IJBEM logo
International Journal of Bioelectromagnetism
Vol. 4, No. 2, pp. 53-54, 2002.

previous paper

next paper

www.ijbem.org

A DOMAIN EMBEDDING STRATEGY FOR SOLVING THE BIDOMAIN EQUATIONS ON COMPLICATED GEOMETRIES

G. T. Lines, J. Sundnes, A. Tveito
Simula Research Laboratory,
P.O. Box 134 Lysaker, N-1325, Norway

Abstract: Simulations with the bidomain equations are often solved on simple geometries due to the complexity introduced by using more realistic boundaries. In this work we propose a method where all the calculations are carried out on squares and boxes even when the tissue boundaries are complex. This is achieved by domain embedding and proper extension of the conductivity tensors in the model.We show in this work how realistic geometries can be implemented for a simulator working on lattice grids only.

INTRODUCTION

In many applications of the bidomain model it is important to use realistic geometries, re-entry simulations being one example. The geometry is then often represented by an unstructured grid, which makes the implementation of the simulator hard compared to the case where the equations are defined over a structured grid. Although there exists tools for dealing with these complications, the fact remains that a lot of work is still carried out on idealized geometries.
 
METHOD
The coupled bidomain[1] and forward problem can be formulated as:
H:
s

t
= F(t,s,v;x)
H:
 C c v

t
+ cIion(v,s) = Ñ·(Mi Ñ(v+ue))
H:
Ñ·((Mi+Me) Ñue) = - Ñ·(Mi Ñv)
H:
nH ·(Mi Ñ(v+ue)) = 0
H:
ue = uT
H:
nH·Me Ñue = - nT·MT ÑuT
T:
Ñ·(MT ÑuT) = 0
T:
nT ·(MT ÑuT) = 0
(1)
where H is the myocardial region and T is the rest of the computational domain. The physical parameters C and c are membrane capacitance and membrane surface area per volume tissue. The matrices Mi,Me and MT are conductivity tensors for intra- and extra-cellular space in H and for extra cardiac tissue in T. The primary unknowns are the transmembrane potential v, the extra cellular potential ue, and the extra cardiac potential uT. To compute the ion-current Iion a set of states s must be known, these are determined from an ODE system[2] with right hand side F.
Our aim is to construct a method were a simple grid can be used to solve (1) even if H and T are complex. We first note that (1) can be reduced to a simpler system[3]. In particular the boundary conditions for uT and ue make it possible to combine the elliptic equations on H and T. To this end we define W = TÈH and
u(x)= ì
í
î
ue(x),
if x Î H
uT(x),
if x Î T
The equation for u becomes
Ñ·(M Ñu) = f,
x Î W
where
f(x) = ì
í
î
-Ñ·(MiÑv)
if x Î H
0
if x Î T
(2)
and
M(x) = ì
ï
í
ï
î
Mi(x)+Me(x),
if x Î H
MT(x),
if x Î T
(3)
The next simplification is to expand the domains H and T into two circumscribing boxes H¢ and T¢, see Figure 1. The parabolic part of the problem is only defined over the myocardial domain H. The boundary condition states that the intra cellular current does not leave the domain. This is implemented by setting Mi=0 on H¢-H. A similar approach is used to implement the no-flow condition on W, i.e. MT=0 on T¢-T. There will be overlap between T and H¢, MT will in this area be defined according to its extra cardiac location.
As a third and final simplification we propose an operator splitting method to solve the reduced system. The ODEs and the two PDEs are thus not solved simultaneously, but in three separate steps.

Figure 1: The original domains H and T embedded in boxes H¢ and T¢ .

 
IMPLEMENTATION

Constructing appropriate grids for the the problem is now fairly simple since the domains are only rectangles or boxes. If the finite difference scheme will be used a lattice grid is suitable. In the test cases presented here we have used the finite element method with triangular elements i 2D.
Since the solution from one domain is needed to compute the solution in the other domain a mapping is needed between the domains. This is particularly easy to construct if the inner grid is a subset of the outer grid, i.e. with common nodes and elements. Such an arrangement is shown in Figure 2.

Figure 2: Cross section of the torso. The drawn boundaries are from outer to inner: W ¢, W, H ¢ and H (both epi- and endocardium). The coarsest grid is shown with dotted lines. Note that the grid over H¢ is a subset of the grid over W¢.

To construct the conductivity tensors the geometries of the true domains must be represented in some form. In a 2D setting it might be given as polygons. In general a function is needed that can decide if a given point is on the inside or on the outside of a domain. We will not go into detail about how such functions can be constructed. It suffice to say that it is in general a lot simpler to implement an indicator function for a surface than it is to construct a grid based on the same surface.
As mentioned above, the boundary conditions on the physical boundaries will be taken care of by the conductivity tensors. The type of boundary condition on the artificial boundaries can be either natural or essential, the choice will not effect the solution on the interior domains.
 
DISCUSSION
To illustrate the technique we have used a 2D slice of the torso with realistic endo- and epicardial boundaries as shown in Figure 2. The domain T in this example is not connected, it consists of the extra cardiac space plus the two blood cavities of the ventricles. This will not complicate the computation since only H and W are relevant for the simulator, both of which are connected and also rectangular after the embedding.
Figure 3 shows the solution late in the depolarization phase. The non-embedded reference solution is show at the top. In the finest level of refinement(bottom), we see that the depolarization sequence is well preserved, which indicates that the error introduced by the embedding is small, provided that a fine resolution is used (approximately 400.000 nodes).

Figure 3: Top: The transmembrane potential in the original simulation. Below: Embedded approach. Two levels of refinement, finest at the bottom. For comparison, completely unexcited tissue is shown in white.

We note that the technique is also applicable for isolated heart simulations, both in the mono- and bidomain case. Although the approach simplifies the implementation there will be computational overhead due to increased tissue volume.

REFERENCES

[1] C.S. Henriquez, "Simulating the Electrical Behavior of Cardiac Tissue Using the Bidomain Model", Critical Reviews in Biomedical Engineering, vol 21, pp. 1-77, 1993.
[2] R.L. Winslow, D.F. Scollan, A. Holmes, C.K. Yung, J. Zhang , M.S. Jafri, "Electrophysiological modeling of cardiac ventricular function: From cell to organ", Ann Rev Biomded Engng, vol 2, pp. 119-155, 2000.
[3] J. Sundnes, G.T. Lines, A. Tveito, "Efficient solution of ordinary differential equations modeling electrical activity in cardiac cells", Math. Biosc., vol 172, pp. 55-72, 2001.

 

previous paper table of contents next paper

© International Society for Bioelectromagnetism