Prediction of ECG Signals Using Feedforward Neural Networks

Abstract

The Electrocardiogram (ECG) signal could be modelled or analysed using time series prediction methods. This study considered neural networks models trained with ECG data, so that the trained models could then predict ECG for the purpose of diagnosis and prevention of cardiac troubles. Predicting the ECG is necessary in order to assist the heart specialist to provide early measures that could avert the likely cardiac crises thereby improving life’s longevity, productivity and standard of living. The research utilised the application of backpropagation algorithm feedforward neural networks to predict the ECG of heart rhythm disorders. The ECG data for very slow heartbeat (sinus bradycardia), low blood reaching the heart (myocardial ischemia) and very fast heartbeat (ventricular tachycardia) were obtained from Massachusetts Institute of Technology, Biomedical Institute of Health Sciences (MIT-BIH).  Feedforward neural networks (FFNN) using Levenberg-Marquart training algorithm were investigated in this research using neural network toolbox in MATLAB and were found to be good predictors of ECG. Feedforward neural network prediction performance however proved that FFNN could effectively predict the ECG. The research was based on short-term prediction of ECG using single-point prediction.