FitzHugh-Nagumo model

This two-dimensional system of PDE models an excitable system such as electric signals on heart tissue. It is solved by means of a finite differences scheme. Also, an inverse problem is established and partially solved: reconstruct cardiac tissue properties by using measurements available from an ECG. Code available on github upon request.

Overview

Heart disease is one of the leading causes of death in the world, thus it is necessary a fully-understanding of its biomechanical behaviour. One attempt to do so is the FitzHugh-Nagumo model, one of the most successful mathematical models that describe excitable media, that is, materials consisting of elementary segments with well-defined rest state, a threshold for excitation and a diffusive-type coupling to its nearest neighbours. Hence, this model is useful for modelling different phenomena such as nerve pulses, spreading of forest fires or certain types of chemical reactions. In particular, it is used in cardiac electrophysiology to qualitatively describe the voltage propagation through the cardiac tissue. Consequently, from this model, it is possible to gain some insights about heart imparities such as arrhythmias.

The most important characteristic of excitable media is the almost immediate damping out of signals below a certain threshold. On the other hand, signals exceeding this threshold propagate without damping. In cardiac cells, ions move through small pores in the cellular membrane which can be either open (excited) or closed (rest) making the heart muscle pumping blood in and out.

In what follows, the equation is described, a direct model is solved with a finite differences scheme and an inverse problem is established to reconstruct properties of the heart tissue from measurements given by an ECG

The equation

The FitzHugh-Nagumo model is a system of two variables and two PDEs. The variables are an activator (the electric potential $V$) and an inhibitor (a variable $R$ that describes the voltage-dependent probability of the pores in the membrane being open and ready to transmit ionic current). The potential follows a diffusion process with a non-linear term.

$$\begin{array}{ll}\dfrac{\partial V}{\partial t}= div(\sigma\nabla V)+\nu V(\alpha-V)(V-1)- R + I & \text{, in }\Omega\times[0,T] \\
\dfrac{\partial R}{\partial t}=\delta (V-Rd) & \text{, in } \Omega\times[0,T] \\
\dfrac{\partial V}{\partial n}= 0 & \text{, on } \partial \Omega \times [0,T] \\
V(x,0) = V_0 & \text{, for } x\in\Omega \\
R(x,0) = R_0 & \text{, for } x\in\Omega \end{array}$$ where

  • $V(t,x)$: the voltage or transmembrane potential in the point $x\in\Omega$ and time $t\in[0,T]$.
  • $R(t,x)$: inhibitor or recovery variable.
  • $\sigma(x)$: conductivity of the tissue that may depends on $x\in\Omega$. Moreover, $\sigma(x)\geq0$ and if $\sigma(x)=0$ then the tissue on the point $x\in\Omega$ is dead and no diffussion may occur.
  • $\nu$: excitation rate, with $\nu>0$.
  • $\alpha$: threshold for excitation, with $0<\alpha<1/2$.
  • $\delta$: excitation decay, with $0<\delta$.
  • $d$: recovery decay, with $d>0$.
  • $I$: stimulating current.

Direct problem. Solving the equation with finite differences

A Crank-Nicolson scheme is used to numerically solve the system, i.e., a convex comibination between the explicit and implicit schemes with parameter $\theta$:

$$\begin{array}{cll} \dfrac{V^{n+1}_{i,j}-V^n_{i,j}}{\Delta t}&=&\theta\sigma_{i,j} \left(\dfrac{V^{n}_{i,j+1}-2V^{n}_{i,j}+V^{n}_{i,j-1}}{h_x^2}+\dfrac{V^{n}_{i,j}-2V^{n}_{i,j}+V^{n}_{i-1,j}}{h_y^2}\right) \\
&&+\theta\left((\sigma_x)_{i,j}\dfrac{V^n_{i,j+1}-V^n_{i,j}}{h_x}+(\sigma_y)_{i,j}\dfrac{V^n_{i+1,j}-V^n_{i,j}}{h_y}\right) \\
&&+(1-\theta)\sigma_{i,j} \left(\dfrac{V^{n+1}_{i,j+1}-2V^{n+1}_{i,j}+V^{n+1}_{i,j-1}}{h_x^2}+\dfrac{V^{n+1}_{i,j}-2V^{n+1}_{i,j}+V^{n+1}_{i-1,j}}{h_y^2}\right) \\
&&+(1-\theta)\left((\sigma_x)_{i,j}\dfrac{V^{n+1}_{i,j+1}-V^{n+1}_{i,j}}{h_x}+(\sigma_y)_{i,j}\dfrac{V^{n+1}_{i+1,j}-V^{n+1}_{i,j}}{h_y}\right) \\
&&+\nu V^{n}_{i,j}(V^{n}_{i,j}-\alpha)(1-V^{n}_{i,j})- \theta \delta R^{n}_{i,j}-(1-\theta)\delta R^{n+1}_{i,j} \\
\dfrac{R^{n+1}_{i,j}-R^n_{i,j}}{\Delta t}&=&\theta\delta (V^{n}_{i,j}-R^{n}_{i,j}d)+(1-\theta)\delta (V^{n+1}_{i,j}-R^{n+1}_{i,j}d) \end{array}$$

Simulations

Normal heart

Reentrant waves

Ventricular fibrilation

Code and Poster

You can download a poster here.

Inverse problem

For this toy model, we could suppose as known the parameters $\num \alpha, \delta$ and $d$ while $\sigma$ remains unknown. This function contains information about the properties of the cardiac tissue, hence, it would be helpful to know it and make a good prognosis. The inverse problem proposed is to reconstruct this function $\sigma$ from measurements obtained from an ECG.

An ECG is a graph of voltage versus time that describes the electrical activity in the heart from electrodes located in the skin.

Pablo Arratia
Pablo Arratia
Young researcher

My research interests include inverse problems, partial differential equations, deep learning, physics-informed neural networks and medical imaging.