Angular Gaussian Supervised Contrastive Learning for Long-Tailed Electrocardiogram Arrhythmia Diagnosis

2026-07-16 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Medical AI · Depth: Expert, quick

Summary

Angular Gaussian Supervised Contrastive Learning (AG-SCL) is a novel framework designed to improve deep learning reliability for long-tailed electrocardiogram (ECG) arrhythmia diagnosis, particularly for rare abnormalities. It addresses limitations of existing methods by considering direction-dependent morphological variability across ECG classes, not just class frequency. AG-SCL integrates an Angular Gaussian contrastive branch, which models full-covariance class uncertainty on unit-normalized embeddings; Adaptive Logit Adjustment, learning bounded label-state-specific prior corrections; and tail-aware augmentation, generating morphology-preserving views while protecting the 7-25 Hz QRS-dominant band. Evaluated on the PTB-XL benchmark and a 1317-hour nocturnal ECG dataset from 141 subjects, AG-SCL achieved superior macro-level performance. On PTB-XL, it obtained a balanced accuracy of 0.838, sensitivity of 0.709, and specificity of 0.968. For Noc-ECG, values were 0.918, 0.889, and 0.947, respectively. The method showed significant gains in rare or morphologically unstable rhythm classes.

Key takeaway

For Machine Learning Engineers developing diagnostic systems for long-tailed medical datasets, AG-SCL offers a robust approach to improve rare class detection. You should consider integrating anisotropic representation learning and adaptive prior calibration to enhance sensitivity without sacrificing specificity. This method, particularly its full-covariance modeling and tail-aware augmentation, can significantly boost performance on morphologically diverse and imbalanced data like ECGs.

Key insights

AG-SCL enhances long-tailed ECG diagnosis by combining prior calibration with anisotropic representation learning for rare arrhythmias.

Principles

• Model full-covariance class uncertainty on embeddings.
• Learn adaptive, label-state-specific prior corrections.
• Use tail-aware augmentation for morphology preservation.

Method

AG-SCL unifies an Angular Gaussian contrastive branch, Adaptive Logit Adjustment, and tail-aware augmentation to improve long-tailed multi-label ECG diagnosis.

In practice

• Apply full-covariance modeling to unit-normalized embeddings.
• Implement adaptive logit adjustments for rare classes.
• Generate morphology-preserving augmented views.

Topics

• Electrocardiogram Arrhythmia Diagnosis
• Long-Tailed Learning
• Supervised Contrastive Learning
• Deep Learning
• Medical AI
• Imbalanced Data

Code references

• Open-EXG/AG-SCL-for-Long-Tailed-ECG

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

Related on AIssential

• SafeECGMatch: Calibration-Aware Joint Frequency and Time Space Semi-Supervised Learning for Open-Set ECG Classification — SafeECGMatch improves ECG classification by safely handling out-of-distribution data within semi-supervised learning.
• Robust training of implicit generative models for multivariate and heavy-tailed distributions with an invariant statistical loss — ISL offers a stable, gradient-based alternative to adversarial training for implicit generative models, enhanced for heavy-tailed and multivariate data.
• Domain-Aware Hierarchical Contrastive Learning for Semi-Supervised Generalization Fault Diagnosis — DAHCL improves semi-supervised fault diagnosis by correcting pseudo-label bias and leveraging uncertain unlabeled data through domain-aware and hierarchical contrastive learning.
• Enhanced fuzzy joint mutual information with Cuckoo Search Algorithm and genetic algorithm for diagnostics of chronic kidney disease prediction — The EFJMI + CSA-GA model significantly enhances CKD prediction by optimizing feature selection in fuzzy, high-dimensional healthcare data, achieving high accuracy.
• ECG-Lens: Benchmarking ML & DL Models on PTB-XL Dataset — Deep learning models, especially complex CNNs, significantly outperform traditional ML for raw 12-lead ECG classification.
• A self-supervised electrocardiogram foundation model for empowering cardiovascular disease prediction and genetic factor discovery — A self-supervised ECG foundation model improves CVD prediction and uncovers novel genetic associations.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.