16. Ashihara, T., Trayanova, N.A.: Cell and tissue responses to electric shocks. Europace 7(s2),
S155–S165 (2005)
17. Aslanidi, O.V., Colman, M.A., Stott, J., Dobrzynski, H., Boyett, M.R., Holden, A.V.,
Zhang, H.: 3D virtual human atria: a computational platform for studying clinical atrial
fibrillation. Prog. Biophys. Mol. Biol. 107, 156–168 (2011)
18. Attin, M., Clusin, W.T.: Basic concepts of optical mapping techniques in cardiac electrophys-
iology. Biol. Res. Nurs. 11(2), 195–207 (2009)
19. Austin, T.M., Trew, M.L., Pullan, A.J.: Solving the cardiac Bidomain equations for discon-
tinuous conductivities. IEEE Trans. Biomed. Eng. 53(7), 1265–1272 (2006)
20. Axelsson, O.: Iterative Solution Methods. Cambridge University Press, Cambridge (1994)
21. Ayache, N., Delingette, H., Sermesant, M. (eds.): Proceedings of the 5th International
Conference on Functional Imaging and Modeling of the Heart, FIMH’09, Nice, 3–5 June
2009. LNCS, vol. 5528. Springer, Berlin (2009)
22. Azzouzi, A., Coudiere, Y., Turpault, R., Zemzemi, N.: A mathematical model of the Purkinje-
muscle junctions. Math. Biosci. Eng. 8(4), 915–930 (2011)
23. Backhvalov, N., Panasenko, G.: Homogenization: Averaging Processes in Periodic Media.
Kluwer Academic, Dordrecht (1990)
24. Balay, S., Buschelman, K., Gropp, W.D., Kaushik, D., Knepley, M., Curfman McInnes, L.,
Smith, B.F., Zhang, H.: PETSc home page. http://www.mcs.anl.gov/petsc (2001)
25. Balay, S., Buschelman, K., Gropp, W.D., Kaushik, D., Knepley, M., Curfman McInnes,
L., Smith, B.F., Zhang, H.: PETSc users manual. Tech. Rep. ANL-95/11 – Revision 2.1.5,
Argonne National Laboratory (2002)
26. Barcilon, V., Chen, D.-P., Eisenberg, R.S., Jerome, J.W.: Qualitative properties of steady-
state Poisson-Nernst-Planck systems: perturbation and simulation study. SIAM J. Appl. Math.
57(3), 631–648 (1997)
27. Barles, G.: Fully non-linear Neumann type boundary conditions for second-order elliptic and
parabolic equations. J. Differ. Equ. 106, 90–106 (1993)
28. Barr, R.C., Plonsey, R.: Propagation of excitation in idealized two-dimensional tissue.
Biophys. J. 45, 1191–1202 (1984)
29. Barr, R.C., Spach, M.S.: Inverse calculation of QRS-T epicardial potentials from body surface
potential distributions for normal and ectopic beats in the intact dog. Circ. Res. 42, 661–675
(1978)
30. Baruffi, S., Spaggiari, S., Stilli, D., Musso, E., Taccardi, B.: The importance of fiber
orientation in determining the features of cardiac electric field. In: Antaloczy, Z. (ed.) Modern
Electrocardiology, pp. 89–92. Excerpta Medica, Amsterdam (1978)
31. Baruscotti, M., Barbuti, A., Bucchi, A.: The cardiac pacemaker current. J. Mol. Cell. Cardiol.
48, 55–64 (2008)
32. Bassetti, F.: Variable time-step discretization of degenerate evolution equations in Banach
space. Numer. Funct. Anal. Optim. 24(3–4), 391–426 (2003)
33. Beeler, G.W., Reuter, H.T.: Reconstruction of the action potential of ventricular myocardial
fibers. J. Physiol. 268, 177–210 (1977)
34. Beg, M.F., Helm, P.A., McVeigh, E., Miller, M.I., Winslow, R.L.: Computational cardiac
anatomy using MRI. Magn. Reson. Med. 52,1167–1174 (2004)
35. Belhamadia, Y.: A time-dependent adaptive remeshing for electrical waves of the heart. IEEE
Trans. Biomed. Eng. 55(2), 443–452 (2008)
36. Belhamadia, Y., Fortin, A., Bourgault, Y.: Towards accurate numerical method for mon-
odomain models using a realistic heart geometry. Math. Biosci. 220(2), 89–101 (2009)
37. Bellettini, G., Paolini, M.: Anisotropic motion by mean curvature in the context of Finsler
geometry. Hokkaido Math. J. 25, 537–566 (1996)
38. Bellettini, G., Paolini, M., Venturini, S.: Some results on surface measures in calculus of
variations. Ann. Math. Pura Appl. 170, 329–359 (1996)
39. Bellettini, G., Colli Franzone, P., Paolini, M.: Convergence of front propagation for
anisotropic bistable reaction-diffusion equations. Asymptot. Anal. 15, 325–358 (1997)
References 369

40. Bellettini, G., Paolini, M., Pasquarelli, F.: Nonconvex mean curvature flow as a formal
singular limit of the nonlinear bidomain model. Adv. Differ. Equ. 18(9–10), 895–934 (2013)
41. Beltrami, A.P., et al.: Adult cardiac stem cells are multipotent and support myocardial
regeneration. Cell 114, 763–776 (2003)
42. Bendahmane, M., Karlsen, K.H.: Analysis of a class of degenerate reaction-diffusion systems
and the bidomain model of cardiac tissue. Netw. Heterog. Media 1, 185–218 (2006)
43. Bendahmane, M., Karlsen, K.H.: Convergence of a finite volume scheme for the bidomain
model of cardiac tissue. Appl. Numer. Math. 59(9), 2266–2284 (2009)
44. Bensoussan, A., Lions, J.-L., Papanicolaou, G.: Asymptotic Analysis for Periodic Structures.
North-Holland, Amsterdam (1978)
45. Berbari, E.J., Lander, P., Scherlag, B.J., Lazzara, R., Geselowitz, D.B.: Ambiguities of
epicardial mapping. J. Electrocardiol. 24(Suppl), 16–20 (1992)
46. Bergmann, O., et al.: Evidence for cardiomyocyte renewal in humans. Science 324, 98–102
(2009)
47. Bers, D.M.: Excitation-Contraction Coupling and Cardiac Contractile Force, 2nd edn.
Kluwer, Dordrecht (2001)
48. Bertran, G., Biagetti, M.O., Valverde, E., Arini, P.D., Quinteiro, R.A., Lack of effect of
conduction direction on action potential durations in anisotropic ventricular strips of pig heart.
J. Cardiovasc. Electrophysiol. 13(4), 380–387 (2002)
49. Biktashev, V.N., Holden, A.V., Mironov, S.F., Pertsov, A.M., Zaitsev, A.V.: Three-dimensional
organisation of re-entrant propagation during experimental ventricular fibrillation. Chaos
Solitons Fract. 13(8), 1713–1733 (2002)
50. Bishop, M.J., Gavahan, D.J., Trayanova, N.A., Rodriguez, B.: Photon scattering effects in
optical mapping of propagation and arrhythmogenesis in the heart. J. Electrocardiol. 40, S75–
S80 (2007)
51. Bishop, M.J., Boyle, P.M., Plank, G., Welsh, D.G., Vigmond, E.J.: Modeling the role of the
coronary vasculature during external field stimulation. IEEE Trans. Biomed. Eng. 57(10),
2335–2345 (2010)
52. Bishop, M.J., Plank, G., Burton, R., Schneider, J., Gavaghan, D., Grau, V., Kohl, P.: Develop-
ment of an anatomically-detailed MRI-derived rabbit ventricular model and assessment of its
impact on simulation of electrophysiological function. Am. J. Physiol. Heart Circ. Physiol.
298, H699–H718 (2010)
53. Bondarenko, V.E., Szigeti, G.P., Bett, G.C., Kim, S.: A computer model for the action
potential of mouse ventricular myocytes. Am. J. Physiol. Heart Circ. Physiol. 278, H1378–
H1403 (2004)
54. Bordas, R., Carpentieri, B., Fotia, G., Maggio, F., Nobes, R., Pitt-Francis, J., Southern, J.:
Simulation of cardiac electrophysiology on next-generation high-performance computers.
Philos. Trans. R. Soc. A 367(1895), 1951–1969 (2009)
55. Bordas, R.M., Gillow, K., Gavaghan, D., Rodriguez, B., Kay, D.: A Bidomain model of the
ventricular specialized conduction system of the heart. SIAM J. Appl. Math. 72(5), 1618–
1643 (2012)
56. Beuter, A., Glass, L., Mackey, M.C., Titcombe, M.S.: Nonlinear Dynamics in Physiology and
Medicine. Springer, New York (2003)
57. Bouchard, S., Jacquemet, V., Vinet, A.: Automaticity in acute ischemia: bifurcation analysis
of a human ventricular model. Phys. Rev. E 83, 011911-1-10 (2011)
58. Boulakia, M., Fernandez, M.A., Gerbeau, J.-F., Zemzemi, N.: Towards the Numerical
Simulation of Electrocardiograms. In: FIMH07, Salt Lake City, 7–9 June 2007. LNCS,
vol. 4466, pp. 240–249. Springer, Berlin (2007)
59. Boulakia, M., Fernandez, M.A., Gerbeau, J.-F., Zemzemi, N.: A coupled system of PDEs
and ODEs arising in electrocardiograms modelling. Appl. Math. Res. Exp. 2, 1–28 (2008).
doi:10.1093/amrx/abn002
60. Bourgault, Y., Coudiere, Y., Pierre, C.: Existence and uniqueness of the solution for the
bidomain model used in cardiac electrophysiology. Nonlinear Anal. Real World Appl. 10,
458–482 (2009)
370 References

61. Brenner, S.C., Scott, L.R.: The Mathematical Theory of Finite Element Methods. Texts in
Applied Mathematics, vol. 15, 2nd edn. Springer, New York (2002)
62. Brezis, H.: Analyse Fonctionnelle, Theory and Applications. Masson, Paris (1983)
63. Britton, N.F.: Reaction – Diffusion Equations and Their Applications to Biology. Academic,
London (1986)
64. Brooks, D.H., Ahmad, G.F., MacLeod, R.S., Maratos, G.M.: Inverse electrocardiography
by simultaneous imposition of multiple constraints. IEEE Trans. Biomed. Eng. 46(1), 3–18
(1999)
65. Bueno-Orovio, A., Cherry, E., Fenton, F.: Minimal model for human ventricular action
potentials in tissue. J. Theor. Biol. 253, 544–560 (2008)
66. Buist, M., Sands, G., Hunter, P., Pullan, A.: A deformable finite element derived finite
difference method for cardiac activation problems. Ann. Biomed. Eng. 31, 577–588 (2003)
67. Burgess, M.J., Steinhaus, B.M., Spitzer, K.W., Ershler, P.R.: Nonuniform epicardial activation
and repolarization properties of in vivo canine pulmonary conus. Circ. Res. 62(2), 233–246
(1988)
68. Burnes, J.E., Taccardi, B., Rudy, Y.: Noninvasive imaging modality for cardiac arrhythmias.
Circulation 102, 2152–2158 (2000)
69. Burnes, J.E., Taccardi, B., Ershler, P.R., Rudy, Y.: Noninvasive electrocardiographic imaging
of substrate and intramural ventricular tachycardia in infarcted hearts. J. Am. Coll. Cardiol.
38, 2071–2078 (2001)
70. Burton, F.L., Cobbe, S.M.: Dispersion of ventricular repolarization and refractory period.
Cardiovasc. Res. 50, 10–23 (2001)
71. Cabo, C., Rosenbaum, D.: Quantitative Cardiac Electrophysiology. Marcel Dekker, New York
(2002)
72. Cai, X.-C.: Additive Schwarz algorithms for parabolic convection-diffusion equations.
Numer. Math. 60(1), 41–61 (1991)
73. Cai, X.-C.: Multiplicative Schwarz methods for parabolic problems. SIAM J. Sci. Comput.
15(3), 587–603 (1994)
74. Camara, O., Pop, M., Rhode, K., Sermesant, M., Smith, N., Young, A. (eds.): Proceedings
of Statistical Atlases and Computational Models of the Heart, STACOM’10, Beijing, 20 Sept
2010. LNCS, vol. 6364. Springer, Berlin (2010)
75. Camara, O., Konukoglu, E., Pop, M., Rhode, K., Sermesant, M., Young, A. (eds.): Proceed-
ings of Statistical Atlases and Computational Models of the Heart, STACOM’11, Toronto,
22 Sept 2011. LNCS, vol. 7085. Springer, Berlin (2012)
76. Camara, O., Mansi, T., Pop, M., Rhode, K., Sermesant, M., Young, A. (eds.): Proceedings of
Statistical Atlases and Computational Models of the Heart, STACOM’12, Nice, 5 Oct 2012.
LNCS, vol. 7746. Springer, Berlin (2013)
77. Camelliti, P., Borg, T.K., Kohl, P.: Structural and functional characterization of cardiac
fibroblasts. Cardiovasc. Res. 65, 40–51 (2005)
78. Carmeliet, E., Vereecke, J.: Cardiac Cellular Electrophysiology. Kluwer, Dordrecht (2002)
79. Casten, R.G., Cohen, H., Lagerstrom, P.A.: Perturbation analysis of an approximation to the
Hodgkin-Huxley theory. Q. Appl. Math. 32(4), 365–402 (1975)
80. Cates, A.W., Pollard, A.E.: A model study of intramural dispersion of action potential
duration in the canine pulmonary conus. Ann. Biomed. Eng. 26, 567–576 (1998)
81. Cerbai, E., Barbieri, M., Mugelli, A.: Characterization of the hyperpolarization-activated
current, If in ventricular myocytes isolated from hypertensive rats. J. Physiol. 481(3), 585–
591 (1994)
82. Cerbai, E., Barbieri, M., Mugelli, A.: Occurrence and properties of the hyperpolarization-
activated current If in ventricular myocytes from normotensive and hypertensive rats during
aging. Circulation 94, 1674–1681 (1996)
83. Cerqueira, M.D., Weissman, N.J., Dilsizian, V., et al.: Standardized myocardial segmentation
and nomenclature for tomographic imaging of the heart – a statement for healthcare
professionals from the Cardiac Imaging Committee of the Council on Clinical Cardiology
of the American Heart Association. Circulation 105(4), 539–542 (2002)
References 371

84. Chauhan, V.S., Downar, E., Nanthakumar, K., Parker, J.D., Ross, H.J., Chan, W., Picton, P.:
Increased ventricular repolarization heterogeneity in patients with ventricular arrhythmia
vulnerability and cardiomyopathy: a human in vivo study. Am. J. Physiol. Heart Circ. Physiol.
290(1), H79–H86 (2006)
85. Chen, X.-Y.: Dynamics of interfaces in reaction diffusion systems. Hiroshima Math. J. 21,
47–83 (1991)
86. Chen, X.-Y.: Generation and propagation of interfaces in reaction-diffusion systems. Trans.
Am. Math. Soc. 334(2), 877–913 (1992)
87. Chen, P.-S., Moser, K.M., Dembitsky, W.P., Auger, W.R., Daily, P.O., Calisi, C.M., Jamieson,
S.W., Feld, G.K.: Epicardial activation and repolarization patterns in patients with right
ventricular hypertrophy. Circulation 83, 104–118 (1991)
88. Cheng, D.K., Tung, L., Sobie, E.A.: Nonuniform responses of transmembrane potential
during electric field stimulation of single cardiac cells. Am. J. Physiol. Heart Circ. Physiol.
277, H351–H362 (1999)
89. Cheng, L.K., Bodley, J.M., Pullan, A.: Comparison of potential- and activation-based
formulations for the inverse problem of electrocardiology. IEEE Trans. Biomed. Eng. 50(1),
11–22 (2003)
90. Cherry, E.M., Greenside, H.S., Henriquez, C.S.: A space-time adaptive method for simulating
complex cardiac dynamics. Phys. Rev. Lett. 84(6), 1343–1346 (2000)
91. Cherry, E.M., Greenside, H.S., Henriquez, C.S.: Efficient simulation of three-dimensional
anisotropic cardiac tissue using an adaptive mesh refinement method. Chaos 13, 853–865
(2003)
92. Ciarlet, P.G.: The finite element method for elliptic problems. Classics in Applied Mathemat-
ics, vol. 40. SIAM, Philadelphia (2002)
93. Clancy, C.E., Rudy, Y.: Linking a genetic defect to its cellular phenotype in a cardiac
arrhythmia. Nature 400 566–569 (1999)
94. Clancy, C.E., Rudy, Y.: NaC channel mutation that causes both Brugada and Long-QT
syndrome phenotypes: a simulation study of mechanism. Circulation 105, 1208–1213 (2002)
95. Clayton, R.H., Holden, A.V.: Propagation of normal beats and re-entry in a computational
model of ventricular cardiac tissue with regional differences in action potential shape and
duration. Prog. Biophys. Mol. Biol. 85(2–3), 473–499 (2004)
96. Clayton, R.H., Holden, A.V.: Dispersion of cardiac action potential duration and the
initiation of re-entry: a computational study. Biomed. Eng. 4, 11 (2005). Online: http://www.
biomedical-engineering-oline.com/content/4/1/11
97. Clayton, R.H., Panfilov, A.V.: A guide to modelling cardiac electrical activity in anatomically
detailed ventricles. Prog. Biophys. Mol. Biol. 96, 19–43 (2008)
98. Clayton, R.H., Zhuchkova, E.A., Panfilov, A.V.: Phase singularities and filaments: simplifying
complexity in computational models of ventricular fibrillation. Prog. Biophys. Mol. Biol. 90,
378–398 (2006)
99. Clements, J.C., Nenonen, J., Li, P.K.J., Horacek, B.M.: Activation dynamics in anisotropic
cardiac tissue via decoupling. Ann. Biomed. Eng. 32(7), 984–990 (2004)
100. Clerc, L.: Directional differences of impulse spread in trabecular muscle from mammalian
heart. J. Physiol. 255, 335–346 (1976)
101. Colli Franzone, P., Guerri, L.: Models of the spreading of excitation in myocardial tissue.
CRC Crit. Rev. Biomed. Eng. 20, 211–253 (1992); Pilkington, T.C., Loftis, B., Palmer, T.,
Budinger, T.F. (eds.) High-Performance Computing in Biomedical Research, pp. 359–401.
CRC, Boca Raton (1993)
102. Colli Franzone, P., Guerri, L.: Spread of excitation in 3-D models of the anisotropic cardiac
tissue I: validation of the eikonal approach. Math. Biosci. 113, 145–209 (1993)
103. Colli Franzone, P., Magenes, E.: On the inverse potential problem of electrocardiology.
Calcolo 16(4), 459–538 (1979)
104. Colli Franzone, P., Pavarino, L.F.: A parallel solver for reaction-diffusion systems in
computational electrocardiology. Math. Model Methods Appl. Sci. 14(6), 883–911 (2004)
372 References

105. Colli Franzone, P., Savaré, G.: Degenerate evolution systems modeling the cardiac electric
field at micro and macroscopic level. In: Lorenzi, A., Ruf, B. (eds.) Evolution Equations,
Semigroups and Functional Analysis, pp. 49–78. Birkhauser, Basel/Boston (2002)
106. Colli Franzone, P., Taccardi, B., Viganotti, C.: An approach to inverse calculation of epicardial
potentials from body surface maps. Adv. Cardiol. 21, 50–54 (1978)
107. Colli Franzone, P., Guerri, L., Taccardi, B., Viganotti, C.: The direct and inverse potential
problem in electrocardiology. Numerical aspects of some regularization methods and appli-
cation to data collected in isolated dog heart experiments. Tech. Rep. No. 222, IAN-CNR
(1979)
108. Colli Franzone, P., Guerri, L., Viganotti, C., Macchi, E., Baruffi, S., Spaggiari, S.,
Taccardi, B.: Potential fields generated by oblique dipole layer modeling excitation wave-
fronts in the anisotropic myocardium. Comparison with potential fields elicited by paced dog
hearts in a volume conductor. Circ. Res. 51(3), 330–346 (1982)
109. Colli Franzone, P., Guerri, L., Viganotti, C.: Oblique dipole layer potentials applied to
electrocardiology. J. Math. Biol. 17(1), 93–124 (1983)
110. Colli Franzone, P., Guerri, L., Magenes, E.: Oblique dipole layer potential for the direct and
inverse problems of electrocardiology. Math. Biosci. 68, 23–55 (1984)
111. Colli Franzone, P., Guerri, L., Viganotti, C., Taccardi, B.: Finite element approximation of
regularized solution of the inverse potential problem of electrocardiography and applications
to experimental data. Calcolo 12(1), 91–186 (1985)
112. Colli Franzone, P., Guerri, L., Rovida, S.: Wavefront propagation in an activation model of
the anisotropic cardiac tissue: asymptotic analysis and numerical simulations. J. Math. Biol.
28, 121–176 (1990)
113. Colli Franzone, P., Guerri, L., Tentoni, S.: Mathematical modeling of the excitation process
in myocardial tissue: influence of fiber rotation on wavefront propagation and potential field.
Math. Biosci. 101, 155–235 (1990)
114. Colli Franzone, P., Guerri, L., Taccardi, B.: Spread of excitation in a myocardial volume.
Simulation studies in a model of anisotropic ventricular muscle activated by point stimulation.
J. Cardiovasc. Electrophysiol. 4, 144–160 (1993)
115. Colli Franzone, P., Guerri, L., Pennacchio, M., Taccardi, B.: Spread of excitation in 3-D
models of the anisotropic cardiac tissue II: effects of fiber architecture and ventricular
geometry. Math. Biosci. 147, 131–171 (1998)
116. Colli Franzone, P., Guerri, L., Pennacchio, M., Taccardi, B.: Spread of excitation in 3-D
models of the anisotropic cardiac tissue III: effects of ventricular geometry and fiber structure
on the potential distribution. Math. Biosci. 151, 51–98 (1998)
117. Colli Franzone, P., Guerri, L., Pennacchio, M., Taccardi, B.: Anisotropic mechanisms for
multiphasic unipolar electrograms. Simulation studies and experimental recordings. Ann.
Biomed. Eng. 28, 1–17 (2000)
118. Colli Franzone, P., Pennacchio, M., Guerri, L.: Accurate computation of electrograms in the
left ventricular wall. Math. Model Methods Appl. Sci. 10(4), 507–538 (2000)
119. Colli Franzone, P., Guerri, L., Taccardi, B.: On the polyphasic character of simulated and
experimental electrograms. Biomed. Tech. 46(2), 16–19 (2001)
120. Colli Franzone, P., Pavarino, L.F., Taccardi, B.: A parallel solver for anisotropic cardiac
models. In: Proceedings of the IEEE Computers in Cardiology, Thessaloniki Chalkidiki,
21–24 Sept 2003. vol. 30, pp. 781–784 (2003)
121. Colli Franzone, P., Guerri, L., Taccardi, B.: Modeling ventricular excitation: axial and
orthotropic effects on wavefronts and potentials. Math. Biosci. 188, 191–205 (2004)
122. Colli Franzone, P., Pavarino, L.F., Taccardi, B.: Monodomain simulations of excitation and
recovery in cardiac blocks with intramural heterogeneity. In: Frangi, A.F., et al. (eds.):
FIMH05: Functional Imaging and Modeling of the Heart. LNCS, vol. 3504, pp. 267–277.
Springer, Berlin (2005)
123. Colli Franzone, P., Pavarino, L.F., Taccardi, B.: Simulating patterns of excitation, repolariza-
tion and action potential duration with cardiac Bidomain and Monodomain models. Math.
Biosci. 197, 35–66 (2005)
References 373

124. Colli Franzone, P., Deuflhard, P., Erdmann, B., Lang, J., Pavarino, L.F.: Adaptivity in space
and time for reaction-diffusion systems in electrocardiology. SIAM J. Sci. Comput. 28(3),
942–962 (2006)
125. Colli Franzone, P., Pavarino, L.F., Savarè, G.: Computational electrocardiology: mathematical
and numerical modeling. In: Quarteroni, A., et al. (eds.) Complex Systems in Biomedicine,
pp. 187–241. Springer, New York (2006)
126. Colli Franzone, P., Pavarino, L.F., Taccardi, B.: Effects of transmural electrical heterogeneities
and electrotonic interactions on the dispersion of cardiac repolarization and action potential
duration: a simulation study. Math. Biosci. 204(1), 132–165 (2006)
127. Colli Franzone, P., Pavarino, L.F., Scacchi, S., Taccardi, B.: Determining recovery times
from transmembrane action potentials and unipolar electrograms in normal heart tissue. In:
Sachse, F.B., Seemann, G. (eds.) FIMH’07, Salt Lake City, 7–9 June 2007. LNCS, vol. 4466,
pp. 139–149. Springer, Berlin (2007)
128. Colli Franzone, P., Pavarino, L.F., Scacchi, S., Taccardi, B.: Monophasic action potentials
generated by bidomain modeling as a tool for detecting cardiac repolarization times. Am. J.
Physiol. Heart Circ. Physiol. 293, H2771–H2785 (2007)
129. Colli Franzone, P., Pavarino, L.F., Scacchi, S.: Dynamical effects of myocardial ischemia in
anisotropic cardiac models in three dimensions. Math. Model Methods Appl. Sci. 17(12),
1965–2008 (2007)
130. Colli Franzone, P., Pavarino, L.F., Scacchi, S., Taccardi, B.: Modeling ventricular repolar-
ization: effects of transmural and apex-to-base heterogeneities in action potential durations.
Math. Biosci. 214, 140–152 (2008)
131. Colli Franzone, P., Pavarino, L.F., Scacchi, S.: Exploring anodal and cathodal make and break
cardiac excitation mechanisms in a 3D anisotropic bidomain model. Math. Biosci. 230(2),
96–114 (2011)
132. Colli Franzone, P., Pavarino, L.F., Scacchi, S.: Cardiac excitation mechanisms, wavefront
dynamics and strength – interval curves predicted by 3D orthotropic bidomain simulations.
Math. Biosci. 235(1), 66–84 (2012)
133. Conrath, C.E., Opthof, T.: Ventricular repolarization: an overview of (patho)physiology,
sympathetic effects and genetic aspects. Prog. Biophys. Mol. Biol. 92(3), 269–307 (2006)
134. Corbin II, L.V., Scher, A.M.: The canine heart as an electrocardiographic generator: depen-
dence on cardiac cell orientation. Circ. Res. 41, 58–67 (1977)
135. Coronel, R., de Bakker, J.M.T., Wilms-Schopman, F.J.G., Opthof, T., Linnenbank, A.C.,
Belterman, C.N., Janse, M.J.: Monophasic action potentials and activation recovery intervals
as measures of ventricular action potential duration: experimental evidence to resolve some
controversies. Heart Rhythm 3(9), 1043–1050 (2006)
136. Cortassa, S., Aon, M., B. O’Rourke, Jacques, R., Tseng, H., Marban, E., et al.: A compu-
tational model integrating electrophysiology, contraction, and mitochondrial bioenergetics in
the ventricular myocytes. Biophys. J. 91, 1564–1598 (2006)
137. Costa, K.D., K. May-Newman, Farr, D., O’Dell, W.G., McCulloch, A.D., Omens, J.H.: Three-
dimensional residual strain in midanterior canine left ventricle. Am. J. Physiol. Heart Circ.
Physiol. 42, H1968–H1976 (1997)
138. Costa, K.D., Holmes, J.W., McCulloch, A.D.: Modelling cardiac mechanical properties in
three dimensions. Philos. Trans. R. Soc. Lond. A 359(1783), 1233–1250 (2001)
139. Coudiere, Y., Pierre, C.: Stability, convergence of a finite volume method for two systems
of reaction-diffusion equations in electro-cardiology. Nonlinear Anal. Real World Appl. 7(4),
916–935 (2006)
140. Coveney, P., Diaz, V., Hunter, P., Viceconti, M.: Computational Biomedicine. Oxford
University Press, Oxford (2014)
141. Cronin, J.: Mathematical Aspects of Hodgkin-Huxley Neural Theory. Cambridge University
Press, Cambridge (1987)
142. Cuppen, J.J.M.: Calculating the isochrones of ventricular depolarization. SIAM J. Sci. Stat.
Comput. 5, 105–120 (1984)
374 References

143. Cuppen, J.J.M., van Oosterom, A.: Model studies with the inversely calculated isochrones of
ventricular depolarization. IEEE Trans. Biomed. Eng. 31(10), 652–659 (1984)
144. De Ambroggi, L., Musso, E., Taccardi, B.: Body-surface mapping. In: Macfarlane, P.W.,
Lawrie, T.D.V. (eds.) Comprehensive Electrocardiology, pp. 1015–1049. Pergamon, Oxford
(1989)
145. DeBruin, K.A., Krassowska, W.: Electroporation and shock-induced transmembrane potential
in a cardiac fiber during defibrillation strength shocks. Ann. Biomed. Eng. 26, 584–596 (1998)
146. Dekker, E.: Direct current make and break thresholds for pacemaker electrodes on the canine
ventricle. Circ. Res. 27, 811–823 (1970)
147. Deuflhard, P., Erdmann, B., Roitzsch, R., Lines, G.T.: Adaptive finite element simulation of
ventricular fibrillation dynamics. Comput. Vis. Sci. 12(5), 201–205 (2009)
148. di Bernardo, D., Murray, A.: Computer model for study of cardiac repolarization. J.
Cardiovasc. Electrophys. 11(8), 895–899 (2000)
149. Di Francesco, D., Noble, D.: A model of cardiac electrical activity incorporating ionic pumps
and concentration changes. Philos. Trans. R. Soc. Lond. B 307(1133), 353–398 (1985)
150. Doi, S., Inoue, J., Pan, Z., Tsumoto, K.: Computational Electrophysiology. Springer, Tokyo
(2010)
151. Dryja, M., Widlund, O.B.: Multilevel additive methods for elliptic finite element problems.
In: Hackbusch, W. (ed.) Parallel Algorithms for Partial Differential Equations – Proceedings
of the Sixth GAMM-Seminar, Kiel, 19–21 Jan 1990. Notes on Numerical Fluid Mechanics,
vol. 31, pp. 58–69. Vieweg, Braunschweig (1991). 3-528-07631-3
152. Dryja, M., Widlund, O.B.: Domain decomposition algorithms with small overlap. SIAM J.
Sci. Comput. 15(3), 604–620 (1994)
153. Dryja, M., Sarkis, M.V., Widlund, O.B.: Multilevel Schwarz methods for elliptic problems
with discontinuous coefficients in three dimensions. Numer. Math. 72(3), 313–348 (1996)
154. Ebihara, L., Johnson, E.A.: Fast sodium current in cardiac muscle: a quantitative description.
Biophys. J. 32 779–790 (1980)
155. Efimov, I.R., Ermentrout, B., Huang, D.T., Salama, G.: Activation and repolarization patterns
are governed by different structural characteristics of ventricular myocardium: experimental
study with voltage-sensitive dyes and numerical simulations. J. Cardiovasc. Electrophysiol.
7, 512–530 (1996)
156. Efimov, I.R., Gheng, Y., Van Eagoner, D.R., Mazgalev, T., Tchou, P.J.: Virtual electrode-
induced phase singularity: a basic mechanism of defibrillation failure. Circ. Res. 82, 918–925
(1998)
157. Efimov, I.R., Gray, R.A., Roth, B.J.: Virtual electrodes and deexcitation: new insights into
fibrillation induction and defibrillation. J. Cardiovasc Electrophysiol. 11, 339–353 (2000)
158. Efimov, I.E., Kroll, M.W., Tcho, P.J. (eds.): Cardiac Bioelectric Therapy. Springer, New York
(2009)
159. Einstein, A.: Eine neue Bestimmung der Moleküldimensionen. Ann. Phys. 324, 289–306
(1906)
160. Ejima, J., Martin, D., Engle, C., Sherman, Z., Kunimoto, S., Gettes, L.: Ability of activation
recovery intervals to assess action potential duration during acute no-flow ischemia in the in
situ porcine heart. J. Cardiovasc. Electrophysiol. 99, 832–844 (1998)
161. El-Sherif, N., Caref, E.B., Yin, H., Estivo, M.: The electrophysiological mechanism of
ventricular arrhythmias in the long QT syndrome. Tridimensional mapping of activation and
recovery patterns. Circ. Res. 79, 474–492 (1996)
162. El-Sherif, N., Chinushi, M., Caref, E.B., Restivo, M.: Electrophysiological mechanism of
the characteristic electrocardiographic morphology of torsade de points tachyarrhythmias in
the long-QT syndrome: detailed analysis of ventricular tridimensional activation patterns.
Circulation 96, 4392–4399 (1997)
163. Endresen, L.P., Hall, K., Hoye, J.S., Myrheim, J.: A theory for the membrane potential of
living cells. Eur. Biophys. J. 29, 90–103 (2000)
164. Entcheva, E., Eason, J., Efimov, I.R., Cheng, Y., Malkin, R., Clayton, F.: Virtual electrode
effects in transvenous defibrillation-modulation by structure and interface: evidence from
bidomain simulations an optical mapping. J. Cardiovasc. Electrophysiol. 9, 949–961 (1998)
References 375

165. Ermentrout, B.: Simulating, analyzing, and animating dynamical systems: a guide to XPPAUT
for researchers and students. SIAM, Philadelphia (2002)
166. Ethier, M., Bourgault, Y.: Semi-implicit time-discretization schemes for the Bidomain model.
SIAM J. Numer. Anal. 46(5), 2443–2468 (2008)
167. Evans, L.C., Gariepy, R.F.: Measure Theory and Fine Properties of Functions. Studies in
Advanced Mathematics. CRC, New York (1997)
168. Fall, C.P., Marland, E.S., Wagner, J.M., Tyson, J.J.: Computational Cell Biology, 3rd edn.
Springer, New York (2005)
169. Fast, V.G., Kléber, A.G.: Role of wavefront curvature in propagation of cardiac impulse.
Cardiovasc. Res. 33, 258–271 (1997)
170. Fenton, F.H., Karma, A.: Vortex dynamics in three-dimensional continuous myocardium with
fiber rotation: filament instability and fibrillation. Chaos 8, 20–47 (1998)
171. Fenton, F.H., Cherry, E.M., Hastings, H.M., Evans, S.J.: Multiple mechanisms of spiral wave
breakup in a model of cardiac electrical activity. Chaos 12(3), 852–892 (2002)
172. Fernandez, M.A., Zemzemi, N.: Decoupled time-marching schemes in computational cardiac
electrophysiology and ECG numerical simulation. Math. Biosci. 226, 58–75 (2010)
173. Fife, P.C.: Mathematical Aspect of Reacting and Diffusing Systems. Springer, Berlin (1979)
174. Fife, P.C.: Dynamics of Internal Layers and Diffusive Interfaces. CBMS-NSF Regional
Conference Series in Applied Mathematics, vol. 53. SIAM, Philadelphia (1988)
175. Fife, P.C., McLeod, J.B.: The approach of solutions of nonlinear diffusion equations to
travelling front solutions. Arch. Ration. Mech. Anal. 65, 335–361 (1977)
176. Fish, J.M., Di Diego, J.M., Nesterenko, V., Antzelevich, C.: Epicardial activation of left ven-
tricular wall prolongs QT interval and transmural dispersion of repolarization. Implications
for biventricular pacing. Circulation 109, 2136–2142 (2004)
177. Fischer, G., Tilg, B., Modre, R., Huiskamp, G.J.M., Fetzer, J., Rucker, W., Wach, P.: A
bidomain model based BEM-FEM coupling formulation for anisotropic cardiac tissue. Ann.
Biomed. Eng. 28(10), 1229–1243 (2000)
178. FitzHugh, R.: Impulses and physiological states in theoretical models of nerve membrane.
Biophys. J. 1, 445–466 (1961)
179. FitzHugh, R.: Mathematical models of excitation and propagation in nerve. In: Schwan, H.P.
(ed.) Biological Engineering, pp. 1–85. MacGraw-Hill, New York (1969)
180. Frangi, A.F., Radeva, P.I., Santos, A., Hernandez, M. (eds.): Proceedings of the Third
International Workshop on Functional Imaging and Modeling of the Heart, FIMH’05,
Barcelona, 2–4 June 2005. LNCS, vol. 3504. Springer, Berlin (2005)
181. Franz, M.R.: Monophasic Action Potentials: Bridging Cells to Bedside, pp. 19–45. Futura
Publishing Company, Armonk (2000)
182. Franz, M.R.: What is a monophasic action potential recorded by Franz contact electrode?
Cardiovasc. Res. 65, 940–941 (2005)
183. Franz, M.R., et al.: Monophasic action potential mapping in human subjects with normal
electrocardiograms: direct evidence for the genesis of the T wave. Circulation 75, 379–386
(1987)
184. Frazier, D.W., Krassowska, W., Chen, P.-S., Wolf, P.D., Danieley, N.D., Smith, W.M., Ideker,
R.E.: Transmural activation and stimulus potentials in three-dimensional anisotropic canine
myocardium. Circ. Res. 63, 135–146 (1988)
185. Frazier, D.W., Wolf, P.D., Wharton, J.M., Tang, A.S.L., Smith, W.M., Ideker, R.E.: Stimulus-
induced critical point: mechanism for electrical initiation of reentry in normal canine
myocardium. J. Clin. Invest. 83, 1039–1052 (1989)
186. Fu, Z., Kirby, R.M., Whitaker, R.T.: A fast iterative method for solving the eikonal equation
on tetrahedral domains. SIAM J. Sci. Comput. 35(5), C473–C494 (2013)
187. Fuller, M.S., Sandor, G., Punske, B., Taccardi, B., MacLeod, R.S., Ershler, P.R., Green, L.S.,
Lux, R.L.: Estimates of repolarization dispersion from electrocardiographic measurements.
Circulation 102(6), 685–691 (2000)
188. Fuller, M.S., Sandor, G., Punske, B., Taccardi, B., MacLeod, R.S., Ershler, P.R., Green, L.S.,
Lux, R.L.: Estimates of repolarization and its dispersion from electrocardiographic
376 References

measurements: direct epicardial assessment in the canine heart. J. Electrocardiol. 33(2), 171–
180 (2000)
189. Fuster, V., Walsh, R.A., Harrington, R.A.: Hurst’s the Heart, 13th edn. MacGraw-Hill,
New York (2011)
190. Garbern, J.C., Lee, R.T.: Cardiac stem cell therapy and the promise of heart regeneration. Cell
Stem Cell 12(6), 689–698 (2013)
191. Garfinkel, A., Kim, Y.-H., Voroshilovsky, O., Qu, Z., Kil, J.R., Lee, M.-H.,
Karagueuzian, H.S., Weiss, J.N., Chen, P.-S.: Preventing ventricular fibrillation by flattening
cardiac restitution. Proc. Natl. Acad. Sci. U S A 97(11), 6061–6066 (2000)
192. Gaudesius, G., Miragoli, M., Thomas, S.P., Rohr, S.: Coupling of cardiac electrical activity
over extended distances by fibroblasts of cardiac origin. Circ. Res. 93, 421–428 (2003)
193. Gauthier, L.D., Greenstein, J.L., Winslow, R.L.: Toward an integrative computational model
of the guinea pig cardiac myocyte. Front. Physiol. 3, 244 (2012)
194. Gepstein, L., Hayam, G., Ben-Haim, S.A.: Activation-recovery coupling in the normal swine
endocardium. Circulation 96(11), 4036–4043 (1997)
195. Gerardo Giorda, L., Mirabella, L., Nobile, F., Perego, M., Veneziani, A.: A model-based
block-triangular preconditioner for the Bidomain system in electrocardiology. J. Comput.
Phys. 228(10), 3625–3639 (2009)
196. Gerardo Giorda, L., Perego, M., Veneziani, A.: Optimized Schwarz coupling of Bidomain and
Monodomain models in electrocardiology. Math. Model. Numer. Anal. 45, 309–334 (2011)
197. Geselowitz, D.B.: On the theory of the electrocardiogram. Proc. IEEE 77, 857–876 (1989)
198. Geselowitz, D.B.: Description of cardiac sources in anisotropic cardiac muscle. Application
of bidomain model. J. Electrocardiol. 25, 65–67 (1992)
199. Geselowitz, D.B., Miller, W.T.: A bidomain model for anisotropic cardiac muscle. Ann.
Biomed. Eng. 11, 191–206 (1983)
200. Ghosh, S., Rudy, Y.: Application of L1-norm regularization to epicardial potential solutions
of the inverse electrocardiography problem. Ann. Biomed. Eng. 37(5), 902–912 (2009)
201. Giaquinta, M., Hildebrandt, S.: Calculus of Variations I. Volume 310 of Grundlehren der
mathematischen Wissenschaften. Springer, Berlin (1996)
202. Giga, Y., Goto, S.: Motion of hypersurfaces and geometric equations. J. Math. Soc. Jpn. 44(1),
99–111 (1992)
203. Giga, Y., Sato, M.-H.: Neumann problem for singular degenerate parabolic equations. Differ.
Integral Equ. 6(6), 1217–1230 (1993)
204. Giga, Y., Goto, S., Ishii, H., Sato, M.-H.: Comparison principle and convexity preserving
properties for singular degenerate parabolic equations on unbounded domains. Ind. Univ.
Math. J. 40(2), 443–469 (1991)
205. Gima, K., Rudy, Y.: Ionic current basis of electrocardiographic waveforms. A model study.
Circ. Res. 90, 889–896 (2002)
206. Goldsmith, E.C., et al.: Organization of fibroblasts in the heart. Dev. Dyn. 230(4), 787–794
(2004)
207. Goto, M., Brooks, C.: Membrane excitability of the frog ventricle examined by long pulses.
Am. J. Physiol. Heart Circ. Physiol. 217, H1236–H1245 (1969)
208. Gotoh, M., Uchida, T., Fan, W., Fishbein, M.C., Karagueuzian, H.S., Chen, P.-S.: Anisotropic
repolarization in ventricular tissue. Am. J. Physiol. (Heart Circ. Physiol.) 41, 107–113 (1997)
209. Grandi, E., Pasqualini, F.S., Bers, D.M.: A novel computational model of the human
ventricular action potential and Ca transient. J. Mol. Cell. Cardiol. 48, 112–121 (2010)
210. Greensite, F.: The mathematical basis for imaging cardiac electrical function. CRC Crit. Rev.
Biomed. Eng. 22, 347–399 (1994)
211. Greensite, F., Huiskamp, G.: An improved method for estimating epicardial potentials from
the body surface. IEEE Trans. Biomed. Eng. 45, 98–104 (1998)
212. Griffith, B.E., Peskin, C.S.: Electrophysiology. Commun. Pure. Appl. Math. 66, 1837–1913
(2013)
213. Guan, S., Lu, Q., Huang, K.: A discussion about the Di Francesco-Noble model. J. Theor.
Biol. 189, 27–32 (1997)
References 377

214. Gulrajani, R.M.: Models of the electrical activity of the heart and computer simulation of the
electrocardiogram. CRC Crit. Rev. Biomed. Eng. 16, 1–66 (1988)
215. Gulrajani, R.M.: Bioelectricity and Biomagnetism. Wiley, New York (1998)
216. Gulrajani, R.M., Roberge, F.A., Savard, P.: The inverse problem of electrocardiography. In:
Macfarlane, P.W., Lawrie, T.T.V. (eds.) Comprehensive Electrocardiology, I: chap. 9, pp. 237–
288. Pergamon, Oxford (1989)
217. Han, C., Pogwizd, S.M., Killingsworth, C.R., He, B.: Noninvasive reconstruction of the three-
dimensional ventricular activation sequence during pacing and ventricular tachycardia in the
canice heart. Am. J. Physiol. Heart Circ. Physiol. 302, H244–H252 (2012)
218. Harrild, D.M., Henriquez, C.S.: A finite volume model of cardiac propagation. Ann. Biomed.
Eng. 28(2), 315–334 (1997)
219. Harrild, D.M., Henriquez, C.S.: A computer model of normal conduction in the human atria.
Circ. Res. 87, e25–e36 (2000)
220. Harrild, D.M., Penland, R., Henriquez, C.: A flexible method for simulating cardiac conduc-
tion in three-dimensional complex geometries. J. Electrocardiol. 33(3), 241–251 (2000)
221. Haws, C.W., Lux, R.L.: Correlation between in vivo transmembrane action potential durations
and activation–recovery intervals from electrograms. Circulation 81, 281–288 (1990)
222. He, B., Li, G., Zhang, X.: Noninvasive imaging of cardiac transmembrane potentials within
three-dimensional myocardium by means of a realistic geometry anisotropic heart model.
IEEE Trans. Biomed. Eng. 50(10), 1190–1202 (2003)
223. Heidenreich, E.A., Rodriguez, J.F., Gaspar, F.J., Doblaré, M.: Fourth-order compact schemes
with adaptive time step for monodomain reaction–diffusion equations. J. Comput. Appl.
Math. 216(1), 39–55 (2008)
224. Heijman, J., Volders, P.G.A., Westra, R.L., Rudy, Y.: Local control of ˇ-adrenergic stimu-
lation: effects on ventricular myocyte electrophysiology and Ca2C -transient. J. Mol. Cell.
Cardiol. 50(5), 863–871 (2011)
225. Henriquez, C.S.: Simulating the electrical behavior of cardiac tissue using the bidomain
model. CRC Crit. Rev. Biomed. Eng. 21, 1–77 (1993)
226. Henriquez, C.S., Muzikant, A.L., Smoak, C.K.: Anisotropy, fiber curvature, and bath loading
effects on activation in thin and thick cardiac tissue preparations: simulations in a three-
dimensional bidomain model. J. Cardiovasc. Electrophysiol. 7(5), 424–444 (1996)
227. Herron, T.J., Lee, P., Jalife, J.: Optical imaging of voltage and calcium in cardiac cells and
tissues. Circ. Res. 110, 609–623 (2012)
228. Hille, B.: Ionic Channels of Excitable Membranes, 2nd edn. Sinauer Associates Inc.,
Sunderland (1982)
229. Hodgkin, A., Huxley, A.: A quantitative description of membrane current and its application
to conduction and excitation in nerve. J. Physiol. (Lond.) 117, 500–544 (1952)
230. Hoff, D.: Stability and convergence of finite difference methods for systems of nonlinear
reaction-diffusion equations. SIAM J. Numer. Anal. 15, 1161–1177 (1978)
231. Holland, R.P., Brooks, H.: Precordial end epicardial surface potentials during Myocardial
ischemia in the pig. A theoretical and experimental analysis of the TQ and ST segments.
Circ. Res. 37, 471–480 (1975)
232. Holzapfel, G.A., Ogden, R.W.: Constitutive modelling of passive myocardium: a structurally
based framework for material characterization. Philos. Trans. R. Soc. A 13(367), 3445–3475
(2009)
233. Hooke, N.: Efficient simulation of action potential propagation in a bidomain. Ph.D. thesis,
Department of Computer Science, Duke University (1992)
234. Hooks, D.A., Tomlinson, K.A., Mardsen, S.G., LeGrice, I.J., Smaill, B.H., Pullan, A.J.,
Hunter, P.J.: Cardiac microstructure. Implications for electrical propagation and defibrillation
in the heart. Circ. Res. 91, 331–338 (2002)
235. Hopenfeld, B., Stinstra, J.G., MacLeod, R.S.: Mechanism for ST depression associated with
contiguous subendocardial ischemia. J. Cardiovasc. Electrophysiol. 29, 1200–1206 (2004)
236. Hormander, L.: The Analysis of Linear Partial Differential Operators. Springer, Berlin (1983)
378 References

237. Hoyt, R.H., Cohen, M.L., Saffitz, J.E.: Distribution and three-dimensional structure of
intercellular junctions in canine myocardium. Circ. Res. 64, 563–574 (1989)
238. Hsiao, G.C., Wendland, W.L.: Boundary Integral Equations. Springer, Berlin (2008)
239. http://commons.wikimedia.org/wiki/File:Wiggers_Diagram.png
240. http://www.texasheartinstitute.org/HIC/Anatomy/images/fig1_crosslg.jpg
241. Huiskamp, G., Greensite, F.: A new method for myocardial activation imaging. IEEE Trans.
Biomed. Eng. 44, 433–446 (1997)
242. Humphrey, J.D.: Cardiovascular Solid Mechanics. Cells, Tissues, and Organs. Springer,
New York (2002)
243. Hund, T.J., Rudy, Y.: Rate transient and regulation of action potential and calcium transient
in a canine cardiac ventricular cell model. Circulation 110, 3168–3174 (2004)
244. Hund, T.J., Kucera, J.P., Otani, N.F., Rudy, Y.: Ionic charge conservation and long-term steady
state in hte Luo–Rudy dynamic cell model. Biophys. J. 81, 3324–3331 (2001)
245. Hunter, P.J., Nash, M.P., Sands, G.B.: Computational electromechanics of the heart. In:
Panfilov, A.V., Holden, A.V. (eds.) Computational Biology of the Heart. Wiley, New York
(1997)
246. Hunter, P.J., McCulloch, A.D., ter Keurs, H.: Modelling the mechanical properties of cardiac
muscle. Prog. Biophys. Mol. Biol. 69(2–3), 289–331 (1998)
247. Hunter, P.J., et al.: A vision and strategy for the virtual physiological human in 2010 and
beyond. Philos. Trans. R. Soc. A 368, 2595–2614 (2010)
248. Hunter, P.J., et al.: A vision and strategy for the virtual physiological human: 2012 update.
Interface Focus 3, 1–9 (2013)
249. Isakov, V.: Inverse Problems for Partial Differential Equations, 2nd edn. Springer, Berlin
(2006)
250. Jack, J.J.B., Noble, D., Tsien, R.W.: Electric Current Flow in Excitable Cells. Clarendon,
Oxford (1983)
251. Jacquemet, V.: An eikonal approach for the initiation of reentrant cardiac propagation in
reaction-diffusion models. IEEE Trans. Biomed. Eng. 57(9), 2090–2098 (2010)
252. Jacquemet, V., Henriquez, C.: Finite volume stiffness matrix for solving anisotropic cardiac
propagation in 2-D and 3-D unstructured meshes. IEEE Trans. Biomed. Eng. 52(8), 1490–
1492 (2005)
253. Jacquemet, V., Kappenberger, L., Henriquez, C.S.: Modeling atrial arrhythmias: impact on
clinical diagnosis and therapies. IEEE Rev. Biomed. Eng. 1, 94–114 (2008)
254. Jafri, S., Rice, J.J., Winslow, R.L.: Cardiac Ca2C dynamics: the role of ryanodine receptor
adaptation and sarcoplasmic reticulum load. Biophys. J. 74, 1149–1168 (1998)
255. Janks D.L., Roth B.J.: Quaterfoil reentry caused by bursting pacing. J. Cardiovasc. Electro-
physiol. 17, 1362–1368 (2006)
256. Janks, D.L., Roth, B.J.: The bidomain theory of pacing. In: Efimov I.R., Kroll M.W., Tchou, J.
(eds.) Cardiac Bioelectric Therapy, chap. 2.1, pp. 63–83. Springer, New York (2009)
257. Janse, M.J.: ST segment mapping and infarct size. Cardiovasc. Res. 45, 190–193 (2000)
258. Janse, M., Sosunov, E.A., Corornel, R., Opthof, T., Anyukhovsky, E.P., de Bakker J.M.T.,
Plotnikov, A.N., Shlapakova, I.N., Danilo, P., J.Tijssen, G.P., Rosen, M.R.: Repolarization
gradients in the canine left ventricle before and after induction of short-term cardiac memory.
Circulation 112, 1711–1718 (2005)
259. Jerome, J.W.: Convergence of successive iterative semidiscretizations for FitzHugh-Nagumo
reaction systems. SIAM J. Numer. Anal. 27, 2054–1065 (1984)
260. Jikov, V.V., Kozlov, S.M., Oleinik, O.A.: Homogenization of Differential Operators and
Integral Functionals. Springer, Berlin (1994)
261. Johnston, P.: Computational Inverse Problems in Electrocardiography. WIT, Southampton
(2001)
262. Johnston, P.R., Kilpatrick, D.: The effect of conductivity values on ST segment shift in
subendocardial ischaemia. IEEE Trans. Biomed. Eng. 50, 150–158 (2003)
263. Johnston, P.R., Kilpatrick, D., Li, C.Y.: The importance of anisotropy in modeling ST segment
shift in subendocardial ischaemia. IEEE Trans. Biomed. Eng. 48, 1366–1376 (2001)
References 379

264. Joyner, R.W.: Modulation of repolarization by electrotonic interactions. Jpn. Heart J. 27, 167–
183 (1986)
265. Jungschleger, J.G., Vos, M.A.: Hybrid action potential etiology. J. Cardiovasc. Electrophysiol.
11(8), 946–948 (2000). (Letter to the Editor)
266. Katila, T., Magnin, I.E., Clarysse, P., Montagnat, J., Nenonen, J. (eds.): Proceedings of the
First International Workshop on Functional Imaging and Modeling of the Heart, FIMH’01,
Helsinki, 15–16 Nov 2001. LNCS, vol. 2230. Springer, Berlin (2001)
267. Katz, A.M.: Physiology of the Heart. Wolters Kluwer, Philadelphia (2011)
268. Keener, J.P.: An eikonal-curvature equation for action potential propagation in myocardium.
J. Math. Biol. 29, 629–651 (1991)
269. Keener, J.P.: Direct activation and defibrillation of cardiac tissue. J. Theor. Biol. 178, 313–324
(1996)
270. Keener, J.P., Bogar, K.: A numerical method for the solution of the bidomain equations in
cardiac tissue. Chaos 8(1), 234–241 (1998)
271. Keener, J.P., Panfilov, A.V.: Three–Dimensional propagation in the heart: the effects of
geometry and fiber orientation on propagation in myocardium. In: Zipes, D.P., Jalife, J. (eds.)
Cardiac Electrophysiology: From Cell to Bedside, pp. 335–347. W. B. Sounders, Philadelphia
(1995)
272. Keener, J.P., Panfilov, A.V.: The effects of geometry and fibre orientation on propagation and
extracellular potentials in myocardium. In: Panfilov, A.V., Holden, A.V. (eds.) Computational
Biology of the Heart, chap. 8, pp. 235–258. Wiley, New York (1997)
273. Keener, J.P., Sneyd, J.: Mathematical Physiology, 2nd edn. Springer, New York (2008)
274. Kerckhoffs, R.C.P. (ed.): Patient-Specific Modeling of the Cardiovascular System:
Technology-Driven Personalized Medicine. Springer, New York (2010)
275. Kerckhoffs, R.C.P., Bovendeerd, B.H.M., Kotte, J.C.S., Prinzen, F.W., Smits, K., Arts, T.:
Homogeneity of cardiac contraction despite physiological asynchrony of depolarization: a
model study. Ann. Biomed. Eng. 31, 536–547 (2003)
276. Kleber, A.G.: ST-segment elevation in the electrocardiogram: a sign of myocardial ischemia.
Cardiovasc. Res. 45, 111–118 (2000)
277. Kleber, A.G., Riegger, C.B.: Electrical constants of arterially perfused rabbit papillar muscle.
J. Physiol. 385, 307–324 (1987)
278. Kleber, A.G., Rudy, Y.: Basic mechanisms of cardiac impulse propagation and associated
arrhythmias. Physiol. Rev. 84(2), 431–488 (2004)
279. Kleber, A.G., Janse, M.J., van Capelle, F.J.L., Durrer, D.: Mechanism and time course of
S-T and T-Q segment changes during acute regional myocardial ischemia in the pig heart
determined by extracellular and intracellular recordings. Circ. Res. 42(5), 603–613 (1978)
280. Kneller, J., Ramirez, R.J., Chartier, D., M.Courtemanche, Nattel, S.: Time-dependent tran-
sients in an ionically based mathematical model of the canine atrial action potential. Am. J.
Physiol. Heart. Circ. Physiol. 282, H1437–H1451 (2002)
281. Knisley, S.B.: Transmembrane voltage changes during unipolar stimulation of rabbit ventri-
cle. Circ. Res. 77(6), 1229–1239 (1995)
282. Koch, C.: Biophysics of Computation. Oxford University Press, New York (1999)
283. Kogan, B.Y.: Introduction to Computational Cardiology. Springer, New York (2010)
284. Kohl, P., Sachs, F., Franz, M.R.: Cardiac Mechano-Electric Coupling and Arrhythmias.
Oxford University Press, Oxford (2011)
285. Kondo, M., Nesterenko, V., Antzelevitch, C.: Cellular basis for the monophasic action
potential. Which electrode is the recording electrode? Cardiovasc. Res. 63, 635–644 (2004)
286. Kjekshus, J.K., Maroko, P.R., Sobel, B.E.: Distribution of myocardial injury and its relation
to epicardial ST-segment changes after coronary artery occlusion in the dog. Cardiovasc. Res.
6, 490–499 (1972)
287. Krogh-Masden, T., Cristini, D.J.: Nonlinear dynamics in cardiology. Ann. Rev. Biomed. Eng.
14, 179–203 (2012)
288. Kuznetsov, Y.: Elements of Applied Bifurcation Theory. Springer, New York (2000)
289. Laflamme, M.A., Murry, C.E.: Heart regeneration. Nature 473 326–335 (2011)
380 References

290. Lang, J.: Adaptive Multilevel Solution of Nonlinear Parabolic PDE Systems. Theory,
Algorithms, and Applications. LNCSE, vol. 16. Springer, Berlin (2000)
291. Leeson, P.: Cardiovascular Imaging. Oxford University Press, Oxford (2011)
292. LeGrice, I.J., Smaill, B.H., Chai, L.Z., Edgar, S.G., Gavin, J.B., Hunter, P.J.: Laminar
structure of the heart: ventricular myocyte arrangement and connective tissue architecture
in the dog. Am. J. Physiol. Heart Circ. Physiol. 269(38), H571–H582 (1995)
293. LeGrice, I.J., Smaill, B.H., Hunter, P.J.: Laminar structure of the heart: a mathematical model.
Am. J. Physiol. Heart Circ. Physiol. 272(41), H2466–H2476 (1997)
294. LeGrice, I.J., Hunter, P.J., Young, A., Smaill, B.H.: The architecture of the heart: a data-based
model. Philos. Trans. R. Soc. Lond. A 359, 1217–1232 (2001)
295. Leon, L.J., Horacek, B.M.: Computer model of excitation and recovery in the anisotropic
myocardium, I: rectangular and cubic arrays of excitable elements. II: excitation in the
simplified left ventricle III: arrhythmogenic conditions in the simplified left ventricle. J.
Electrocardiol. 14, 1–15, 17–31, 33–41 (1991)
296. Leri, A., Kajstura, J., Anversa, P.: Role of cardiac stem cells in cardiac phatophysiology: a
paradigm shift in human myocardial biology. Circ. Res. 109, 941–961 (2011)
297. Lesh, M.D., Spear, J.F., Simson, M.B.: A computer model of the electrogram: what causes
fractionation? J. Electrocardiol. 21(Suppl), S69–S73 (1988)
298. Li, D., Li, C.Y., Yong, A.C., Johnston, P.R., Kilpatrick, D.: Epicardial ST depression in acute
myocardial ischemia. Circ. Res. 85, 959–964 (1999)
299. Li, L., Niederer, S., et al.: A mathematical model of the murine ventricular myocyte: a data-
driven biophysically based approach applied to mice overexpressing the canine NCX isoform.
Am. J. Physiol. HC 299, H1045–H1063 (2010)
300. Lindemans, F.W., van der Gon, J.J.D.: Current threshold and liminal size in excitation of heart
muscle. Cardiovasc. Res. 12(8), 477–485 (1978)
301. Lindemans, F.W., Heethaar, R.M., van der Gon, J.J.D., Zimmerman, A.N.E.: Site of initial
excitation and current threshold as a function of electrode radius in heart muscle. Cardiovasc.
Res. 9, 95–104 (1975)
302. Lines, G.T., Grottum, P., Tweito, A.: Modeling the electric activity of the heart: a bidomain
model of the ventricles embedded in a torso. Comput. Vis. Sci. 5, 195–213 (2003)
303. Linge, S., Sundnes, J., Hanslien, M., Lines, G.T., Tveito, A.: Numerical solution of the
bidomain equations. Philos. Trans. R. Soc. A 367(1895), 1931–1950 (2009)
304. Lions, J.L.: Optimal Control of Systems Governed by Partial Differential Equations. Springer,
Berlin (1971)
305. Lions, J.L., Magenes, E.: Nonhomogeneous boundary value problems and applications.
I. Springer, Berlin (1972)
306. Livshitz, L.M., Rudy, Y.: Regulation of Ca2C and electrical alternans in cardiac myocytes:
role of CAMKII and repolarizing currents. Am. J. Physiol. Heart Circ. Physiol. 292, H2854–
H2866 (2007)
307. Livshitz, L., Rudy, Y.: Uniqueness and stability of action potential models during rest, pacing,
and conduction using problem-solving environment. Biophys. J. 97, 1265–1276 (2009)
308. Luo, C., Rudy, Y.: A model of the ventricular cardiac action potential: depolarization,
repolarization, and their interaction. Circ. Res. 68(6), 1501–1526 (1991)
309. Luo, C., Rudy, Y.: A dynamic model of the cardiac ventricular action potential. I. Simulations
of ionic currents and concentration changes. Circ. Res. 74(6), 1071–1096 (1994)
310. Luo, C., Rudy, Y.: A dynamic model of the cardiac ventricular action potential. II. Afterde-
polarizations, triggered activity, and potentiation. Circ. Res. 74(6), 1097–1113 (1994)
311. Lux, R.L., Gettes, L., Mason, J.W.: Understanding proarrhythmic potential in therapeutic drug
development: alternate strategies for measuring and tracking repolarization. J. Electrocardiol.
39, S161–S164 (2006)
312. Macfarlane, P.W., van Oosterom, A., Janse, M., Kligfield, P., Camm, J., Pahlm, O. (eds.):
Basic Electrocardiology. Cardiac Electrophysiology, ECG Systems and Mathematical Mod-
eling. Springer, New York (2012)
References 381

313. MacLachlan, M.C., Sundnes, J., Lines, G.T.: Simulation of ST segment changes during
subendocardial ischemia using a realistic 3-D cardiac geometry. IEEE Trans. Biomed. Eng.
52, 799–807 (2005)
314. MacLeod, R.S., Punske, B., Yilmaz, B., Shome, S., Taccardi, B.: The role of heart rate in
myocardial ischemia from restricted coronary perfusion. J. Electrocardiol. 34, 43–51 (2001)
315. Mahajan, A., Shiferaw, Y., Sato, D., Baher, A., Olcese, R., Xie, L.H., Yang, M.J., Chen, P.S.,
Restrepo, J.G., Karma, A., Garfinkel, A., Qu, Z., Weiss, J.N.: A rabbit ventricular action
potential model replicating cardiac dynamics at rapid heart rates. Biophys. J. 94(2), 392–410
(2008)
316. Magnin, I.E., Montagnat, J., Clarysse, P., Nenonen, J., Katila, T. (eds.): Proceedings of the
Second International Workshop on Functional Imaging and Modeling of the Heart, FIMH’03,
Lyon, 5–6 June 2003. LNCS, vol. 2674. Springer, Berlin (2003)
317. Malmivuo, J., Plonsey, R.: Bioelectromagnetism. Oxford University Press, Oxford (1995)
318. Mardal, K.A., Sundnes, J., Langtangen, H.P., Tveito, A.: Systems of PDEs and block
preconditioning. In: Langtangen, H.P., Tveito, A. (eds.) Advanced Topics in Computational
Partial Differential Equations. LNCSE, vol. 33, chap. 5, pp. 200–236. Springer, Berlin (2004)
319. Mardal, K.-A., Nielsen, B.F., Cai, X., Tveito, A.: An order optimal solver for the discretized
bidomain equations. Numer. Linear Algebr. Appl. 14(2), 83–98 (2007)
320. Markowich, P.A.: The Stationary Semiconductor Device Equations. Springer, Berlin (1986)
321. Mascagni, M.: The backward euler method for numerical solution of the Hodgkin–Huxley
equations of nerve conduction. SIAM J. Numer. Anal. 27(4), 941–962 (1990)
322. Matta, R.J., Verrier, R.L., Lown, B.: Repetitive extrasystole as an index of vulnerability to
ventricular fibrillation. Am. J. Physiol. 230, 1469–1473 (1976)
323. McAllister, R.E., Noble, D., Tsien, R.W.: Reconstruction of the electrical activity of cardiac
Purkinje fibres. J. Physiol. 251, 1–59 (1975)
324. Mehra, R., Furman, S.: Comparison of cathodal, anodal, and bipolar strength-interval curves
with temporary and permanent pacing electrodes. Br. Heart J. 41, 468–476 (1979)
325. Messnarz, B., Seger, M., Modre, R., Fischer, G., Hanser, F., Tilg, B.: A comparison of
noninvasive reconstruction of epicarial versus transmembrane potentials in consideration of
the null space. IEEE Trans. Biomed. Eng. 51(9), 1609–1618 (2004)
326. Messnarz, B., Tilg, B., Modre, R., Fischer, G., Hanser, F.: A new spatiotemporal regularization
approach for reconstruction of cardiac transmembrane potential patterns. IEEE Trans.
Biomed. Eng. 51(2), 273–281 (2004)
327. Metaxas, D.N., Axel, L. (eds.): Proceedings of the 6th International Conference on Functional
Imaging and Modeling of the Heart, FIMH’11, New York City, 25–27 May 2011. LNCS,
vol. 6666. Springer, Berlin (2011)
328. Millar, C.K., Kralios, F.A., Lux, R.L.: Correlation between refractory periods and activation-
recovery intervals from electrograms – effects of rate and adrenergic interventions. Circula-
tion 72, 1372–1379 (1985)
329. Miller, W.T., Geselowitz, D.B.: Simulation studies of the electrocardiogram I. The normal
heart. Circ. Res. 43(2), 301–315 (1978)
330. Miragoli, M., Gaudesius, G., Rohr, S.: Electrotonic modulation of cardiac impulse conduction
by myofibroblasts. Circ. Res. 98, 801–810 (2006)
331. Mirams, G.R., et al.: Chaste: an open source C plus plus library for computational physiology
and biology. PLoS Comput. Biol. 9(3), e100297 (2013)
332. Miranda, C.: Partial Differential Equations of Elliptic Type. Springer, Berlin (1970)
333. Mitchell, C.C., Schaeffer, D.G.: A two-current model for the dynamics of cardiac membrane.
Bull. Math. Biol. 65(5), 767–793 (2003)
334. Miura, R.H.: Accurate computation of stable solitary waves for the FitzHugh–Nagumo
equations. J. Math. Biol. 13, 247–269 (1982)
335. Modre, R., Tilg, B., Fischer, G., Wach, P.: Noninvasive myocardial activation time imaging:
a novel inverse algorithm applied to clinical ECG mapping data. IEEE Trans. Biomed. Eng.
49(10), 1153–1161 (2002)
382 References

336. Moore, P.K.: An adaptive finite element method for parabolic differential systems: some
algorithmic considerations in solving in three space dimensions. SIAM J. Sci. Comput. 21(4),
1567–1586 (2000)
337. Morris, C., Lecar, H.: Voltage oscillations in the barnacle giant muscle. Biophys. J. 35, 193–
213 (1981)
338. Munteanu, M.: Overlapping additive Schwarz methods for nonlinear parabolic reaction-
diffusion problems. Ph.D. thesis, Department of Mathematics, University of Milano (2008)
339. Munteanu, M., Pavarino, L.F.: Implicit parallel solvers in computational electrocardiology.
In: Carja, O., Vrabie, I.I. (eds.) Applied Analysis and Differential Equations, pp. 255–266.
World Scientific, Singapore (2007)
340. Munteanu, M., Pavarino, L.F.: Decoupled Schwarz algorithms for implicit discretization of
nonlinear Monodomain and Bidomain systems. Math. Model Methods Appl. Sci. 19(7),
1065–1097 (2009)
341. Munteanu, M., Pavarino, L.F., Scacchi, S.: A scalable Newton-Krylov-Schwarz method for
the Bidomain reaction-diffusion system. SIAM J. Sci. Comput. 31(5), 3861–3883 (2009)
342. Murillo, M., Cai, X.: A fully implicit parallel algorithm for simulating the nonlinear electrical
activity of the heart. Numer. Linear Algebr. Appl. 11, 261–277 (2004)
343. Murthy, M.K.V., Stampacchia, G.: Le problème de Dirichlet pour les équations elliptiques du
second ordreà coefficients discontinus. Ann. Ist. Fourier XV, 189–258 (1965)
344. Muzikant, A., Hsu, E.W., Wolf, P.D., Henriquez, C.S.: Region specific modeling of cardiac
muscle: comparison of simulated and experimental potentials. Ann. Biomed. Eng. 30, 867–
883 (2002)
345. Nayak, A.R., Shajahan, T.K., Panfilov, A.V., Pandit, R.: Spiral-wave dynamics in a mathemat-
ical model of human ventricular tissue with myocytes and fibroblasts. PLoS One 8(9), e72950
(2013)
346. Nelson, C.V., Gezelowitz, D.B.: The Theoretical Basis of Electrocardiology. Clarendon,
Oxford (1976)
347. Nesterenko, V.V., Kondo, M., Antzelevitch, C.: Biophysical basis for monophasic action
potential. Cardiovasc. Res. 65, 942–944 (2005)
348. Neu, J.S., Krassowska, W.: Homogenization of syncitial tissues. CRC Crit. Rev. Biomed. Eng.
21(2), 137–199 (1993)
349. Neunlist, M., Tung, L.: Spatial distribution of cardiac transmembrane potentials around an
extracellular electrode: dependence on fiber orientation. Biophys. J. 68, 2310–2311 (1995)
350. Ni, Q., MacLeod, R.S., Punske, B.B., Taccardi, B.: Computing and visualizing electric
potentials and current pathways in the Thorax. J. Electrocard. 33, 189–197 (2000)
351. Niederer, S., Smith, N.: A mathematical model of the slow force response to stretch in rat
ventricular myocytes. Biophys. J. 92, 4030–4044 (2007)
352. Nielsen, I.J., Le Grice, P.M.F., Hunter, P.J., Smaill, B.H.: Mathematical model of geometry
and fibrous structure of the heart. Am. J. Physiol. Heart Circ. Physiol. 260, H1365–H1378
(1991)
353. Nielsen, B.F., Cai, X., Sundnes, J., Tveito, A.: Toward a computational method for imaging
the extracellular potassium concentration during regional ischemia. Math. Biosci. 220, 118–
130 (2009)
354. Nielsen, B.F., Lysaker, M., Grøttum, P.: Computing ischemic regions in the heart with the
bidomain model: first step toward validation. IEEE Trans. Med. Imag. 32(6), 1085–1096
(2013)
355. Noble, D.: A modification of the Hodgkin-Huxley equations applicable to Purkinje fibre
action and pacemaker potentials. J. Physiol. 160, 317–352 (1962)
356. Noble, D., Rudy, Y.: Models of cardiac ventricular action potentials: iterative interaction
between experiment and simulation. Philos. Trans. R. Soc. Lond. A 359, 1127–1142 (2001)
357. Noble, D., Noble, S., Bett, C., Earm, Y.E., Ko, W.K., So, I.K.: The role of sodium-calcium
exchange during the cardiac action potential. Ann. NY Acad. Sci. 639, 334–354 (1991)
References 383

358. Noble, D., Varghese, A., Kohl, P., Noble, P.J.: Improved guinea-pig ventricular cell model
incorporating a diadic space, iKr & iKs, and lenght- & tension-dependent processes. Can.
J. Cardiol. 14, 123–134 (1998)
359. Nochetto, R.H., Savarè, G., Verdi, C.: A posteriori error estimates for variable time-step
discretizations of nonlinear evolution equations. Commun. Pure Appl. Math. 53(5), 525–589
(2000)
360. Nygren, A., Fiset, C., Firek, L., Clark, J.W., Lindblad, D.S., Clark, R.B., Giles, W.R.:
Mathematical Model of an Adult Human Atrial Cell. Circ. Res. 82, 63–81 (1998)
361. O’Hara, T., Virag, L., Varro, A., Rudy, Y.: Simulation of the undiseased human cardiac
ventricular action potential: model formulation and experimental validation. PLoS Comput.
Biol. 7(5), 1–29 (2011)
362. Oleinik, O.A., Shamaev, A.S., Yosifian, G.A.: Mathematical Problems in Elasticity and
Homogenization. North-Holland, Amsterdam (1992)
363. Opthof, T., Coronel, R., Wilms-Schopman, F.J.G., Plotnikov, A.N., Shlapakova, I.N.,
Danilo, P., Rosen, M.R., Janse, M.J.: Dispersion of repolarization in canine ventricle and
the electrocardiographic T wave: Tpe interval does not reflect transmural dispersion. Heart
Rhythm 4, 341–348 (2007)
364. Osaka, T., Kodama, I., Tsuboi, N., Toyama, J., Yamada, K.: Effects of activation sequence and
anisotropic cellular geometry on the repolarization phase of action potential of dog ventricular
muscle. Circulation 76(1), 226–236 (1987)
365. Osher, S., Fedkin, R.: Level Set Methods and Dynamic Implicit Surfaces. Applied Mathemat-
ical Sciences, vol. 153. Springer, New York (2003)
366. Ourselin, S., Rueckert, D., Smith, N. (eds.): Proceedings of the 7th International Conference
on Functional Imaging and Modeling of the Heart, FIMH’13, London, 20–22 June 2013.
LNCS, vol. 7945. Springer, Berlin (2013)
367. Pandit, S.V., Clark, R.B., Giles, W.R., Demir, S.S.: A mathematical model of action potential
heterogeneity in adult rat left ventricular myocytes. Biophys. J. 81(6), 3029–3051 (2001)
368. Panfilov, A.V.: Spiral breakup as a model of ventricular fibrillation. Chaos 8, 57–64 (1998)
369. Panfilov, A.V., Holden, A.V.: Computational Biology of the Heart. Wiley, New York (1997)
370. Park, J.-H., Jerome, J.W.: Qualitative properties of steady-state Poisson-Nernst-Planck sys-
tems: mathematical study. SIAM J. Appl. Math. 57(3), 609–630 (1997)
371. Pavarino, L.F., Colli Franzone, P.: Parallel solution of cardiac reaction-diffusion models. In:
Kornhuber, R., et al. (eds.) Domain Decomposition Methods in Science and Engineering.
LNCSE, vol. 40, pp. 669–676. Springer, Berlin (2004)
372. Pavarino, L.F., Scacchi, S.: Multilevel additive Schwarz preconditioners for the Bidomain
reaction-diffusion system. SIAM J. Sci. Comput. 31(1), 420–443 (2008)
373. Pavarino, L.F., Scacchi, S.: Parallel multilevel Schwarz and block preconditioners for the
Bidomain parabolic-parabolic and parabolic-elliptic formulations. SIAM J. Sci. Comput.
33(4), 1897–1919 (2011)
374. Payne, L.E.: Improperly Posed Problems in Partial Differential Equations. PA Saunders,
Philadelphia (1975)
375. Penland, R., Harrild, D., Henriquez, C.: Modeling impulse propagation and extracellular
potential distributions in anisotropic cardiac tissue using a finite volume element discretiza-
tion. Comput. Vis. Sci. 4, 215–226 (2002)
376. Pennacchio, M.: A nonconforming domain decomposition method for the cardiac potential
problem. In: Proceedings of IEEE Computers in Cardiology, Rotterdam, 23–26 Sept 2001,
vol. 28, pp. 537–540 (2001)
377. Pennacchio, M.: The mortar finite element method for the cardiac “bidomain” model of
extracellular potential. J. Sci. Comput. 20(2), 191–210 (2004)
378. Pennacchio, M., Savarè, G., Colli Franzone, P.: Multiscale modeling for the electrical activity
of the heart. SIAM J. Math. Anal. 37(4), 1333–1370 (2006)
379. Pennacchio, M., Simoncini, V.: Efficient algebraic solution of reaction-diffusion systems for
the cardiac excitation process. J. Comput. Appl. Math. 145, 49–70 (2002)
384 References

380. Pennacchio, M., Simoncini, V.: Substructuring preconditioners for mortar discretization of
degenerate evolution problem. J. Sci. Comput. 36, 391–419 (2008)
381. Pennacchio, M., Simoncini, V.: Algebraic multigrid preconditioners for the bidomain
reaction–diffusion system. Appl. Numer. Math. 59, 3033–3050 (2009)
382. Pennacchio, M., Simoncini, V.: Fast structured AMG preconditioning for the bidomain model
in electrocadiology. SIAM J. Sci. Comput. 33, 721–745 (2011)
383. Peskin, C.S.: Mathematical Aspects of Heart Physiology. Lecture Notes of the Courant
Institute of Mathematical Sciences, New York University, New York (1975)
384. Peskin, C.S.: Numerical analysis of blood flow in the heart. J. Comput. Phys. 25(3), 220–252
(1977)
385. Peskin, C.S., McQueen, D.M.: Cardiac fluid dynamics. Crit. Rev. Biomed. Eng. 20, 451–459
(1992)
386. Pilkington, T.C., Plonsey, R.: Engineering Contributions to Biophysical Electrocardiography.
IEEE, New York (1982)
387. Pilkington, T.C., Loftis, B., Palmer, T., Budinger, T.F.: High-Performance Computing in
Biomedical Research. CRC, Boca Raton (1993)
388. Plank, G., Burton, R.A.B., Hales, P., Bishop, M., Mansoori, T., Bernabeu, M.O., Garny,
A., Prassl, A.J., Bollendorsff, C., Mason, F., Mahmood, F., Rodriguez, B., Grau, V.,
Schneider, J.E., Gavaghan, D., Kohl, P.: Generation of histo-anatomically representative
models of the individual heart: tools and application. Philos. Trans. R. Soc. A 367(1895),
2257–2292 (2009)
389. Plank, G., Liebmann, M., Weber dos Santos, M.O., Vigmond, E.J., Haase, G.: Algebraic
Multigrid Preconditioner for the Cardiac Bidomain Model. IEEE Trans. Biomed. Eng. 54(4),
585–596 (2007)
390. Plank, G., Prassl, A., Hofer, E., Trayanova, N.A.: Evaluating intramural virtual electrodes in
the myocardial wedge preparation: simulations of experimental conditions. Biophys. J. 94,
1904–1915 (2008)
391. Plonsey, R.: Bioelectric sources arising in excitable fibers (Alza lecture). Ann. Biomed. Eng.
16, 519–546 (1988)
392. Plonsey, R., Barr, R.C.: Current flow patterns in two-dimensional anisotropic bisyncitia with
normal and extreme conductivities. Biophys. J. 45, 557–571 (1984)
393. Plonsey, R., Barr, R.C.: Bioelectricity: A Quantitative Approach. Springer, New York (2007)
394. Plonsey, R., Heppner, D.: Consideration of quasi-stationarity in electrophysiological systems.
Bull. Math. Biophys. 29, 657–664 (1967)
395. Poelzing, S., Rosenbaum, D.S.: Heterogeneous connexin43 expression produces electrophys-
iological heterogeneities across ventricular wall. Am J. Physiol. Heart Circ. Physiol. 286,
H2001–H2009 (2004)
396. Poelzing, S., Rosenbaum, D.S.: Altered connexin43 expression produces arrhythmia substrate
in heart failure. Am. J. Physiol (Heart Circ. Physiol) 287, H1762–H1770 (2004)
397. Pollard, A.E., Hooke, N., Henriquez, C.S.: Cardiac propagation simulation. CRC Crit. Rev.
Biomed. Eng. 20(3–4), 171–210 (1992)
398. Pollard, A.E., Burgess, M.J., Spitzer, K.W.: Computer simulations of three-dimensional
propagation in ventricular myocardium. Circ. Res. 72(4), 744–756 (1993)
399. Potse, M., Dubè, B., Richer, J., Vinet, A., Gulrajani, R.: A comparison of Monodomain and
Bidomain reaction–diffusion models for action potential propagation in the human heart.
IEEE Trans. Biomed. Eng. 53(12), 2425–2434 (2006)
400. Potse, M., Vinet, A., Opthof, T., Coronel, R.: Validation of a simple model for the morphology
of the T wave in unipolar electrograms. Am. J. Physiol. Heart Circ. Physiol. 297(2), H792–
H801 (2009)
401. Pressler, M.L., Munster, P.N., Huang, X.-D.: Gap junction distribution in the heart: functional
relevance. In: Zipes, D., Jalife, J. (eds.) Cardiac Electrophysiology, chap. 16, pp. 144–151.
W. B. Sauders, Philadelphia (1995)
402. Priebe, L., Beuckelmann, D.J.: Simulation study of cellular electrical properties in heart
failure. Circ. Res. 82, 1206–1223 (1998)
References 385

403. Prinzmetal, M., Toyoshima, A., Ekmekci, Y., Mizumo, Y., Nagaya, T.: Myocardial ischemia.
Nature of ischemic electrocardiographic patterns in the mammalian ventricles as determined
by intracellular electrographic and metabolic changes. Am. J. Cardiol. 8, 493–503 (1961)
404. Prior, P., Roth, B.J.: Calculation of optical signal using three-dimensional bidomain/diffusion
model reveals distortion of the transmembrane potential. Biophys. J. 95, 2097–2102 (2008)
405. Puglisi, J.L., Bers, D.M.: LabHEART: an interactive computer model of rabbit ventricular
myocyte ion channels and Ca transport. Am. J. Physiol. Cell Physiol. 281(6), C2049–C2060
(2001)
406. Pullan, A.J., Buist, M.L., Cheng, L.K.: Mathematically Modelling the Electrical Activity of
the Heart. World Scientific, Singapore (2005)
407. Pullan, A.J., Cheng, L.K., Nash, M.P., Ghodrati, A., MacLeod, R.S., Brooks, D.H.: The
inverse problem of electrocardiography. In: Comprehensive Electrocardiology, pp. 299–344.
Springer, New York (2010)
408. Punske, B.B., Ni, Q., Lux, R.L., MacLeod, R.S., Ershler, P.R., Dustman, T.J., Allison, M.J.,
Taccardi, B.: Spatial methods of epicardial activation time determination in normal hearts.
Ann. Biomed. Eng. 31(7), 781–792 (2003)
409. Puwal, S., Roth, B.J.: Forward Euler stability of the bidomain model of cardiac tissue. IEEE
Trans. Biomed. Eng. 54(5), 951–953 (2007)
410. Qu, Z., Garfinkel, A.: An advanced algorithm for solving partial differential equation in
cardiac conduction. IEEE Trans. Biomed. Eng. 46(9), 1166–1168 (1999)
411. Qu, Z., Kill, J., Xie, F., Garfinkel, A., Weiss, J.N.: Scroll wave dynamics in a three-
dimensional cardiac tissue model: roles of restitution, thickness, and fiber rotation. Biophys.
J. 78, 2761–2775 (2000)
412. Qu, Z., Xie, F., Garfinkel, A., Weiss, J.N.: Origins of spiral wave meander and breakup in a
two-dimensional cardiac tissue model. Ann. Biomed. Eng. 28, 755–771 (2000)
413. Quan, W., Evans, S.J., Hastings, H.M.: Efficient integration of a realistic two-dimensional
cardiac tissue model by domain decomposition. IEEE Trans. Biomed. Eng. 45, 372–385
(1998)
414. Quarteroni, A., Valli, A.: Numerical Approximation of Partial Differential Equations.
Springer, Berlin (1994)
415. Ramanathan, C., Ghanem, R.N., Jia, P., Ryu, K., Rudy, Y.: Noninvasive electrocardiographic
imaging for cardiac electrophysiology and arrhythmia. Nat. Med. 10(4), 422–428 (2004)
416. Ranjan, R., Tomaselli, G.F., Marban, E.: A novel mechanism of anode-break stimulation
predicted by bidomain modeling. Circ. Res. 84, 153–156 (1999)
417. Rappel, W.J.: Filament instability and rotational tissue anisotropy: a numerical study using
detailed cardiac models. Chaos 11(1), 71–80 (2001)
418. Rasmusson, R.I., Clark, J.W., Giles, W.R., Robinson, K., Clark, R.B., Shibata, E.F., Campbell,
D.L.: A mathematical model of electrophysiological activity in a bullfrog atrial cell. Am. J.
Physiol. 259, H370–H389 (1990)
419. Rinzel, J., Ermentrout, B.: Analysis of neural excitability and oscillations. In: Kock, C., et al.
(eds.) Methods in Neuronal Modelling: From Synapses to Networks. MIT, Boston (1998)
420. Roberts, D., Scher, A.M.: Effect of tissue anisotropy on extracellular potential fields in canine
myocardium in situ. Circ. Res. 50, 342–351 (1982)
421. Roberts, D., Hersch, L.T., Scher, A.M.: Influence of cardiac fiber orientation on wave front
voltage, conduction velocity and tissue resistivity. Circ. Res. 44, 701–712 (1979)
422. Rogers, J.M., McCulloch, A.D.: A collocation-Galerkin finite element model of cardiac action
potential propagation. IEEE Trans. Biomed. Eng. 41, 743–757 (1994)
423. Rohr, S.: Role of gap junctions in the propagation of the cardiac action potential. Cardiovasc.
Res. 62, 309–322 (2004)
424. Rohr, S.: Cardiac fibroblasts in cell culture systems: myofibroblasts all along? J. Cardiovasc.
Pharmacol. 57(4), 389–399 (2011)
425. Romero, D., Sebastian, R., Bijnens, B.H., et al.: Effects of the Purkinje system and cardiac
geometry on biventricular pacing: a model study. Ann. Biomed. Eng. 38(4), 1388–1398
(2010)
386 References

426. Roth, B.J.: The electrical potential produced by a strand of cardiac muscle: a bidomain
analysis. Ann. Biomed. Eng. 16, 609–637 (1988)
427. Roth, B.J.: A mathematical model of make and break electrical stimulation of cardiac tissue
by a unipolar anode or cathode. IEEE Trans. Biomed. Eng. 42, 1174–1184 (1995)
428. Roth, B.J.: Strength-Interval curve for cardiac tissue predicted using the bidomain model. J.
Cardiovasc. Electrophysiol. 7, 722–737 (1996)
429. Roth, B.J.: Nonsustained reentry following successive stimulation of cardiac tissue through a
unipolar electrode. J. Cardiovasc. Electrophysiol. 8, 768–778 (1997)
430. Roth, B.J., Chen, J.: Mechanism of anode break excitation in the heart: the relative influence
of membrane and electrotonic factors. J. Biol. Syst. 7(4), 541–552 (1999)
431. Roth, B.J., Krassowska, W.: The induction of reentry in cardiac tissue. The missing link: how
electric fields alter transmembrane potential. Chaos 8, 204–220 (1998)
432. Roth, B.J., Patel, S.G.: Effects of elevated extracellular potassium ion concentration on anodal
excitation of cardiac tissue. J. Cardiovasc. Electrophysiol. 14, 1351–1355 (2003)
433. Roth, B.J., Pertsov, A.M.: Hybrid modeling of electrical and optical behavior in the heart.
Physica D 238, 1019–1027 (2009)
434. Roth, B.J., Wikswo, J.P., Jr.: A bidomain model for the extracellular potential and magnetic
field of cardiac tissue. IEEE Trans. Biomed. Eng. 33(4), 467–469 (1986)
435. Roth, B.J., Wikswo, J.P., Jr.: Electrical stimulation of cardiac tissue: a bidomain model with
active membrane properties. IEEE Trans. Biomed. Eng. 41(3), 232–240 (1994)
436. Roth, B.J., Lin, S.-F., Wikswo, J.P., Jr.: Unipolar stimulation of cardiac tissue. J. Electrocar-
diol. 31(Suppl), 6–12 (1998)
437. Roux, B., Allen, T., Berneche, S.: Theoretical and computational models of biological ion
channels. Q. Rev. Biophys. 37(1), 15–103 (2001)
438. Rubinstein, I.: Electro-Diffusion of Ions. SIAM, Philadelphia (1990)
439. Rudy, Y.: The electrocardiogram and its relationship to excitation of the heart. In: Sperelakis,
N. (ed.) Physiology and Pathophysiology of the Heart, 3rd edn., chap. 11, pp. 201–239.
Kluwer Academic, Dordrecht (1995)
440. Rudy, Y.: Noninvasive electrocardiographic imaging of arrhythmogenic substrates in humans.
Circ. Res. 112, 863–874 (2013)
441. Rudy, Y., Messinger-Rapport, B.J.: The inverse problem in electrocardiography: solutions in
terms of epicardial potentials. CRC Crit. Rev. Biomed. Eng. 16(3), 215–268 (1988)
442. Rudy, Y., Oster, H.S.: The electrocardiographic inverse problem. CRC Crit. Rev. Biomed.
Eng. 20, 25–45 (1992)
443. Rudy, Y., Silva, J.R.: Computational biology in the study of cardiac ion channels and cell
electrophysiology. Q. Rev. Biophys. 39(1), 57–116 (2006)
444. Rushmer, R.F.: Structure and Function of the Cardiovascular System, 2nd edn.
W. B. Saunders, Philadelphia (1976)
445. Saad, Y.: Iterative Methods for Sparse Linear Systems, 2nd edn. SIAM, Philadelphia (2003)
446. Sachse, F.B.: Computational Cardiology. Modeling of Anatomy, Electrophysiology, and
Mechanics. LNCS, vol. 2966. Springer, Berlin (2004)
447. Sachse, F.B., Seemann, G. (eds.): Proceedings of the 4th International Conference on
Functional Imaging and Modeling of the Heart, FIMH’07, Salt Lake City, 7–9 June 2007.
LNCS, vol. 4466. Springer, Berlin (2007)
448. Saffitz, J.E., Kanter, H.L., Green, K.G., Tolley, T.K., Beyer, E.C.: Tissue-specific determinants
of anisotropic conduction velocity in canine atrial and ventricular myocardium. Circ. Res. 74,
1065–1070 (1994)
449. Saleheen, H.I., Ng, K.T.: A new three-dimensional finite-difference bidomain formulation for
inhomogeneous anisotropic cardiac tissues. IEEE Trans. Biomed. Eng. 45(1), 15–25 (1998)
450. Sambelashvili, A., Efimov, I.R.: Dynamics of virtual electrode-induced scroll-wave reentry in
a 3D bidomain model. Am. J. Physiol Heart Circ. Physiol. 287, H1570–H1581 (2004)
451. Sambelashvili, A., Nikolsky, V.P., Efimov, I.R.: Virtual electrode theory explains pacing
threshold increase caused by cardiac tissue damage. Am. J. Physiol Heart Circ. Physiol. 286,
H2183–H2194 (2004)
References 387

452. Sampson, K.J., Henriquez, C.S.: Electrotonic influences on action potential duration disper-
sion in small hearts: a simulation study. Am. J. Physiol. Heart Circ. Physiol. 289, H350–H360
(2005)
453. Samson, W., Scher, A.: Mechanism of ST-segment alteration during acute myocardial injury.
Circ. Res. 8, 780–787 (1960)
454. Sanchez-Palencia, E., Zaoui, A.: Homogenization Techniques for Composite Media. Lectures
Notes in Physics, vol. 272. Springer, Berlin (1987)
455. Sanfelici, S.: Convergence of the Galerkin approximation of a degenerate evolution problem
in electrocardiology. Numer. Methods Part. Differ. Equ. 18, 218–240 (2002)
456. Sanfelici, S.: Numerical simulations of fractioned electrograms and pathological cardiac
action potential. J. Theor. Med. 4(3), 167–181 (2002)
457. Savaré, G.: Weak solutions and maximal regularity for abstract evolution inequalities. Adv.
Math. Sci. Appl. 6, 377–418 (1996)
458. Scacchi, S.: A hybrid multilevel Schwarz method for the bidomain model. Comput. Methods
Appl. Mech. Eng. 197(45–48), 4051–4061 (2008)
459. Scacchi, S.: A multilevel hybrid Newton-Krylov-Schwarz method for the Bidomain model of
electrocardiology. Comput. Methods Appl. Mech. Eng. 200(5–8), 717–725 (2011)
460. Scacchi, S., Colli Franzone, P., Pavarino, L.F., Taccardi, B.: A reliability analysis of cardiac
repolarization time markers. Math. Biosci. 219(2), 113–128 (2009)
461. Scacchi, S., Colli Franzone, P., Pavarino, L.F., Taccardi, B.: Computing cardiac recovery
maps from electrograms and monophasic action potentials under heterogeneous and ischemic
conditions. Math. Model Methods Appl. Sci. 20(7), 1089–1127 (2010)
462. Scher, A.M.: Excitation of the heart. In: Nelson, C.V., Geselowitz, D.B. (eds.) The Theoretical
Basis of Electrocardiology, pp. 44–67. Clarendon, Oxford (1976)
463. Schmitt, O.H.: Biological information processing using the concept of interpenetrating
domains. In: Leibovich, K.N. (ed.) Information Processing in the Nervous System, pp. 325–
331. Springer, New York (1969)
464. Schuss, Z., Nadler, B., Eisenberg, R.S.: Derivation of Poisson and Nernst-Planck equations in
a bath and channel from a molecular model. Phys. Rev. E 64, 036116 (2001)
465. Scollan, D.F., Holmes, A., Zhang, J., Winslow, R.L.: Reconstruction of cardiac ventricular
geometry and fiber orientation using magnetic resonance imaging. Ann. Biomed. Eng. 28(8),
934–944 (2000)
466. Seemann, G., Hoeper, C., Sachse, F.B., Doessel, O., Holden, A.V.: Heterogeneous three-
dimensional anatomical and electrophysiological model of human atria. Philos. Trans. R. Soc.
A 364, 1465–1481 (2006)
467. Seemann, G., Sachse, F.B., Weiss, D.L., Dossel, O.: Quantitative reconstruction of cardiac
electromechanics in human myocardium: regional heterogeneity. J. Cardiovasc. Electrophys-
iol. 14(10), S219–S228 (2003)
468. Seger, M., Fischer, G., Modre, R., Messnarz, B., Hanser, F., Tilg, B.: Lead field computation
for the electrocardiographic inverse problem – finite element versus boundary elements.
Comput. Methods Prog. Biomed. 77, 241–252 (2005)
469. Sepulveda, N.G., Roth, B.J., Wikswo, J.P., Jr.: Current injection into a two-dimensional
anisotropic bidomain. Biophys. J. 55, 987–999 (1989)
470. Sermesant, M., Konukoglu, E., Delingette, H., Coudiere, Y., Khinchapatnam, P., Rhode,
K.S., Razzavi, R., Ayache, N.: An anisotropic multi-front fast marching method for real-time
simulation in cardiac electrophysiology. In: Sachse, F.B., Seemann, G. (eds.) FIMH’07, Salt
Lake City, 7–9 June 2007. LNCS, vol. 4466, pp. 160–169. Springer, Berlin (2007)
471. Sethian, J.A.: Level Set Methods: Evolving Interfaces in Geometry, Fluid Mechanics,
Computer Vision and Material Sciences, 2nd edn. Cambridge University Press, Cambridge
(1999)
472. Severi, S., Fantini, M., Charawi, L.A., Di Francesco, D.: An updated computational model
of rabbit sinoatrial action potential to investigate the mechanisms of heart rate modulation.
J. Physiol. 590, 4483–4499 (2012)
388 References

473. Shannon, T.R., Wang, F., Puglisi, J., Weber, C., Bers, D.M.: A mathematical treatment of
integrated Ca dynamics within the ventricular myocyte. Biophys. J. 87(5), 3351–3357 (2004);
Erratum Biophys. J. 102(8), 1996–2001 (2012)
474. Shaw, R.M., Rudy, Y.: Ionic mechanisms of propagation in cardiac tissue. Role of the sodium
and L-type calcium currents during reduced excitability and decreased gap junction coupling.
Circ. Res. 81(5), 727–741 (1997)
475. Shenasa, M., Hindricks, G., Borggrefe, M., Breithardt, G., Josephson, M.E.: Cardiac Map-
ping, 4th edn. Wiley-Blackwell, Chichester (2012)
476. Shibata, N., Chen, P.S., Dixon, E.G., Wolf, P.D., Danieley, N.D., Smith, W.M., Ideker, R.E.:
Influence of shock strength and timing on induction of ventricular arrhythmias in dogs. Am.
J. Physiol. 255, H891–H901 (1988)
477. Shou, G., Xia, L., Jiang, M.: Total variation regularization in electrocardiographic mapping.
In: Li, K., et al. (eds.) Life System Modeling and Intelligence Computing. LNMI, vol. 6330,
pp. 51–59. Springer, Berlin (2010)
478. Skipa, O., Nalbach, M., Sachse, F., Werner, C., Dossel, O.: Transmembrane potential
reconstruction in anisotropic heart model. Int. J. Bioelectromagn. 4(2), 17–18 (2002)
479. Sicouri, S., Antzelevich, C.: A subpopulation of cells with unique electrophysiological
properties in the deep subepicardium of the canine ventricle. The M cell. Circ. Res. 68, 1729–
1741 (1991)
480. Sidorov, V.Y., Woods, M.C., Baudenbacher, P., Baudenbacher, F.: Examination of stimulation
mechanism and strength-interval curve in cardiac tissue. Am. J. Physiol. Heart Circ. Physiol.
289, H2602–H2615 (2005)
481. Sigg, D.C., Iaizzo, P.A., Xiao, Y.-F., He, B.: Cardiac Electrophysiology Methods and Models.
Springer, New York (2010)
482. Silva, J.R., Pan, H., Wu, D., Nekouzadeh, A., Decker, K.F., Cui, J., Baker, N.A., Sept, D.,
Rudy, Y.: A multiscale model linking ion-channel molecular dynamics and electrostatics to
the cardiac action potential. Proc. Nat. Acad. Sci. 106(27), 11102–11106 (2009)
483. Simms, H.D., Geselowitz, D.B.: Computation of heart surface potentials using the surface
source model. J. Cardiovasc. Electrophysiol. 6, 522–531 (1995)
484. Singer, A., Gillespie, D., Norbury, J., Eisenberg, R.S.: Singular perturbation analysis of the
steady-state Poisson-Nernst-Planck system: applications to ion channels. Eur. J. Appl. Math.
19, 541–560 (2008)
485. Skouibine, K., Krassowska, W.: Increasing the computational efficiency of a bidomain model
of defibrillation using a time-dependent activating function. Ann. Biomed. Eng. 28, 772–780
(2000)
486. Skouibine, K., Trayanova, N., Moore, P.: Anode/cathode make and break phenomena in a
model of defibrillation. IEEE Trans. Biomed. Eng. 46(7), 769–777 (1999)
487. Skouibine, K., Trayanova, N., Moore, P.: A numerically efficient model for the simulation
of defibrillation in an active bidomain sheet of myocardium. Math. Biosci. 166(1), 85–100
(2000)
488. Smaill, B.H., Zhao, J., Trew, M.L.: Three-dimensional impulse propagation in myocardium.
Arrhythmogenic mechanisms at the tissue level. Circ. Res. 112, 834–848 (2013)
489. Smith, B.F., Bjørstad, P., Gropp, W.D.: Domain Decomposition: Parallel Multilevel Methods
for Elliptic Partial Differential Equations. Cambridge University Press, Cambridge (1996)
490. Smith, N.P., Pullan, A.J., Hunter, P.J.: An anatomically based model of transient coronary
blood flow in the heart. SIAM J. Appl. Math. 62(3), 990–1018 (2002)
491. Smith, N.P., Nickerson, D.P., Crampin, E.J., Hunter, P.J.: Multiscale computational modelling
of the heart. Acta Numer. 371–431 (2004)
492. Soravia, J.P., Souganidis, P.E.: Phase-field theory for Fitzhugh-Nagumo type systems. SIAM
J. Math. Anal. 27(5), 1341–1359 (1996)
493. Spach, M.S., Dolber, P.C.: Relating extracellular potentials and their derivatives to anisotropic
propagation at microscopic level in human cardiac muscle. Evidence for electrical uncoupling
of side-to-side fiber connections with increasing age. Circ. Res. 58, 356–371 (1986)
References 389

494. Spiteri, R.J., Dean, R.C.: On the performance of an implicit-explicit Runge-Kutta method in
models of cardiac electrical activity. IEEE Trans. Biomed. Eng. 55(5), 1488–1495 (2008)
495. Stampacchia, G.: Boundary value problems for some degenerate-elliptic operators. Ann.
Math. Pura Appl. LXXX(IV), 1–122 (1968)
496. Steinhaus, B.M.: Estimating cardiac transmembrane activation and recovery times from
unipolar and bipolar extracellular electrograms: a simulation study. Circ. Res. 64(3), 449–
462 (1989)
497. Steinhaus, B.M., Spitzer, K.W., Isomura, S.: Action potential collision in heart tissue.
Computer simulations and tissue experiments. IEEE Trans. Biomed. Eng. 32(10), 731–742
(1985)
498. Streeter, D.: Gross morphology and fiber geometry in the heart. In: Berne, R.M. (ed.)
Handbook of Physiology, vol. 1, sect. 2, pp. 61–112. Williams & Wilkins, Philadelphia (1979)
499. Sundnes, J., Lines, G.T., Mardal, K., Tveito, A.: Multigrid block preconditioning for a coupled
system of partial differential equations modeling the electrical activity in the heart. Comput.
Methods Biomech. Biomed. Eng. 5, 397–409 (2002)
500. Sundnes, J., Lines, G., Lines, G.T., Grottum, P., Tveito, A.: Electrical activity in the human
heart. In: Langtangen, H.P., Tveito, A. (eds.) Advanced Topics in Computational Partial
Differential Equations. LNCSE, vol. 33, chap. 10, pp. 401–449. Springer, Berlin (2004)
501. Sundnes, J., Lines, G.T., Tveito, A.: An operator splitting method for solving the bidomain
equations coupled to a volume conductor model for the torso. Math. Biosci. 194(2), 233–248
(2005)
502. Sundnes, J., Lines, G.T., Cai, X., Nielsen, B.F., Mardal, K.-A., Tveito, A.: Computing the
Electrical Activity of the Heart. Springer, Berlin (2006)
503. Sundnes, J., Nielsen, B.F., Mardal, K.A., Lines, G.T., Mardal, K.A., Tveito, A.: On the
computational complexity of the bidomain and the monodomain models of electrophysiology.
Ann. Biomed. Eng. 34(7), 1088–1097 (2006)
504. Taccardi, B., Punske, B.B.: Body surface potential mapping. In: Zipes, D., Jalife, J. (eds.)
Cardiac Electrophysiology. From cell to Bedside, 4th edn., pp. 803–811. W. B. Saunders,
Philadelphia (2004)
505. Taccardi, B., de Ambroggi, L., Viganotti, C.: Body surface mapping of heart potentials. In:
Nelson, C.V., Geselowitz, D.B. (eds.) The Theoretical Basis of Electrocardiology, pp. 436–
466. Clarendon, Oxford (1976)
506. Taccardi, B., Lux, R.L., Ershler, P.R., MacLeod, R.S., Vyhmeister, Y.: Effect of myocardial
fiber direction on 3-D shape of excitation wavefronts and associated potential distributions in
ventricular walls. Circulation 86, I-752 (1992)
507. Taccardi, B., Macchi, E., Lux, R.L., Ershler, P.R., Spaggiari, S., Baruffi, S., Vyhmeister,
Y.: Effect of myocardial fiber direction on epicardial potentials. Circulation 90, 3076–3090
(1994)
508. Taccardi, B., Lux, R.L., Ershler, P.R., MacLeod, R.S., Vyhmeister, Y.: Modern views on the
spread of excitation in anisotropic heart muscle. Jpn. Heart J. 35, 31–35 (1994)
509. Taccardi, B., Lux, R.L., Ershler, P.R., MacLeod, R.S., Dustman, T.J., Ingebrigtsen, N.:
Anatomical architecture and electrical activity of the heart. Acta Cardiol. 52, 91–105 (1997)
510. Taccardi, B., Veronese, S., Colli Franzone, P., Guerri, L.: Multiple components in the
unipolar electrocardiogram: a simulation study in a three-dimensional model of ventricular
myocardium. J. Cardiovasc. Electrophysiol. 9, 1062–1084 (1998)
511. Taccardi, B., Punske, B., Lux, R., MacLeod, R., Ershler, P., Dustman, T., Vyhmeister, Y.:
Useful lesson from body surface mapping on body. J. Cardiovasc. Electrophysiol. 9(7), 773–
786 (1998)
512. Taccardi, B., Punske, B., Helie, F., MacLeod, R., Lux, R., Ershler, P., Dustman, T.,
Vyhmeister, Y.: Epicardial recovery sequences and excitation recovery intervals during paced
beats. Role of myocardial architecture. PACE 22(4), part II: 833 (1999)
513. Taccardi, B., Punske, B.B., MacLeod, R.S., Ni, Q.: Visualization, analysis and physiological
interpretation of three-dimensional cardiac electric fields. In: Proceedings of the 2nd Joint
EMBS/BMSE Conference, Houston, Oct 2002. vol. 2, pp. 1366–1367 (2002)
390 References

514. Taccardi, B., Punske, B.B., Colli Franzone, P.: Cardiac potential mapping. In: Proceedings
of EMBS/25th IEEE Annual International Conference, Cancun, 17–21 Sept 2003. vol. 4,
pp. 3749–3752 (2003)
515. Taccardi, B., Punske, B.B., Sachse, F., Tricoche, X., Colli Franzone, P., Pavarino, L.F.,
Zabawa, C.: Intramural activation and repolarization sequences in canine ventricles. Experi-
mental and simulation studies. J. Electrocardiol. 38, 131–137 (2005)
516. Taccardi, B., Punske, B., Macchi, E., MacLeod, R.S., Ershler, P.R.: Epicardial and intramural
excitation during ventricular pacing: effects of myocardial structure. Am. J. Physiol. Heart.
Circ. Physiol. 294, H1753–H1766 (2008)
517. Taggart, P., Sutton, P., Opthof, T., Coronel, R., Trimlett, R., Pugsley, W., Kallis, P.: Transmural
repolarization in the left ventricle in humans during normoxia and ischemia. Cardiovasc. Res.
50, 454–462 (2001)
518. Tarkhanov, N.N.: The Cauchy Problem for Solutions of Elliptic Equations. Akademic Verlag
Gmbh, Berlin (1995)
519. ten Tusscher, K.H.W.J., Panfilov, A.V.: Alternans and spiral breakup in a human ventricular
tissue model. Am. J. Physiol. Heart Circ. Physiol. 291(3), H1088–H1100 (2006)
520. ten Tusscher, K.H.W.J., Panfilov, A.V.: Modelling of the ventricular conduction system. Prog.
Biophys. Mol. Biol. 96(1–3), 152–170 (2008)
521. ten Tusscher, K., Noble, D., Noble, P.J., Panfilov, A.V.: A model for human ventricular tissue.
Am. J. Phys. Heart. Circ. Physiol. 286(4), H1573–H1589 (2004)
522. Tomlinson, K.A., Hunter, P.J., Pullan, A.J.: A finite element method for an eikonal equation
model of myocardial excitation wavefront propagation. SIAM J. Appl. Math. 63(1), 324–350
(2002)
523. Toselli, A., Widlund, O.B.: Domain Decomposition Methods: Algorithms and Theory.
Computational Mathematics, vol. 34. Springer, Berlin (2004)
524. Transgenstein, J.A., Kim, C.: Operator splitting and adaptive mesh refinement for the Luo–
Rudy I model. J. Comput. Phys. 196, 645–679 (2004)
525. Trayanova, N.A.: Defibrillation of the heart: insights into mechanisms from modelling
studies. Exp. Physiol. 91(2), 323–337 (2006)
526. Trayanova, N., Eason, J., Aguel, F.: Computer simulations of cardiac defibrillation: a look
inside the heart. Comput. Vis. Sci. 4, 259–270 (2002)
527. Trayanova, N.A., Constantino, J., Gurev, V.: Electromechanical models of the ventricles. Am.
J. Physiol. Heart Circ. Physiol. 301(2), H279–H286 (2011)
528. Trew, M., Le Grice, I., Smaill, B., Pullan, A.: A finite volume method for modeling
discontinuous electrical activation in cardiac tissue. Ann. Biomed. Eng. 33(5), 590–602
(2005)
529. Trew, M., Smaill, B., Bullivant, D., Hunter, P., Pullan, A.: A generalized finite difference
method for modeling cardiac electrical activation on arbitrary, irregular computational
meshes. Math. Biosci. 198(2), 169–189 (2005)
530. Trew, M.L., Caldwell, B.J., Sands, G.B., Hooks, D.A., Tai, D.C.-S., Austin, T.M., LeGrice,
I.J., Pullan, A.J., Smaill, B.H.: Cardiac electrophysiology and tissue structure: bridging the
scale gap with a joint measurement and modelling paradigm. Exp. Physiol. 91(2), 355–370
(2006)
531. Tung, L.: A bidomain model for describing ischemic myocardial D.C. potentials. Ph.D.
dissertation, MIT, Cambridge, MA (1978)
532. Tyson, J.J., Keener, J.P.: Singular perturbation theory of traveling waves in excitable media.
(A review). Physica D 32, 327–361 (1988)
533. Ueda, N., Zipes, D.P., Wu, J.: Functional and transmural modulation of M cell behavior in
canine ventricular wall. Am. J. Physiol. (Heart Circ. Physiol) 287, H2569–H2575 (2004)
534. van Dam, P.M., Oostendorp, T.F., Linnenbank, A.C., van Oosterom, A.: Non-invasive imaging
of cardiac activation and recovery. Ann. Biomed. Eng. 37(9), 1739–1756 (2009)
535. van Oosterom, A.: Cell models–Macroscopic source descriptions. In: Macfarlane, P.W.,
Lawrie, T.D.V. (eds.) Comprehensive Electrocardiology, pp. 155–179. Pergamon, Oxford
(1989)
References 391

536. van Oosterom, A.: Forward and inverse problems in electrocardiography. In: Panfilov, A.V.,
Holden, A.V. (eds.) Computational Biology of the Heart. Wiley, New York (1997)
537. van Oosterom, A.: Genesis of the T wave as based on an equivalent surface source model. J.
Electrocardiol. 34(Suppl), 217–227 (2001)
538. van Oosterom, P.: Genesis of the T wave as based on an equivalent surface source model. J.
Electrocardiol. 34, 217–227 (2001)
539. Varghese, A., Sell, G.R.: A conservation principle and its effect on the formulation of Na–Ca
exchanger current in cardiac cells. J. Theor. Biol. 189, 33–40 (1997)
540. Veneroni, M.: Reaction-diffusion systems for the microscopic cellular model of the cardiac
electric field. Math. Methods Appl. Sci. 29, 1631–1661 (2006)
541. Veneroni, M.: Reaction-Diffusion systems for the macroscopic Bidomain model of the cardiac
electric field. Nonlinear Anal. Real World Appl. 10(2), 849–868 (2009)
542. Vigmond, E.J.: The electrophysiologic basis of MAP recordings. Cardiovasc. Res. 68, 502–
503 (2005)
543. Vigmond, E.J., Leon, L.J.: Computational efficient model for simulating electrical activity in
cardiac tissue with fiber rotation. Ann. Biomed. Eng. 27, 160–170 (1999)
544. Vigmond, E.J., Aguel, F., Trayanova, N.A.: Computational techniques for solving the
bidomain equations in three dimensions. IEEE Trans. Biomed. Eng. 49(11), 1260–1269
(2002)
545. Vigmond, E.J., Weber dos Santos, R., Prassl, A.J., Deo, M., Plank, G.: Solvers for the cardiac
bidomain equations. Prog. Biophys. Mol. Biol. 96, 3–18 (2008)
546. Viswanathan, P.C., Shaw, R.M., Rudy, Y.: Effects of IKr and IKs heterogeneity on action
potential duration and its rate dependence. A simulation study. Circulation 99(18), 2466–
2474 (1999)
547. Wagner, J., Keizer, J.: Effects of rapid buffers on Ca2C diffusion and CaC2 oscillations.
Biophys. J. 67, 447–456 (1994)
548. Wang, Y., Rudy, Y.: Action potential propagation in inhomogeneous cardiac tissue: safety
factor considerations and ionic mechanism. Am. J. Physiol. Heart Circ. Physiol. 278, H1019–
H1029 (2000)
549. Wang, D., Kirby, R.M., Johnson, C.R.: Finite-element-based discretization and regularization
strategies for 3-D inverse electrocardiography. IEEE Trans. Biomed. Eng. 58(6), 1827–1838
(2011)
550. Wang, L., Wong, K.C., Zhang, H., Liu, H., Shi, P.: Noninvasive computational imaging
of cardiac electrophysiology for 3-D infarct. IEEE Trans. Biomed. Eng. 58(4), 1033–1043
(2011)
551. Wang, L., Dawoud, F., Yeung, S.-K., Shi, P., Wong, K., Lardo, A.: Transmural imaging of
ventricular action potential and post-infarction scars in swine hearts. IEEE Trans. Med. Image
32(4), 731–747 (2013)
552. Weber dos Santos, R., Plank, G., Bauer, S., Vigmond, E.J.: Parallel multigrid preconditioner
for the cardiac bidomain model. IEEE Trans. Biomed. Eng. 51(11), 1960–1968 (2004)
553. Weissenburger, J., Nesterenko, V., Antzelevitch, C.: Transmural heterogeneity of ventricular
repolarization under baseline and long QT conditions in the canine heart in vivo: torsade de
Pointes develops with halotane but not pentobarbital anesthesia. J. Cardiovasc. Electrophys-
iol. 11, 290–304 (2000)
554. White, C.S., Haramati, L.B., Jen-Sho Chen, J., Levsky, J.M.: Cardiac Imaging. Oxford
University Press, Oxford (2014)
555. Whiteley, J.P.: An efficient numerical technique for the solution of the monodomain and
bidomain equations. IEEE Trans. Biomed. Eng. 53(11), 2139–2147 (2006)
556. Whiteley, J.P.: Physiology driven adaptivity for the numerical solution of the bidomain
equations. Ann. Biomed. Eng. 35(9), 1510–1520 (2007)
557. Wikswo, J.P., Jr.: Tissue anisotropy, the cardiac bidomain, and the virtual cathode effect.
In: Zipes D.P., Jalife J. (eds.) Cardiac Electrophysiology: From Cell to Bedside, 2nd edn.,
pp. 348–361. W. B. Saunders, Philadelphia (1994)
392 References

558. Wikswo, J.P., Jr., Roth, B.J.: Virtual electrode theory of pacing. In: Efimov, I.R., Kroll, M.W.,
Tchou, P.J. (eds.) Cardiac Bioelectric Therapy, Chap. 4.3, pp. 283–330. Springer, New York
(2009)
559. Wikswo, J.P., Jr., Lin, S.-F., Abbas, R.A.: Virtual electrodes in cardiac tissue: a common
mechanism for anodal and cathodal stimulation. Biophys. J. 69, 2195–2210 (1995)
560. Wilson, L.D., Jeyaraj, D.: Controversies in measuring repolarization using extracellular
recordings: why should be care. Heart Rhythm 3(9), 1051–1052 (2006)
561. Winfree, A.T.: Sudden cardiac death: a problem in topology. Sci. Am. 248, 144–161 (1983)
562. Winfree, A.T.: The Geometry of Biological Time, 2nd edn. Springer, New York (2001)
563. Winslow, R.L., Rice, J., Jafri, S., Marban, E., O’Rourke, B.: Mechanisms of altered excitation-
contraction coupling in canine tachycardia-induced heart failure, II – model studies. Circ. Res.
84(5), 571–586 (1999)
564. Wit, A.L., Janse, M.J.: The Ventricular Arrhythmias of Ischemia and Infarction: Electrophys-
iological Mechanisms. Futura Publishing Co., New York (1993)
565. Wolferth, C.C., Bettet, S., Livezey, M.M., Murphy, F.: Negative displacement of the RS-T seg-
ment in the electrocardiogram and its relationships to positive displacement: an experimental
study. Am. Heart J. 29, 220–244 (1945)
566. Wyatt, R.P.: Comparison of estimates of activation and recovery times from bipolar and
unipolar electrograms to in vivo transmembrane action potential durations. In: Proceedings
of IEEE Engineering in Medicine and Biology Society, 2nd Annual Conference, Washington,
DC, pp. 22–25 (1980)
567. Xia, Y., Kongstad, O., Hertvig, E., Li, Z., Holm, M., Olsson, B., Yuan, S.: Activation
recovery time measurements in evaluation of global sequence and dispersion of ventricular
repolarization. J. Electrocardiol. 38, 28–35 (2005)
568. Xie, F., Qu, Z.L., Yang, J., Baher, A., Weiss, J.N., Garfinkel, A.: A simulation study of the
effects of cardiac anatomy in ventricular fibrillation. J. Clin. Invest. 113, 686–693 (2004)
569. Xu, A., Guevara, M.R.: Two forms of spiral-wave reentry in an ionic model of ischemic
ventricular myocardium. Chaos 8(1), 157–174 (1998)
570. Yamashita, Y., Geselowitz, D.B.: Source–field relationships for cardiac generators on the heart
surface based on their transfer coefficients. IEEE Trans. Biomed. Eng. 32, 964–970 (1985)
571. Yan, G.-X., Antzelevitch, C.: Cellular basis for the normal T wave and the electrocardio-
graphic manifestations of the Long-QT syndrome. Circulation 98, 1928–1936 (1998)
572. Yan, G.X., Shimizu, W., Antzelevitch, C.: Characteristics and distribution of M cells in
arterially perfused canine left ventricular wedge preparations. Circulation 98(18), 1921–1927
(1998)
573. Yehia, A.R., Jeandupeaux, D., Alonso, F., Guevara, M.R.: Hysteresis and bistability in the
direct transition from 1:1 to 2:1 rhythm in periodically driven single ventricular cells. Chaos
9, 916–931 (1999)
574. Ying, W.J., Rose, D.J., Henriquez, C.S.: Efficient fully implicit time integration methods for
modeling cardiac dynamics. IEEE Trans. Biomed. Eng. 55(12), 2701–2711 (2008)
575. Young, R.J., Panfilov, A.: Anisotropy of wave propagation in the heart can be modeled by a
Riemannian electrophysiological metric. Proc. Natl. Acad. Sci. 107(34), 15063–15068 (2010)
576. Young, A.A., LeGrice, I.J., Young, M.A., Smaill, B.H.: Extended confocal microscopy of
myocardial laminae and collagen network. J. Microscop. 192, 139–150 (1998)
577. Yu, H.: Solving parabolic problems with different time steps in different regions in space
based on domain decomposition methods. Appl. Numer. Math. 30(4), 475–491 (1999)
578. Yu, H.: A local space-time adaptive scheme in solving two-dimensional parabolic problems
based on domain decomposition methods. SIAM J. Sci. Comput. 23(1), 304–322 (2001)
579. Yu, H., Chang, F., Cohen, I.S.: Pacemaker i(f) in adult canine cardiac ventricular myocytes. J.
Physiol. 485, 469–483 (1995)
580. Yue, A.M., Betts, T.R., Roberts, P.R., Morgan, J.M.: Global dynamic coupling of activation
and repolarization in human ventricle. Circulation 112, 2592–2601 (2005)
581. Zampini, S.: Balancing Neumann-Neumann methods for the cardiac Bidomain model. Numer.
Math. 123, 363–393 (2013)
References 393

582. Zampini, S.: Dual-primal methods for the cardiac bidomain model. Math. Model Methods
Appl. Sci. 24(4), 667–696 (2014)
583. Zeng, J., Laurita, K.R., Rosenbaum, D.S., Rudy, Y.: Two Components of the delayed rectifier
K C current in ventricular myocytes of the guinea pig type. Theoretical formulation and their
role in repolarization. Circ. Res. 77, 140–152 (1995)
584. Zhang, X.: Multilevel Schwarz methods. Numer. Math. 63(4), 521–539 (1992)
585. Zhang, H., Holden, A.V., Kodama, I., Honjo, H., Lei, M., Varghese, T., Boyett, M.R.:
Mathematical models of action potentials in the periphery and center of the rabbit sinoatrial
node. Am. J. Physiol. Heart. Circ. Physiol. 279, H397–H421 (2000)
586. Zipes, D., Jalife, J.: Cardiac Electrophysiology, 2nd edn. W. B. Saunders, Philadelphia (1995)
587. Zipes, D., Jalife, J.: Cardiac Electrophysiology, 3rd edn. W. B. Saunders, Philadelphia (2000)
588. Zipes, D., Jalife, J.: Cardiac Electrophysiology: From Cell to Bedside, 4th edn.
W. B. Saunders, Philadelphia (2004)
589. Zipes, D., Jalife, J.: Cardiac Electrophysiology, 5th edn. W. B. Saunders, Philadelphia (2009)
590. Zipes, D., Jalife, J.: Cardiac Electrophysiology: From Cell to Bedside, 6th edn. W. B.
Saunders, Philadelphia (2013)
591. Zozor, S., Blanc, O., Jacquemet, V., Virag, N., Vesin, J., Pruvot, E., Kappenberger, L.,
Henriquez, C.: A numerical scheme for modeling wavefront propagation on a monolayer
of arbitrary geometry. IEEE Trans. Biomed. Eng. 50(4), 412–420 (2003)
Index

Additive Schwarz preconditioners, 220 Block preconditioners, 241

Aliev-Panfilov model, 60 block - diagonal, 242
Angiocardiography, 17 block - factorized, 242
Anode break, 250 Body Surface Maps (BSM), 175
Anode make, 250 Bundle-branch block, 16
APD dispersion, 10
apex-to-base dispersion, 11
righ-left ventricle, 10 Cable equation, 77
transmural dispersion, 11 Cardiac imaging, 17
ATPase, 21 Cardiac regeneration, 6
Atrial fibrillation, 16 Cardiac source splitting, 162
Atrial flutter, 16 Cardiac sources, 149
Atrial hypertrophy, 17 Cardiac stem cells, 6
Atrioventricular block, 16 Cathode break, 250
Atrioventricular node (AVN), 3 Cathode make, 250
Augmented leads, 12 Cellular membrane, 21
electrical circuit model, 35
Collagen, 5
Beeler-Reuter model, 45 Computed tomography angiography (CTA),
Bidomain model 18
parallel solvers, 207, 227 Connexins, 6
PE-formulation, 108 Coronary catheterization, 18
PP-formulation, 107
Schwarz preconditioners, 220
space discretization, 193 Diffusion tensor magnetic resonance imaging
time discretization, 194 (DT-MRI), 7
decoupled implicit, 195 Domain decomposition methods, 215
fully implicit, 194
semi-implicit (IMEX), 196
variational formulation, 192 Echocardiogram, 18
well-posedness results, 108 Eikonal models, 125
Bifurcation diagram, 70 numerical approximation, 204
FHN model, 70, 71 Eikonal-curvature equation, 130
HH model, 73 Eikonal-diffusion equation, 131
Morris-Lecar model, 72 Einthoven standard leads, 12

© Springer International Publishing Switzerland 2014 395

P. Colli Franzone et al., Mathematical Cardiac Electrophysiology, MS&A 13,
DOI 10.1007/978-3-319-04801-7
396 Index

Electrocardiogram (ECG), 13 Multilevel Schwarz preconditioners, 223

bipolar, 320 Multiplicative Schwarz preconditioners, 234
morphology, 318
unipolar, 318
Extracellular matrix, 6 Nernst potential, 24
Nernst-Planck equation, 23

Fiber architecture, 7
Oblique dipole layer, 168
Fibroblasts, 5
Optical mapping, 19
FitzHugh-Nagumo (FHN) model, 60
phase-plane analysis, 62
Frequency diagram, 71 P wave, 13
FHN, 72 Parallel Bidomain solvers, 227
Poisson-Nernst-Planck (PNP) equation, 29
Precordial leads, 13
Gap junctions, 6 Purkinje fibers, 4
Gating variables, 36
Goldman-Hodgkin-Katz equation, 24
QRS complex, 14
QT interval, 15
Heart surface source model, 157
Hodgkin-Huxley (HH) model, 41
Homogenization, 98 Reentry phenomena, 348
Hybrid monophasic action potential (HMAP), anatomical reentry, 348
325 circus movement, 348
Hybrid Schwarz preconditioners, 234 critical point theory, 352
cross gradient stimulation, 353
figure-8 reentry, 358
Infarction, 17 functional reentry, 348
Inverse problem of Electrocardiology, 175 pinwheel experiment, 348
in terms of potential, 176 rotors, 351
in terms of wavefront, 187 scroll wave, 352
Ionic channels, 21 scroll wave breakup, 352, 353
Ionic currents, 36 spiral tip, 353
Ionic models, 40 spiral wave, 353
Relaxed Monodomain model, 132
Ischemia, 17
Roger-McCulloch model, 60

Luo-Rudy model Schwarz theory, 215

LR1 model, 46 Sinoatrial node (SAN), 2
LRd dynamic model, 49 Sinus arrhythmia, 15
model updates, 54 ST segment, 14

M cell, 11 T wave, 15
Magnetic resonance imaging (MRI), 19 morphology, 315
Mitchell-Schaeffer model, 60 Transmembrane action potential (TAP), 323
Monodomain model activation markers, 324
space discretization, 193 recovery markers, 324
variational formulation, 192 Transmembrane potential, 21
Monophasic action potential (MAP), 325 Traveling waves, 86
recovery markers, 325 fronts, 86
Morris-Lecar model, 60 pulses, 89
Index 397

U wave, 15 Voltage clamp, 37

Ventricular fibrillation, 16 Wiggers diagram, 1

Ventricular hypertrophy, 17 Wilson central terminal, 13
Ventricular tachycardia, 16 Wolff-Parkinson-White (WPW) syndrome,
Virtual electrode polarization (VEP), 250 16
MS&A – Modeling, Simulation and Applications

Series Editors:

Alfio Quarteroni Anthony T. Patera

École Polytechnique Fédérale Massachusetts Institute of Technology
de Lausanne (Switzerland) Cambridge, MA (USA)
Tom Hou Enrique Zuazua
California Institute of Technology Basque Center for Applied
Pasadena, CA (USA) Mathematics
Bilbao (Spain)
Claude Le Bris
École des Ponts Paris Tech
Paris (France)

Editor at Springer:
Francesca Bonadei
[email protected]

1. L. Formaggia, A. Quarteroni, A. Veneziani (Eds.)

Cardiovascular Mathematics
2009, XIV+522 pp, ISBN 978-88-470-1151-9
2. A. Quarteroni
Numerical Models for Differential Problems
2009, XVI+602 pp, ISBN 978-88-470-1070-3
3. M. Emmer, A. Quarteroni (Eds.)
MATHKNOW
2009, XII+264 pp, ISBN 978-88-470-1121-2
4. A. Alonso Rodríguez, A. Valli
Eddy Current Approximation of Maxwell Equations
2010, XIV+348 pp, ISBN 978-88-470-1934-8
5. D. Ambrosi, A. Quarteroni, G. Rozza (Eds.)
Modeling of Physiological Flows
2012, X+414 pp, ISBN 978-88-470-1934-8
6. W. Liu
Introduction to Modeling Biological Cellular Control Systems
2012, XII+268 pp, ISBN 978-88-470-2489-2
 7. B. Maury
The Respiratory System in Equations
2013, XVIII+276 pp, ISBN 978-88-470-5213-0
8. A. Quarteroni
Numerical Models for Differential Problems, 2nd Edition
2014, XX+656pp, ISBN 978-88-470-5521-6
9. A. Quarteroni, G. Rozza (Eds.)
Reduced Order Methods for modeling and computational reduction
2014, X+332pp, ISBN 978-3-319-02089-1
10. J. Xin, Y. Qi
An Introduction to Mathematical Modeling and Signal Processing in Speech
and Hearing Sciences
2014, XII+208pp, ISBN 978-3-319-03085-2
11. L. Beirão da Veiga, K. Lipnikov, G. Manzini
The Mimetic Finite Difference Method for Elliptic Problems
2014, XVI+392pp, ISBN 978-3-319-02662-6
12. E. Cristiani, B. Piccoli, A.Tosin
Multiscale Modeling of Pedestrian Dynamics
2014, XVI+260pp, ISBN 978-3-319-06619-6
13. P. Colli Franzone, L.F. Pavarino, S. Scacchi
Mathematical Cardiac Electrophysiology
2014, XIV+399pp, ISBN 978-3-319-04800-0

For further information, please visit the following link:

http://www.springer.com/series/8377