DiffusionLite-ECG: A Lightweight Diffusion-Based Framework for Joint R-Peak Detection and Arrhythmia Diagnosis
Year: 2026
Type: conference-paper
Abstract: Automatic electrocardiogram (ECG) classification for arrhythmia detection remains a challenging task in clinical practice, largely due to its strong dependence on accurate R-peak detection and complex feature extraction pipelines. Conventional ECG analysis systems typically adopt a two-stage framework, in which R-peak detection and arrhythmia diagnosis are performed independently, leading to error propagation and limited robustness under abnormal or noisy conditions. To address these limitations, this paper proposes DiffusionLite-ECG, a lightweight one-dimensional diffusion-based framework that jointly performs R-peak detection and arrhythmia diagnosis in an end-to-end manner. By unifying these two tasks within a single generative learning framework, DiffusionLite-ECG enables task-aware representation learning directly from raw ECG signals and effectively mitigates the structural dependency inherent in conventional pipelines. The proposed model further integrates Dilated Self-Organized Operational Neural Networks (DilSelfONN), a biologically inspired heterogeneous operator that enhances nonlinear modeling capability while significantly reducing model complexity. The proposed method is evaluated on the large-scale CPSC-DB dataset, achieving an R-peak detection F1-score of 99.14% and an arrhythmia classification accuracy of 98.99% with substantially fewer parameters than standard diffusion-based models. In addition, cross-dataset validation on the MIT-BIH Arrhythmia Database demonstrates strong robustness and generalization under distribution shifts. These results indicate that DiffusionLite-ECG provides an effective and practical solution for robust ECG analysis in resource-constrained real-world settings.
Institutions: Chery Automobile (China), Anhui University
Language: English
Topic: ECG Monitoring and Analysis
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: gold
OpenAlex ID: W7224940546