HeartBeatAI: Deep Learning Framework for ECG Arrhythmia Detection
HeartBeatAI is a deep learning framework for multi-label ECG arrhythmia detection, combining domain generalization, multi-scale feature aggregation, and clinical explainability. It integrates a Squeeze-and-Excitation (SE) ResNet and a Multi-Layer Concentration Pipeline to capture both macro-rhythm and micro-morphological anomalies. The framework uses MixStyle regularization and Label Smoothing to mitigate domain shift. Benchmarking on four large-scale datasets achieved 98% Macro F1-score under intra-source conditions, but Leave-One-Domain-Out evaluations showed significant degradation in detecting rare anomalies. The paper is available on arXiv.
Key facts
- HeartBeatAI is a deep learning framework for 12-lead ECG classification.
- It combines domain generalization, multi-scale feature aggregation, and clinical explainability.
- Uses Squeeze-and-Excitation (SE) ResNet and Multi-Layer Concentration Pipeline.
- Employs MixStyle regularization and Label Smoothing to mitigate domain shift.
- Benchmarked on four large-scale datasets with intra-source and LODO protocols.
- Achieved 98% Macro F1-score under intra-source conditions.
- LODO evaluations showed significant degradation in detecting rare anomalies.
- Paper published on arXiv with ID 2605.24588.
Entities
Institutions
- arXiv