Interpretable deep learning for automatic diagnosis of 12-lead electrocardiogram

• DOI: 10.1016/j.isci.2021.102373
• Publication Date: 2021-03-29T22:48:57Z
• Authors: Dongdong Zhang; Samuel Yang; Xiaohui Yuan; Ping Zhang
• Keywords: Lead (geology); Deep Neural Networks

Electrocardiogram (ECG) is a widely used reliable, non-invasive approach for cardiovascular disease diagnosis. With the rapid growth of ECG examinations and insufficiency cardiologists, accurate automatic diagnosis signals has become hot research topic. In this paper, we developed deep neural network classification cardiac arrhythmias from 12-lead recordings. Experiments on public dataset showed effectiveness our method. The proposed model achieved an average F1 score 0.813. superior performance than 4 machine learning methods learned extracted expert features. Besides, models trained single-lead ECGs produce lower using all 12 leads simultaneously. best-performing are lead I, aVR, V5 among leads. Finally, employed SHapley Additive exPlanations method to interpret model's behavior at both patient level population level.