A Laboratory-Based Federated Learning Deployment on Real Devices for ECG-Based Clinical Decision Support Systems

Integrating Federated Learning (FL) in the Internet of Medical Things (IoMT) enables the development of privacy-preserving Clinical Decision Support Systems (CDSS) for real-world healthcare applications. However, most FL research has been confined to simulated environments. This work presents a fully deployed real-world FL deployment for CDSS, validated through an electrocardiogram (ECG) arrhythmia detection scenario running on heterogeneous Internet of Things (IoT) edge devices. Unlike simulated approaches, the proposed system implements an IoT setup that allows collaborative model training without sharing or centralizing sensitive ECG data. Performance evaluations on a maximum of eight devices with varying computational capabilities demonstrate the presented framework's adaptability, achieving an F1 score of 93% compared to a centralized approach (97%). The results indicate that real-world FL deployment achieves competitive accuracy while significantly enhancing data privacy, system scalability, and practical feasibility. By bridging the gap between FL theory and real-world implementation, these findings validate the potential of FL for Artificial Intelligence (AI) driven CDSS for personalized patient-centric healthcare.Clinical relevance- This work demonstrates that AI models for early diagnosis and personalized treatment can be trained on decentralized patient data using a Federated Learning approach that preserves privacy, achieves competitive performance comparable to centralized methods, and enables scalable clinical decision support across heterogeneous healthcare environments.