Privacy-Preserving Lightweight Federated Learning For IOMT

Abstract

The rapid expansion of the Internet of Medical Things (IoMT) has transformed modern healthcare by
enabling continuous patient monitoring, real-time diagnostics, and personalized treatment. However, the
sensitive nature of medical data raises significant privacy, security, and regulatory concerns, particularly
in distributed healthcare environments. Traditional centralized machine learning approaches require data
aggregation, increasing the risk of data breaches and violating privacy norms such as those outlined in
frameworks like the Health Insurance Portability and Accountability Act (HIPAA). To address these
challenges, this study proposes a privacy-preserving lightweight federated learning (FL) framework for IoMT
systems. The proposed model enables decentralized training of machine learning models across resource-
constrained IoMT devices while ensuring that raw patient data never leaves local nodes. A lightweight
architecture is designed to minimize computational overhead, communication cost, and energy
consumption, making it suitable for wearable devices and edge-based healthcare systems. Furthermore,
the framework integrates privacy-enhancing techniques such as differential privacy and secure aggregation
to protect sensitive information from inference attacks and model inversion threats. Optimization
strategies, including model compression and adaptive communication protocols, are incorporated to
enhance efficiency without compromising model accuracy. Our method reduces communication by 80% and
offers strong privacy assurances while nonetheless attaining 95% of the accuracy of centralized learning.
This paper offers a useful federated learning method to provide effective distributed machine learning with
respect for privacy for next-generation IoMT systems.