Federated Contrastive Learning for Privacy-Preserving ECG Arrhythmia Classification
Students & Supervisors
Student Authors
Supervisors
Abstract
Accurate detection of ECG arrhythmias plays a critical role in early diagnosis of cardiovascular diseases, yet traditional methods that rely on centralized data aggregation raise concerns about patient privacy and data sharing regulations. This study proposes a novel framework that integrates federated learning and contrastive learning to address these challenges in a decentralized setting. The proposed approach enables multiple healthcare clients to collaboratively train deep models without sharing sensitive data, while simultaneously leveraging contrastive learning to enhance the quality of ECG feature representations. Experiments are conducted on a public ECG arrhythmia classification dataset using five non-IID data partitions reflecting various class imbalance scenarios. The results demonstrate that the federated contrastive learning framework achieves competitive classification performance, with an F1-score of 0.9789 in the fully balanced setting and improved recall in imbalanced scenarios, highlighting its robustness in detecting minority classes. Key techniques such as client-specific augmentations, mixup regularization, and federated batch normalization contribute to improved generalization under data heterogeneity. This work provides empirical insights into the convergence behavior, representational quality, and privacy-preserving benefits of contrastive objectives in federated environments. The proposed solution shows strong potential for scalable and secure deployment in real-world clinical settings where data diversity and privacy are paramount.
Keywords
Publication Details
- Type of Publication:
- Conference Name: International Conference on Computer and Information Technology
- Date of Conference: 19/12/2025 - 19/12/2025
- Venue: Long Beach Hotel, Cox's Bazar, Bangladesh
- Organizer: IEEE, Bangladesh Section