BiLSTM-Attention Deep neural networks for Electrocardiogram arrhythmia classification

Abstract

Arrhythmia refers to an abnormal rhythm of heartbeat; heart may beat too fast, too slow, or with an irregular rhythm. In medicine, the electrocardiogram (ECG) is prevalently used for detecting and classifying arrhythmias. In this paper, we present an approach based on the combination of Convolutional Neural Networks (CNNs), Bidirectional Long Short-Term Memory (BiLSTM) networks and Attention Mechanism for the classification of heartbeats, which can accurately classify five different types of arrhythmias. We evaluated the proposed method on the PhysioNet MIT-BIH dataset. According to the results, our method was evaluated using 10-fold cross-validation, achieving the accuracy of 95.17% in arrhythmia classification.