c Pleiades Publishing, Ltd., 2007.

ISSN 1990-4789, Journal of Applied and Industrial Mathematics, 2007, Vol. 1, No. 4, pp. 18.
c S.Yu. Andreev, V.A. Kochegurov, 2005, published in Sibirskii Zhurnal Industrialnoi Matematiki, 2005, Vol. VIII, No. 2(22), pp. 311.
Original Russian Text

Algorithms for Intraoperative Modeling

of the Atrial Excitation Dynamics
S. Yu. Andreev* and V. A. Kochegurov
Tomsk Polytechnic University, pr. Lenina 30, Tomsk, 634050, Russia
Received March 2, 2005

AbstractThe problem of the atrial excitation modeling in clinical conditions is studied. The basic
model requirements and the existing methods for modeling the excitation dynamics of contractile
myocardium are considered. The cellular automata theory serves as the basis of the model. The
calculations are carried out on a rectangular grid. For each pair of the cellular automaton elements,
the time delay of the excitation transfer is computed. Such an approach allows to adapt the model
to the particular characteristics of the research subject taking into account all data obtained by the
electrophysiological tests.
DOI: 10.1134/S1990478907040011

INTRODUCTION
An intensive enhancement of the diagnostic and treatment methods for arrhythmias at the beginning
of 90ths gave rise to the development of some methods of endocardial mapping of the heart cavities and
the methods of evaluation of the excitation spread along the myocardium. However, it is necessary to
note that in some situations the only intraoperative evaluation of the excitation spread is not enough.
In this connection, the methods of modeling the excitation spread along the myocardium, including the
post-ablation ones, started to gain wide acceptance.
While performing a surgical procedure, the doctor needs to know what result exactly he or she must
achieve. To solve the posed problems, he or she relies on the data obtained during the preliminary
electrophysiological testing, on his or her own skills and experience. The doctor makes a decision
concerning the method and tactics of the further treatment being conducted by him or her. However,
it is necessary to note that the decision is influenced most of all by the kind of arrhythmia that requires
correction. There is no doubt that a graphic representation of the process of impulse spread along the
myocardium enables a better understanding of the arrhythmia mechanism and possible changes after
performing an intervention.
Presently, there are some projects to create the excitation dynamics models both for the entire heart
and for its parts separately; but they are oriented either to the scientific studies of the object (heart) or to
the modeling some properties of the active medium and particular effects [13]. These technologies are
not used clinically owing to their complexity and a high cost of the individual model creation, whereas the
use of the ready-made stereotyped solutions is not possible because of the uniqueness of each individual
case.
In connection with this, the goal was set to create a mathematical model of the auricle myocardium,
which would be capable for functioning in the clinic conditions and forecasting the process of the excitation wave spread after a surgical procedure. (The problem was formulated in the Tomsk Cardiological
Research Institute of the Siberian Division of the Russian Academy of Medical Sciences.)
*

E-mail: riftas@rambler.ru

ANDREEV, KOCHEGUROV

Fig. 1.

1. PROPERTIES OF THE MEDIUM

The authors of the project [4] were among the first to consider the problem of modeling the excitation
spread in the atrial tissues. They formulated the basic principles of origination of both the spiral waves
and the re-entry waves around circular obstacles, and also the law of an excitation spread.
The law of an impulse spread in a homogeneous two-dimensional system is the Huygens principle
in its simplest form: the successive wavefronts are orthogonal to an imaginary system of rays, i.e., of
strained threads which originate at the excited point and bend round all the obstacles. The back front of
the refractory wave is another curve of the same shape, that follows the forward wave front at the distance
measured along those rays. In this case, the wavefront can propagate only into a region which is in the
quiescent state. Since the rays coincide with the strained threads and the spread velocity is constant;
therefore, all points of the wave front are at the same distance (measured along the corresponding
strained thread) from the source of impulses. On an infinite surface without holes and obstacles, the
forward wave front propagating from an excitation point is a circle with the center at that point, and it
will go to infinity. In a convex finite domain, a single excitation point also gives an expanding circular
wavefront that disappears without reflection on reaching the boundary. The strained thread can deviate
and, therefore, change the circular shape of the wavefront for the two reasons: because of an obstacle,
i.e., a hole in the surface, and because of some concave parts of the boundary.
Consider now a convex obstacle inside a convex domain. Figure 1 represents the wavefronts after an
excitation of the point A. The external convex boundary of the domain is not shown, because it simply
cuts the wavefronts everywhere they reach it.
The lines AT1 and AT2 are, in fact, the tangent lines drawn from A to the obstacle. In the domain
bounded by the dotted lines and not containing the obstacle, a wavefront looks like a circle or a circular
arc. Outside this domain, the shape of the wavefront can be obtained as follows: From an arbitrary point
P on the wavefront, draw a tangent line to the obstacle. Then the sum of the distances AT2 + T2 Q + QP
equals the radius AT3 of the circular arc which forms a part of the same wavefront. In other words, the
geometrical locus of the points P is the involute of the obstacle that consists of the two circular arcs:
one with the center at T1 and the other with the center at T2 . At the only point R on the obstacle, the
distances from it to A measured by the thread stretched through T1 and T2 coincide (they are equal to the
sum of the length of the radius AT1 and the distance from T1 to R along the boundary of the obstacle). If
the distance AT3 is less than the corresponding distance to R then the wavefront intersects the obstacle
in two points. If AT3 is greater than the distance to R then two involute arcs of the wavefront intersect
each other under a certain angle, which gradually increases with the increase of the distances. The parts
of the involutes lying behind the wavefront do not belong to it, because otherwise the wavefront moves
along the refractory region.
JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS

Vol. 1 No. 4 2007

ALGORITHMS FOR INTRAOPERATIVE MODELING

2. REVIEW OF THE EXISTING METHODS

The works on modeling the myocardium excitation have been carried out for a long time already, and
nowadays there exist several approaches to solution of this problem:
bidomain model;
reaction-diffusion equations for the excitable media;
cellular automata.
The bidomain model has gained wide acceptance as a basic approach for studying the macroscopic
electrical phenomena in the heart tissue. The heart muscle is represented as a couple of interconnected
spaces, intracellular and extracellular, each of which has different conductivity coefficients along and
across the fibers [5]. The model is based on two main equations:
(in in ) = Imem Isin ,

(en en ) = Imem Isen ,

where in and en are the intracellular and extracellular potentials, in and en are the intracellular and
extracellular conductivities, Isin and Isen are the densities of the intracellular and extracellular currents,
and is the capacity coefficient of the cell surface. Note that in is defined by the conductivity of the
intracellular space and the gap junctions. The domains are connected with each other through the
membrane current Imem . The ionic current Iion (V, t) is calculated according to the cellular membrane
model (see [2, 6, 7]).
The second method for modeling the behavior of the excitable medium utilizes the reaction-diffusion
equations:
E
g
= D1 E + F (E, g),
= D2 E + F (E, g).
t
t
The first equation describes the variations of the autowave E in time, it is the fast variable; and the second
characterizes the slow variable g, i.e., the change in time of the active medium. The reaction-diffusion
equations are used to describe many autowave processes regardless of their physical nature. The last
terms constitute the physical component of the equation. For the excitation waves in the myocardium,
E is the membrane potential and g is the conductivity of the slow component of the ionic current; D1
2
2
2
2
2
and D2 are the diffusion coefficients; is the Laplace operator; =
+
or =
+
+
x y
x y z
for the three-dimensional case [8].
Another approach to describe the behavior of dynamical systems (also well-known and common) is
the cellular automata method. Based on this method, the chemical and diffusive processes, as well as
the movement of the liquid flow and other complicated nonlinear systems are modeled. Now we consider
this method in more detail.
3. THE MATHEMATICAL MODEL
A cellular automaton is a discrete dynamical system, formed by a collection of cells connected with
each other in the same way. All the cells form the so-called cellular automaton lattice. The lattices can
be of various types according to their dimension and the shape of the cells. Each cell (or node) is a finite
automaton, whose state is determined by the states of the neighboring cells and also by its own state.
Each node is characterized by a certain discrete set of the integer-valued quantities (variables) which can
assume a finite number of possible values. The state of variables at each node synchronously changes
after discrete time intervals according to the local rules, which may depend on the state of variables at the
nearest neighboring nodes. In computer science, the cellular automata are an analogue of the physical
notion of a field.
Cellular automata differ from the differential equations by locality of the rules through which the
dynamics of the system is described. In the case of the differential equations, some rules of change of
the quantities averaged over the entire system are used. At that, it is assumed that such rules exist. In
the cellular automata case, the existence of such generalized rules is not necessary. It suffices to know
the laws of the system evolution on a microlevel, i.e., in small spatial regions (cells) that comprise the
JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS

Vol. 1 No. 4

2007

ANDREEV, KOCHEGUROV

Fig. 2.

macrosystem. It is important that those rules are the same for all cells [9]. With respect to the problem
of modeling the excitation dynamics of the active tissue, this property is especially important, because it
allows to take into account the biological nature of the contractile myocardium functioning.
For the cellular automata with two-dimensional lattices of regular polygons, there exist only three
kinds of lattices: triangular, square, and hexagonal (see Fig. 2).
For each lattice type, there exists its own neighborhood of a cell. As a rule, the closest neighbors
are used as a neighborhood (the Moor neighborhood). Among all the neighboring cells, the cells
with common sides are distinguished into a separate class, they are called the main neighbors. A
neighborhood formed only by the main neighbors is called the von Neumann neighborhood. The table
represents the neighborhoods for all kinds of planar lattices.
The neighborhoods for all kinds of planar lattices
Number

Lattice type

The number of main neighbors

Total number of neighbors

Rectangular

Triangular

12

Hexagonal

Consider an automaton capable of receiving a finite number of signals S (s1 , s2 , . . . , sN ) at

each moment of time t = 1, 2 . . . and capable of changing its internal state depending on them. The
automaton can perform a finite number of actions f (f1 , f2 , . . . , fx ). The choice of action is determined
by the internal state of the automaton, and it has a finite number of them (1 , 2 , . . . , m ). The
number m is called the memory capacity of the automaton.
Assume that the automaton is situated in a certain medium and that the actions f give rise to the
responses S of the medium C. These responses, in their turn, are the input signals for the automaton; it
uses them for making decisions about the further actions.
In the problem for modeling the atrial excitation dynamics, the automata approach has a number of
serious advantages:
fine-grain parallelism that the cellular automata have;
a possibility to use the modern multiprocessor stations for computations on a high-dimensional
grid;
relative simplicity of realization;
a possibility to adapt the model to a real object based on the data becoming available during the
surgical procedure.
JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS

Vol. 1 No. 4 2007

ALGORITHMS FOR INTRAOPERATIVE MODELING

Introduce some notation. Consider a two-dimensional realization of a cellular automaton as the

simplest case. For the calculations, we will use discrete time values separated by the same interval t.
In this model, t must assume the value equal to the time interval in which the excitation is transmitted
from one cell to another. The discrete values of time, for which the system states must be computed, are
represented by t = {t1 , t2 , . . . , tn }, where tn is the value on the boundary of the interval of interest.
Let A(i, j, t) be the state of the cell with the spatial coordinates i and j at the time moment t. The
automaton under consideration can be in one of four states. At that, the quantity 1 corresponds to the
state of quiescence; 2 , to the excitation state; 3 , to the refractory state; and 4 , to the state in which the
cell does not have the active behavior properties. Thus, the automaton possesses the memory capacity
m = 4. The considered automaton has a bounded number of actions f (f1 , f2 , . . . , f4 ), where f1 is a
transition from the state 1 to the state 2 ; f2 , from 2 to 3 ; and f3 , from 3 to 1 .
The action of the automaton in the state 4 should be considered separately. In this case, only the
transition to 4 is possible.
The change of the states may occur in a certain strictly determined sequence: if a cell is in the
quiescent phase, then in the next moment of time it may become active, then refractory, and only after
that it may return to the initial state.
The states of all cells are preset as the initial conditions. By default, each cell is in the quiescent
phase. In the model under consideration, the self-excitation capability of the cells is taken into account
by introduction of the pacemaker-cells that initiate the movement of the excitation wave. Initially those
cells can be in the active phase or take the excitation state at the given moment of time, which also is set
by the initial conditions A(i, j, t) = 2 . This allows moving the system of cells out of equilibrium. It is
assumed that the cells, to which the value 2 is assigned in the initial conditions, are the rhythm drivers
and therefore can experience excitation spontaneously. In order to define the non-conducting sections,
the state A(i, j, t) = 4 is assigned to necessary cells.
To solve the considered problem, a cellular automaton must meet several requirements:
all the cells must have the same structure of the connections;
each wave generation must be continuous.
In other words, there should be no cells behind the wavefront that have not taken part in the excitation
process. The only exceptions may be non-excitable cells A(i, j) = 4 and the cells that at the moment
of the wavefront passing were in the refractory state A(i, j, t) = 3 .
4. CELLULAR AUTOMATON LATTICE
Regardless of the lattice type, there exists a neighborhood for each cell that constitutes a ring around
the central cell. The notion of a ring is one of the key notions in construction of the auricle excitation
model. For a homogeneous medium, a ring can be identified with the wavefront. It is exactly the ring
shape that will determine the precision of the cellular automaton approximation of the excitation front
shape. On the zero step of computation the pacemaker-cells serve as rings; on the first, the cells
bordering them, and so on.
To construct the model, a rectangular grid was used. Consider the possible variants of the neighborhood for a rectangular grid. The simplest ones are four- and eight-connected automata (see Fig. 3).
Introduce notation:
A(i, j + 1, t) = 1 = A1 (t), A(i, j 1, t) = 1 = A2 (t), A(i + 1, j, t) = 1 = A3 (t),
A(i 1, j, t) = 1 = A4 (t), A(i + 1, j + 1, t) = 1 = A5 (t), A(i 1, j + 1, t) = 1 = A6 (t),
A(i 1, j 1, t) = 1 = A7 (t), A(i + 1, j 1, t) = 1 = A8 (t),
f (A(i, j, t)) is the action of the cell A(i, j, t) at the time moment t.
Consider now the possible variants of the cells connection (Fig. 4).
A four-connected automaton is the automaton with four connected cells which change the states
under condition that on the previous step the central cell was excited. For the automaton with four
neighboring cells, two variants may be considered (Figs. 4, a and 4, b):
JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS

Vol. 1 No. 4

2007

ANDREEV, KOCHEGUROV

Fig. 3.

Fig. 4.

1) the von Neumann neighborhood

A1 (t 1) A2 (t 1) A3 (t 1) A4 (t 1) f1 (A(i, j, t)),

(1)

2) neighborhood with the diagonal cells

A5 (t 1) A6 (t 1) A7 (t 1) A8 (t 1) f1 (A(i, j, t)).
Both considered variants satisfy the first condition; however, the neighborhood with diagonal cells does
not meet the second condition.
An eight-connected automaton is the automaton with the Moor neighborhood:
A1 (t 1) A2 (t 1) A3 (t 1) A4 (t 1) A5 (t 1)


A6 (t 1) A7 (t 1) A8 (t 1) f1 (A(i, j, t)).

(2)

The form of connections (1) and (2) will allow construction of the wave satisfying both conditions.
Despite the automata (1) and (2) working within the set conditions, they approximate the wavefront
poorly; thus, consider a combination of automata (1) and (2). To this end, introduce the notions of even
and odd generation of the wavefront. For the even generation, the excitation is transmitted according
to the supposition that the cells are eight-connected; and for odd generation, in accordance to the
four-connectedness. Such a cellular automaton approximates the wavefront better than (1) and (2) (see
Fig. 3).
It should be stated explicitly that the separation of the computation process into even and odd steps
does not contradict the first rule; because all the cells are connected with each other in the same way,
and one and the same cell may, under different conditions, work according to both variants (1) and (2).
In order to take into account the refractoriness of the active medium and the possibility of absence of
the excitation, write down the cell evolution rules in the following form:
A(i, j) = 4 = Aps (i, j) is the cell with the coordinates i and j without any excitation capacity;
A(i, j, t) = 3 = Arf (i, j, t) is the cell with the coordinates i and j in the refractory state.
JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS

Vol. 1 No. 4 2007

ALGORITHMS FOR INTRAOPERATIVE MODELING

A complete rule of the cell transition into the active state for even and odd generation of the wavefront
can be written in the following form:
for even generation of the wavefront:
(A1 (t 1) A2 (t 1) A3 (t 1) A4 (t 1) A5 (t 1)
A6 (t 1) A7 (t 1) A8 (t 1)) (Aps (i, j) Arf (i, j, t)) f1 (A(i, j, t)),
for odd generation of the wavefront:
(A1 (t 1) A2 (t 1) A3 (t 1) A4 (t 1)) (Aps (i, j) Arf (i, j, t)) f1 (A(i, j, t)).
The condition of the cell transition to the state of refractoriness can be represented in the form
f1 (A(i, j, t)) f2 (A(i, j, t + 1));

(3)

whereas the refractory period of the cell can be described by the rule
f2 (A(i, j, t)) f3 (A(i, j, t + n)),

(4)

where n is the duration of the refractory period.

Equations (3) and (4) are applicable both for even and odd calculation steps.
5. MODEL ADAPTATION TO A REAL RESEARCH OBJECT
For a model to be adequate, it is necessary to introduce a mechanism of its adaptation to the data
coming from the research object (data received by the electrophysiological tests).
On account of a limited possibility of obtaining some input information, the values of the transition
time into the excitation state are set only for some of the cells located arbitrarily. The problem of the
model adaptation to real conditions is reduced to the calculation of the excitation time for each cell and
subsequent calculation of the delay interval in the excitation transmission between the neighboring cells:
  n 


n
X
R hi p X R hj p
t(x, y) =
wi fi ,
wi =
,
(5)
Rhi
Rhj
i=1

j=1

where n is the number of known points, wi is a weight function, fi is a given value of the function at
the point i, R is the distance from the interpolated point to the farthest point with a known value of the
excitation time, hi is thep
distance between the interpolated and the given point, p is a power parameter
(usually equals 2); hi = (x xi )2 + (y yi )2 , x and y are the cells coordinates.
The delay interval in the excitation transmission between the cells A(x, y) = A and B(x + 1, y) = B
is calculated as tAB = t(x, y) t(x + 1, y). In the direction of other neighboring cells t is calculated
analogously. The duration of the refractory period rf is calculated by (5) for each cell.
After all parameters for the cellular automaton are obtained, it should be taken out of equilibrium. To
do that, it suffices to move one or several cells into the excitation state. However, it should be done taking
into account the data obtained in the process of the surgical procedures. One can take as a pacemaker
the cell with the earliest excitation time t(x, y)min or a group of cells, when several cells have this value.
CONCLUSION
In the process of modeling, a software product was created which enables to model the dynamics of
the contractile myocardium excitation. This product is planned to be included in a medical-diagnostic
complex Elkart 2 Navigator developed by the medical production company Electropulse.
The solution of the posed problem enables to realize the following in medicine:
to carry out control of the surgical procedure and evaluate its effectiveness before the procedure;
to carry out the search for new, more efficient schemes of the catheter ablation;
to train the medical personnel using the data obtained during the surgical procedures.
JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS

Vol. 1 No. 4

2007

ANDREEV, KOCHEGUROV

REFERENCES
1. A. V. Holden and V. N. Biktashev, Computational Biology of Propagation in Excitable Media Models of
Cardiac Tissue, Chaos. Solitons & Fractals, No. 8, 16431658 (2000).

2. K. Simelius, J. Nenonen, and M. Horacek,

Modeling Cardiac Ventricular Activation, Internat. J. Bioelectromagnetism, No. 2, 5158 (2001).
3. D. T. Kaplan, J. M. Smith, B. E. H. Saxberg, and R. J. Cohen, Nonlinear Dynamics in Cardiac Conduction,
Math. Biosci., No. 90, 1948 (1988).
4. N. Wienier and A. Rosenblueth, The Mathematical Formulation of the Problem of Conduction of Impulses in
a Network of Connected Excitable Elements, Specifically in Cardiac Muscle, Arch. Inst. Cardiology (Mexico),
No. 16, 205265 (1946).
5. C. H. Luo and Y. Rudy, A Model of the Ventricular Cardiac Action Potential: Depolarization, Repolarization,
and Their Interaction, Circ. Res., No. 6, 15011526 (1991).
6. C. H. Luo and Y. Rudy, A Dynamic Model of the Cardiac Ventricular Action Potential. I: Simulations of Ionic
Currents and Concentration Changes, Circ. Res., No. 6, 10711096 (1994).
7. C. H. Luo and Y. Rudy, A Dynamic Model of the Cardiac Ventricular Action Potential. II: Afterdepolarizations,
Triggered Activity, and Potentiation, Circ. Res., No. 6, 10971113 (1994).
8. G. R. Ivanitskii, Biophysics at the Turn of the Century: Autowaves, Biophysics, No. 5, 773795 (1999).
9. V. K. Vanag, Study of Spatially Distributed Dynamical Systems using Probabilistic Cellular Automata,
Uspekhi Fiz. Nauk, No. 5, 481503 (1999).

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS

Vol. 1 No. 4 2007