Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Advancing Few-Shot Pediatric Arrhythmia Classification with a Novel Contrastive Loss and Multimodal Learning

Yiqiao Chen, Zijian Huang, Zhenghui Feng

Abstract

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 that integrates surface ECG and intracardiac electrogram (IEGM) signals for pediatric arrhythmia classification. The model combines dual-branch feature 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), which incorporates prototype-based alignment, class-frequency reweighting, and globally informed hard-class modulation to improve intra-class compactness and inter-class separability under class imbalance. We evaluate the proposed method on the pediatric subset of the Leipzig Heart Center ECG-Database and establish a reproducible preprocessing pipeline including rhythm-segment construction, denoising, and label grouping. The proposed approach achieves 96.22% Top-1 accuracy and improves macro precision, macro recall, macro F1 score, and macro F2 score by 4.48, 1.17, 6.98, and 7.34 percentage points, respectively, over the strongest baseline. These results indicate improved minority-sensitive classification performance on the current benchmark. However, further validation under subject-independent and multicenter settings is still required before clinical translation.

Advancing Few-Shot Pediatric Arrhythmia Classification with a Novel Contrastive Loss and Multimodal Learning

Abstract

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 that integrates surface ECG and intracardiac electrogram (IEGM) signals for pediatric arrhythmia classification. The model combines dual-branch feature 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), which incorporates prototype-based alignment, class-frequency reweighting, and globally informed hard-class modulation to improve intra-class compactness and inter-class separability under class imbalance. We evaluate the proposed method on the pediatric subset of the Leipzig Heart Center ECG-Database and establish a reproducible preprocessing pipeline including rhythm-segment construction, denoising, and label grouping. The proposed approach achieves 96.22% Top-1 accuracy and improves macro precision, macro recall, macro F1 score, and macro F2 score by 4.48, 1.17, 6.98, and 7.34 percentage points, respectively, over the strongest baseline. These results indicate improved minority-sensitive classification performance on the current benchmark. However, further validation under subject-independent and multicenter settings is still required before clinical translation.

Paper Structure

This paper contains 16 sections, 15 equations, 11 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Dataset and Preprocessing Pipeline
  4. Proposed Multimodal Architecture
  5. ResNet-based Modality Encoding Process
  6. Attention-based Cross-modal Semantic Alignment Fusion
  7. Transformer-based Classification Head
  8. Adaptive Global Class-aware Contrastive Loss
  9. Training Strategies
  10. Experimental Setup
  11. Experimental Results and Discussion
  12. Comparison with State-of-the-Art (SOTA) Reproduced Baselines
  13. Effectiveness of AGCACL: Comparison with Contrastive Losses
  14. Ablation Studies
  15. Limitations and Generalizability
  16. ...and 1 more sections

Figures (11)

  • Figure 1: Overview of the proposed framework for long-tailed pediatric arrhythmia classification. The framework comprises dataset preprocessing, multimodal ECG–IEGM feature learning, and optimization with the proposed Adaptive Global Class-Aware Contrastive Loss (AGCACL).
  • Figure 2: Representative single-lead ECG segment before and after denoising. The denoised waveform preserves the main morphological structure while reducing high-frequency noise.
  • Figure 3: Illustration of the ECG preprocessing pipeline. Step 1 discards unannotated signal segments. Step 2 segments the remaining signal based on rhythm labels provided by experts. Step 3 slices each rhythm segment into fixed 2-second windows to ensure consistent labeling. Only windows fully contained within a single rhythm type are preserved. This pipeline ensures the generation of structurally consistent and label-pure ECG segments for model training.
  • Figure 4: Performance comparison of the proposed method and reproduced state-of-the-art baselines on the pediatric arrhythmia classification task. Results are reported in terms of Top-1 accuracy and macro-averaged specificity, precision, recall, F1, and F2.
  • Figure 5: Radar-chart comparison of the proposed method and reproduced state-of-the-art baselines across six evaluation metrics.
  • ...and 6 more figures