Convolutional Block Attention BiLSTM for Arrhythmia Detection

Abstract

Cardiovascular diseases represent a significant cause of mortality, with millions of electrocardiograms being recorded each year. Therefore, methods of automated diagnosis for electrocardiograms are of particular interest. Since electrocardiograms have recognizable features and are time-dependent, we propose a model using convolutional layers, convolutional attention, and long short-term memory units. The model is trained and validated on the MIT-BIH Arrhythmia database, and achieves an accuracy of 99.10%, a precision of 99.09%, a specificity of 99.64%, a sensitivity of 95.90%, and an F1 score of 97.47%.