Multi-label Arrhythmia Classification From 12-Lead Electrocardiograms

In participation of the PhysioNet/Computing in Cardiology Challenge 2020, we developed a novel computational approach for efficiently identifying cardiac abnormalities from 12-lead electrocardiogram (ECG) data. The developed methodology is composed of three processes: selecting representation, generating features, and predicting outcomes. We proposed a cache-inspired method to select a 12-lead ECG heartbeat representation. Moreover, we devised a physiologically interpretable feature generator for segmented 12-lead ECG signals. For multi-label arrhythmia classification, we innovated an efficient arrhythmia outcome prediction procedure that is adaptable to ECG data of variant lengths. Our team, JuJuRock, received a score of 0.402 using 5-fold cross-validation on the full training data and a score of 0.244 on the final full test data. Team JuJuRock ranked 16th out of the 41 teams that participated in this year's Challenge.