Identification of 27 abnormalities from multi-lead ECG signals: an ensembled SE_ResNet framework with Sign Loss function

Abstract Objective . Cardiovascular disease is a major threat to health and one of the primary causes death globally. The 12-lead ECG cheap commonly accessible tool identify cardiac abnormalities. Early accurate diagnosis will allow early treatment intervention prevent severe complications cardiovascular disease. Our objective develop an algorithm that automatically identifies 27 abnormalities from databases. Approach Firstly, series pre-processing methods were proposed applied on various data sources in order mitigate problem divergence. Secondly, we ensembled two SE_ResNet models rule-based model enhance performance abnormalities’ classification. Thirdly, introduce Sign Loss tackle class imbalance, thus improve model's generalizability. Main results In PhysioNet/Computing Cardiology Challenge (2020), our approach achieved challenge validation score 0.682, full test 0.514, placed us 3rd out 40 official ranking. Significance We robust predictive framework combines deep neural networks clinical knowledge classify multiple able multi-lead signals regardless discrepancies imbalance labeling. trained five datasets validated it six countries. outstanding demonstrate effectiveness framework.