Enhancing cardiovascular disease classification through attention-based deep learning methods

The field of electrocardiography (ECG) monitoring and diagnosis systems has experienced notable advancements in software and hardware innovations, particularly in the realm of automatic ECG classification algorithms. In this study, we propose a novel Squeeze-and-Excitation ResNet and Transformer encoder model (RSETd) that integrates gender and age demographic features to classify seven categories, including normal and six abnormalities. We also explore an extended version of RSETd to classify 39 categories. Our model demonstrates satisfactory accuracy and time-cost performance: The proposed model achieves an impressive F1-score of 0.779, with a time-cost of 10.862ms per record. To enhance the classification capabilities further, we expand the detection classes to 39 categories of abnormalities, resulting in an F1-score of 0.732, with a time-cost of 104.561ms per record.