SafeECGMatch: Calibration-Aware Joint Frequency and Time Space Semi-Supervised Learning for Open-Set ECG Classification

2026-06-06 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Health & Medical Research · Depth: Expert, quick

Summary

SafeECGMatch is a novel calibration-aware semi-supervised learning (SSL) framework designed for single-label Electrocardiogram (ECG) classification, specifically addressing label scarcity and the presence of out-of-distribution (OOD) anomalies in unlabeled data pools. Standard SSL methods often misclassify OOD data, leading to overconfident, incorrect predictions. SafeECGMatch mitigates this by employing a dual-branch architecture that extracts time-frequency latent representations using ECG-specific augmentations. It dynamically aligns confidence with empirical accuracy through adaptive label smoothing and temperature scaling, calibrating both the multiclass classifier and the OOD detector across temporal and spectral domains. This joint optimization enables trustworthy OOD rejection and reliable pseudo-labeling, achieving state-of-the-art accuracy and calibration on the PTB-XL and PhysioNet/CinC Challenge benchmarks.

Key takeaway

For Machine Learning Engineers developing ECG classification models with limited labeled data, SafeECGMatch offers a robust approach to integrate unlabeled data safely. Its calibration-aware semi-supervised learning prevents overconfident predictions on out-of-distribution anomalies, ensuring more reliable diagnostic support. Consider adopting its dual-branch, time-frequency approach to enhance both accuracy and trustworthiness in clinical applications, particularly when dealing with diverse or anomalous physiological time-series data.

Key insights

SafeECGMatch improves ECG classification by safely handling out-of-distribution data within semi-supervised learning.

Principles

• Standard SSL overconfidently mislabels unseen OOD classes.
• Aligning confidence with accuracy improves OOD detection and pseudo-labeling.

Method

SafeECGMatch uses a dual-branch architecture for time-frequency representations, applying adaptive label smoothing and temperature scaling for calibration across temporal and spectral domains.

In practice

• Utilize ECG-specific augmentations for robust feature extraction.
• Implement confidence-accuracy alignment for OOD detection.

Topics

• ECG Classification
• Semi-Supervised Learning
• Out-of-Distribution Detection
• Calibration
• Time-Frequency Analysis
• Physiological Time-Series

Code references

• labhai/SafeECGMatch

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• A spectral audit framework reveals task-dependent aperiodic reliance across EEG and ECG deep learning — Deep learning models on physiological data often rely on aperiodic signals, impacting task performance.
• ROBUST-WT: Robust Uncertainty-aware Segmentation Transform via Whitening and Training Enhancements — ROBUST-WT demonstrates that targeted training enhancements can substantially improve medical image segmentation robustness without altering core architecture.
• Practical estimation of the optimal classification error with soft labels and calibration — Isotonic calibration enables consistent Bayes error estimation even with corrupted soft labels, provided their order is preserved.
• Joint Score-Threshold Optimization for Interpretable Risk Assessment — A new framework optimizes healthcare risk assessment by addressing partial supervision and asymmetric misclassification costs.
• A multi-task learning approach combining regression and classification tasks for joint feature selection — MTLComb balances mixed regression and classification tasks for unbiased joint feature selection by aligning regularization paths.
• Entropy-Guided Self-Supervised Learning for Medical Image Classification — Combining ImageNet and entropy-guided MAE pre-training via ensemble significantly boosts medical image classification accuracy and robustness.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.