ECG Signal Classification Using Long Short-Term Memory Neural Networks

This study proposes a neural-based approach to classify ECG signals acquired from a low-cost wearable device as an early warning system for possible heart diseases. The aim is to recognize ECG signals into respective classes, including normal sinus rhythm and different types of arrhythmia, while adhering to FDA approved IEC standards for wearable medical devices. The proposed strategy aims to design a robust classifier for the ECG watch, which is part of the “TeleHcart” currently under development by the authors. The proposed hybrid model of Wavelet Transform Strategy (WST) fused with a Bidirectional Long Short-Term Memory (Bi-LSTM) neural network achieves an overall classification accuracy of 91% for all classes of arrhythmia. The study focuses on ECG datasets and classifier development aspects only, and the classification system is designed to assist cardiologists and general practitioners in making accurate diagnoses by acting as an assistive/recommendation tool.