Application of Bidirectional LSTM Neural Networks and Noise Pre-cleaning for Feature Extraction on an Electrocardiogram

Abstract

Cardiac diseases are one of the most common diseases on the planet. Thousands of people die from this disease every year. For prompt diagnosis, an automated system for processing electrocardiograms is required. The standard model of an automated system consists of signal preprocessing, feature extraction, and classification. In this article, unidirectional and bidirectional network models with long short-term memory were considered for the classification of electrocardiogram signals. The simulation results showed that the use of both methods without preliminary signal processing and feature extraction on them is not advisable. Also, the simulation result showed that models that include the removal of noise from electrocardiograms have more accurate training results for bidirectional networks with a long short-term memory. The simulation was carried out in the MatLab 2020b mathematical environment based on the PhysioNet Computing in Cardiology Challenge 2017 database, taken from an open source. The best result was obtained in the classification of atrial fibrillation.