Neural Networks on Noninvasive Electrocardiographic Imaging Reconstructions: Preliminary Results

Abstract

In the reverse electrocardiography (ECG) problem, the objective is to reconstruct the heart’s electrical activity from a set of body surface potentials by solving the direct model and the geometry of the torso. Over the years, researchers have used various approaches to solve this problem, from direct, iterative, probabilistic, and those based on deep learning. The interest of the latter, among the wide range of techniques, is because the complexity of the problem can be significantly reduced while increasing the precision of the estimation. In this article, we evaluate the performance of a deep learning-based neural network compared to the Tikhonov method of zero order (ZOT), first (FOT), and second (SOT). Preliminary results show an improvement in performance over real data when Pearson’s correlation coefficient (CC) and (RMSE) are calculated. The CC’s mean value and standard deviation for the proposed method were 0.960 (0.065), well above ZOT, which was 0.864 (0.047).