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

arXiv cs.AI · 6 hours ago

Researchers developed Angular Gaussian Supervised Contrastive Learning (AG-SCL), a machine learning method to improve diagnosis of rare heart arrhythmias from ECG data when training datasets have imbalanced class distributions. The method achieved a balanced accuracy of 0.838 on the PTB-XL benchmark and 0.918 on a nocturnal ECG dataset of 1317 hours from 141 subjects. The approach combines uncertainty modeling with adaptive adjustment techniques to better detect rare arrhythmias while maintaining high specificity for common conditions.