SafeECGMatch: Calibration-Aware Joint Frequency and Time Space Semi-Supervised Learning for Open-Set ECG Classification
SafeECGMatch introduces a calibration-aware semi-supervised learning framework for ECG classification that addresses the critical challenge of handling out-of-distribution anomalies in unlabeled medical data. Using dual-branch time-frequency architecture with adaptive confidence calibration, the method achieves state-of-the-art accuracy while maintaining reliable OOD rejection, advancing trustworthy AI deployment in clinical diagnostics.
SafeECGMatch tackles a fundamental problem in medical AI: deploying classification models with limited labeled data while managing the reality that unlabeled patient pools contain diagnostic conditions absent from training sets. Standard semi-supervised learning frameworks fail in this scenario by forcing pseudo-labels onto anomalous or novel cases, generating false confidence that clinicians cannot safely rely upon. This research bridges that gap by integrating calibration mechanisms directly into the learning process.
The framework's dual-branch architecture exploiting both temporal and frequency domains reflects domain-specific understanding of ECG analysis. Traditional SSL methods treat all unlabeled data as potentially useful; SafeECGMatch distinguishes between in-distribution samples suitable for pseudo-labeling and out-of-distribution anomalies requiring rejection. The adaptive label smoothing and temperature scaling components ensure predicted confidence scores align with actual accuracy, a critical requirement for clinical decision-support systems where overconfident misclassifications pose patient safety risks.
For healthcare AI development, this work demonstrates that reliability and safety concerns need not sacrifice performance. Evaluation on PTB-XL and PhysioNet/CinC Challenge benchmarks validates practical effectiveness. The open-source release enables broader adoption and reproducibility. This advances the maturation of medical AI from purely accuracy-focused metrics toward clinically deployable systems where practitioners understand and trust model limitations.
- →SafeECGMatch combines semi-supervised learning with out-of-distribution detection to prevent dangerous pseudo-labeling of anomalous patient data
- →Dual-branch time-frequency architecture with adaptive calibration achieves state-of-the-art accuracy on ECG classification benchmarks
- →Framework aligns confidence predictions with empirical accuracy through temperature scaling, enabling clinically trustworthy AI deployment
- →Open-source release on GitHub accelerates adoption across medical AI research and clinical applications
- →Method addresses broader challenge of deploying AI with label scarcity while managing distribution mismatch in real-world healthcare data