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

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Internet of Medical Things (IoMT)

Background:

• Federated Learning (FL) is crucial for privacy-preserving Clinical Decision Support Systems (CDSS) in healthcare.
• Most FL research is limited to simulations, lacking real-world validation.
• Integrating FL into IoMT offers potential for decentralized, secure medical data analysis.

Purpose of the Study:

• To present a fully deployed, real-world Federated Learning (FL) system for Clinical Decision Support Systems (CDSS).
• To validate the system using an electrocardiogram (ECG) arrhythmia detection scenario on heterogeneous Internet of Things (IoT) edge devices.
• To demonstrate collaborative model training without sharing sensitive patient data.

Main Methods:

• Implemented a real-world FL deployment on heterogeneous IoT edge devices for CDSS.
• Utilized an electrocardiogram (ECG) arrhythmia detection task for validation.
• Evaluated system performance on up to eight devices with varying computational capacities.

Main Results:

• Achieved an F1 score of 93% for ECG arrhythmia detection, comparable to a centralized approach (97%).
• Demonstrated the framework's adaptability and scalability across devices with varying capabilities.
• Confirmed significant enhancement in data privacy, system scalability, and practical feasibility compared to simulated environments.

Conclusions:

• Real-world FL deployment for CDSS offers competitive accuracy while prioritizing data privacy.
• The system validates the potential of FL for AI-driven, patient-centric healthcare solutions.
• Bridges the gap between FL theory and practical application in IoMT for personalized medicine.