Abstract

Info

Title:: ECG-based identification and classification of myocardial infarction

Author(s):: WANG Xinfeng1; 2; QI Mengling3; XU Hongzhi1; 1. Software College, Jishou University, Jishou 416000, China2. School of Data and Computer Science, Sun Yat-sen University, Guangzhou 510275, China 3. Medical Research Center, Sun Yat-sen Memorial Hospital, Sun Yat-sen University, Guangzhou 510120, China

Keywords:: Keywords: myocardial infarction electrocardiogram machine learning deep learning interpretability

Abstract:: Abstract: Myocardial infarction (MI) is a kind of severe heart diseases. Pre-symptomatic health examination can detect MI diseases at early stage. Electrocardiogram (ECG) is a common-used non-invasive health examination and diagnostic tool. Some studies that use ECG to predict MI have some shortcomings such as private data sets-based, small sample sizes, simple analysis methods. To solve these issues, the first benchmark prediction experiment of MI on the available collection of biological information resource platform UK Biobank is proposed, which covers machine learning methods based on clinical features and deep learning methods based on ECG signals. The results show that the AUC based on clinical characteristics is 0.690, and the AUC of deep learning using the original ECG signals is 0.728, which is an increase of nearly 4%. It is proved that deep learning based on the original ECG can learn more information than clinical features. In addition, a preliminary interpretability analysis is performed on the results obtained by XGBoost and ResNet methods, and it is found that ST wave is more closely related to MI.