Automated Detection of Cardiac Arrhythmia using Recurrent Neural Network

Abstract

Recurrent Neural Networks have recently emerged as a boon for analysing time-series data. The primary aim of this paper is to perform automated classification of seven types of ECG beats using Gated Recurrent Unit (GRU) and Long Short-term Memory (LSTM). To gather ECG Signals, MIT-BIH arrhythmia database is used. Noise is removed using Generative adversarial network (GAN) and ECG beat segmentation is done to get labelled database. Using these extracted ECG beats, our designed models are trained from scratch and then tested. Investigating the results obtained by training process, it is observed that the designed network with LSTM layer obtained best results when compared to the network with GRU layer. The network with GRU layer achieved an accuracy of 96.72% and the network with LSTM layer achieved an accuracy of 98.22%.