An inverse method of estimating parameter distributions based on a heat muscle model

The authors present a method for estimating parameters on a nonlinear system that would otherwise be difficult to measure directly. The method is based on an extended backpropagation technique where the evolution of the measured field variables over time is mapped to an artificial neural network. The connections within the network are defined by the mathematical model that represents the system. The model is then used to run forward simulations and inverse refinements iteratively until errors are within acceptable bounds. As an example, the performance of this method on a simulated 2-D myocardial tissue is investigated. A modified FitzHugh-Nagumo model was used where both the electrical potential and the generalized current were described over time. The task assigned to the method was to determine the cell-to-cell coupling or diffusion coefficients of the simulated tissue. >