Deep Learning-Based Algorithm for Prediction of Heart Disease using Electrocardiogram

Abstract

Disease diagnosis in the hospitals is usually carried out by experts and experienced medical practitioners. For an expert cardiologist, any anomaly over the heart rate can easily be detected using Electrocardiogram (ECG) report. ECG is reliably used as a measure to monitor the functionality of the cardiovascular system. The primary challenge in manually analyzing ECG signals lies in the intricate task of detecting and categorizing diverse waveforms and morphologies present within the signal. The change in the morphological pattern over a recorded ECG could be sometimes confusing for cardiologist and highly challenging. In this work, a single learning model is created by utilizing an adaptive implementation of 1-D Convolutional Neural Networks (CNNs), which combines the two primary components of classification, namely feature extraction and classification. Our system uses deep-learning techniques to predict the heart disease accurately by developing a ECG signal classification methodology. Deep CNN is used to accurately classify five different arrhythmias in accordance with AAMI EC57 standard. The CNN is trained and tested using ECG dataset obtained from MIT-BIH resulting in an accuracy of 94.31%.