Advancing long-tailed pediatric arrhythmia classification with a novel contrastive loss and multimodal learning

Arrhythmias are a major cause of sudden cardiac death in children, making automated rhythm classification from electrocardiograms (ECGs) clinically important. However, pediatric arrhythmia analysis remains challenging because of age-dependent waveform variability, limited data availability, and a pronounced long-tailed class distribution that hinders recognition of rare but clinically important rhythms. To address these issues, we propose a multimodal end-to-end framework for pediatric arrhythmia classification that integrates surface ECG and intracardiac electrogram (IEGM) signals. The model uses dual-branch encoders, attention-based cross-modal fusion, and a lightweight Transformer classifier to learn complementary electrophysiological representations. We further introduce an Adaptive Global Class-Aware Contrastive Loss (AGCACL), combining prototype alignment, class-frequency reweighting, and globally informed hard-class modulation to improve class compactness and separability under imbalance. We evaluate the method on the pediatric subset of the Leipzig Heart Center ECG-Database and establish a reproducible preprocessing pipeline for rhythm-segment construction, denoising, and label grouping. Against nine reproduced state-of-the-art (SOTA) arrhythmia classification baselines covering convolution-based, lightweight, Transformer-based, and multimodal fusion frameworks, the proposed approach achieves 96.22% Top-1 accuracy and improves macro precision, macro recall, macro F1 score, and macro F2 score by 4.42, 1.17, 4.97, and 4.08 percentage points, respectively, over the strongest reproduced baseline. Together with the established pediatric preprocessing and evaluation pipeline, these results suggest practical advantages over existing frameworks for pediatric long-tailed rhythm analysis: the ECG–IEGM design leverages complementary electrophysiological information, improves recognition of rare but clinically important rhythm categories, and may serve as a clinician-in-the-loop tool for assisted interpretation and priority screening.