Training a NN with ECG to diagnose the hypertrophic portions of HCM

Abstract

In this study the authors try to construct a neural network trained with electrocardiogram (ECG) information to diagnose the hypertrophic portions of hypertrophic cardiomyopathy (HCM). Computer electrocardiography remains a fundamental diagnostic method for both contour and rhythm analysis. In almost all patients with HCM, there are more or less abnormal ECG findings, but it is very difficult to diagnose the hypertrophic portions of HCM relying solely on ECG findings, even for an experienced cardiologist. The data used in this study are from seventy-nine patients with HCM. Their ECGs were used to test and train a neural network, and the criteria of teaching data depended on the results of echocardiography. This study was completed using the Neural Works Professional II/PLUS of Neural Ware Inc., on a personal computer. A three-layer neural network trained by back-propagation algorithm showed better ability for diagnosing the hypertrophic portions of HCM depending only on ECG information.