Inception based GAN for ECG arrhythmia classification

Abstract

Cardiovascular diseases are the world's principal reason for death, accounting it about 17.9 million people per year, as reported by World Health Organization(WHO). Arrhythmia is often a heart disease that is interpreted by a variation in the linearity of the heartbeat. The goal of this study would be to develop a new deep learning technique to accurately interpret arrhythmia utilizing a one-second segment. This paper introduces a novel method for automatic GAN-based arrhythmia classification. The input ECG signal is derived from the fusion of well known Physionet dataset from MIT-BIH and some Hospital ECG databases. The ECG segment over time is used to detect 15 different classes of arrhythmias. The GAN network uses an attention-based generator to learn local essential features and to maintain data integrity for both time and frequency domains. Among these, the highest accuracy obtained is 98%. It can be inferred from the results that the proposed approach is smart enough to make meaningful predictions and produces excellent performance on the related metrics.