🤖 AI Summary
This study addresses the poor generalization and clinical deployment challenges in automated electrocardiogram (ECG) analysis caused by class imbalance and cross-institutional domain shifts. The authors propose a deep learning framework for multi-label arrhythmia detection using 12-lead ECGs, eschewing conventional image-based approaches in favor of direct modeling of both macroscopic rhythm and microscopic morphological abnormalities in the raw time-series signal. The method innovatively integrates a Squeeze-and-Excitation ResNet with a multi-layer concentration pipeline to identify diagnostically critical leads, while incorporating MixStyle regularization and label smoothing to enhance cross-domain robustness. Evaluated on four large-scale datasets, the model achieves a Macro F1-score of 98% under in-domain settings; however, performance on rare arrhythmias degrades substantially in cross-institutional scenarios, highlighting a key challenge for real-world deployment.
📝 Abstract
While Deep Learning (DL) enhances automated electrocardiogram (ECG) analysis, clinical deployment is hindered by class imbalance and the generalization gap. This paper presents HeartBeatAI, a deep learning framework combining domain generalization, multi-scale feature aggregation, and clinical explainability for robust 12-lead ECG classification. Moving beyond image-based paradigms, HeartBeatAI integrates a Squeeze-and-Excitation (SE) ResNet to isolate diagnostic leads alongside a Multi-Layer Concentration Pipeline to capture macro-rhythm and micro-morphological anomalies. To mitigate domain shift, the framework employs MixStyle regularization and Label Smoothing. Rigorous benchmarking across four large-scale datasets using intra-source and Leave-One-Domain-Out (LODO) protocols demonstrates high performance (98% Macro F1-score) under intra-source conditions. However, LODO evaluations reveal significant degradation in detecting rare anomalies, highlighting a persistent challenge in cross-institutional deployment.