Classification of Heart Diseases Based On ECG Signals Using Median Complex and XGBoost

Rui Yu, Guanghong Bin, Guangyu Bin
Department of Biomedical Engineering, Faculty of Environmental and Life Sciences, Beijing University of Technology, Beijing, China


The electrocardiogram (ECG) contains a wide range of heart-related information, the identification of which can help to detect, localize and treat heart disease. The 2021 PhysioNet/CinC Challenge aims to identify different heart diseases using different numbers of leads. In this work, a total of 43,101 ECG data was used to build the model. From these data, we proposed a new feature called median complex, which could reduce noise interference and maintain the morphological characteristics of the ECG. In addition, 350 features including rhythm features were extracted in the feature extraction phase. After that, the extracted features were fed into XGBoost containing 1000 trees for classification. In the last submission, we (team name = BJUT) achieved a Challenge metric score of 0.372, -0.406, -0.406 and -0.406 on 12-lead, 6-lead, 3-lead, 2-lead validation datasets respectively.