Privacy-Preserving Lightweight Federated Learning for Heterogeneous Data in Internet of Medical Things

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Data Privacy

Background:

• Internet of Medical Things (IoMT) enables continuous health monitoring via connected devices.
• Arrhythmia detection using electrocardiogram (ECG) is vital for patient care.
• Federated Learning (FL) addresses privacy concerns in distributed healthcare data.

Purpose of the Study:

• To develop a lightweight FL model for arrhythmia detection that handles data heterogeneity.
• To improve the efficiency and privacy of ECG-based arrhythmia detection in IoMT.
• To evaluate the model's performance against existing methods.

Main Methods:

• Proposed a novel lightweight Federated Learning (FL) architecture tailored for ECG data.
• Trained and evaluated the model on two public ECG datasets (PTBD and MIT-BIH arrhythmia).
• Compared computational and communication efficiency against a baseline ResNet model.

Main Results:

• Achieved high accuracy (0.95-0.98) in arrhythmia detection.
• Demonstrated robustness against heterogeneous data distributions.
• Showcased significant reductions in computational cost (MFLOPS) and communication overhead compared to ResNet.

Conclusions:

• The proposed lightweight FL model offers an efficient and privacy-preserving solution for arrhythmia detection in IoMT.
• The model's ability to handle data heterogeneity makes it suitable for diverse clinical applications.
• This approach is promising for resource-constrained environments requiring real-time health monitoring.