This paper presents ProtoECGNet, an interpretable multi-label deep learning model for electrocardiogram (ECG) classification. ProtoECGNet uses prototype-based inference, which bases decisions on similarity to learned representations of real ECG segments, to provide transparent and reliable case-based explanations. It employs a structured multi-branch architecture that mirrors clinical interpretation workflows, integrating a 1D CNN with global prototypes for rhythm classification, a 2D CNN with temporally localized prototypes for morphology-based inference, and a 2D CNN with global prototypes for diffusion anomalies. Each branch is trained with a prototype loss designed for multi-label learning and combines a novel contrastive loss that promotes clustering, separation, diversity, and proper separation between prototypes of unrelated classes. We evaluate ProtoECGNet on all 71 diagnostic labels of the PTB-XL dataset, demonstrating that it performs competitively against state-of-the-art black-box models while providing structured case-based explanations. The representativeness and clarity of the prototypes were verified through structured clinician reviews. ProtoECGNet demonstrates that prototype learning can effectively scale to complex multi-label time series classification, providing a practical path toward transparent and reliable deep learning models for clinical decision support.
