The ANN as a technique to solve the inverse problem ofelectrocardiography: the effect of the training margin on the errorscaused by geometric uncertainties in an eccentric homogenous sphericalmodel

摘要

Artificial neural networks (ANN) were previously proposed by theauthor as a technique for solving the inverse problem ofelectrocardiography. The aim of the paper is to find the training limitsof the input data for this problem and the related probability of outputerrors. Using a model of a homogeneous spherical body, the cardiacsource was represented by six current dipoles located on the surface ofan eccentric heart sphere. Body surface potentials were calculated at 26measuring points for (19×64) cases representing the all possiblecombinations of dipole status (64) in the following (19) cases; thebasic case, 8 cases of uncertainty in heart radius r_o, 8cases of angular uncertainty in θ, and 2 cases in Φ of thespherical coordinates. The data base for 64×19 cases was used totest the response of the ANN for each group of training. The statisticaloutput error E was calculated in each case (E=Nf×100/64×6,where Nf is the number of false outputs). The obtained results showedthat the ANN accepted the training data within a limited margin, beyondwhich the data would appear to be contradictory