+ <p>ProtoECGNet is a prototype-based deep learning model for multi-label ECG classification that mirrors clinical interpretation by learning distinct prototypes for rhythm, morphology, and global abnormalities. It introduces a contrastive loss to structure the prototype space based on diagnostic co-occurrence, achieving near state-of-the-art performance on PTB-XL while providing faithful, case-based explanations grounded in real ECG segments from the training set. ProtoECGNet delivers transparent reasoning across all 71 labels, spanning a comprehensive range of cardiac abnormalities in 12-lead ECGs.
0 commit comments