A ResNet-BiLSTM Multi-lead ECG Classification Method Embedded with Attention Mechanism

Abstract

Computer-assisted electrocardiogram analysis has important clinical significance for the prevention and treatment of cardiovascular diseases. A multi-lead electrocardiogram (ECG) classification method based on the residual network and bidirectional long short-term memory neural network was proposed. In order to extract more effective ECG features, the Squeeze-and-Excitation (SE) attention mechanism was embedded into the depth model. Finally, the effectiveness of the proposed method was verified on the Chinese cardiovascular disease database (CCDD) and the internationally recognized MIT-BIH-AR database. The accuracy, sensitivity and specificity of normal and abnormal heartbeats classification on the MIT-BIH-AR database that contains 48 recordings were 99.52%, 99.46% and 99.54%, respectively. The accuracy, sensitivity and specificity of the classification of normal and abnormal ECG recordings on the CCDD that contains more than 150000 recordings were 84.44%, 79.27% and 88.45%, respectively. The overall experimental results show that the classification performance of the proposed method is good on both small-scale and large-scale data sets.