Secure Latency-Aware Task Offloading Using Federated Learning and Zero Trust in Edge Computing for IoMT

The exponential growth of wearable medical devices generates massive healthcare data within the Internet of Medical Things (IoMT) framework, necessitating efficient and secure task processing. Although offloading medical tasks to Edge Servers (ESs) improves the agility of medical services and enhances real-time processing capabilities, it simultaneously introduces significant security and trust concerns in heterogeneous and delay-sensitive IoMT environments. Existing task offloading approaches either compromise on performance or fail to ensure sufficient safeguards against sophisticated cyber intrusions. To address the dual challenge of minimizing latency while ensuring robust security during task offloading in IoMT systems, this paper proposes a Trust and Latency-aware Task Offloading (TLTO) approach in heterogeneous IoMT by integrating ZTA with FL. TLTO achieves decentralized intelligence without compromising data confidentiality and system integrity. Before offloading, ZTA enforces authentication among the ESs and cloud entities, orchestrated through a dedicated zero-trust orchestrator. Within FL, an improved on-policy temporal difference control algorithm is leveraged for local model training. Moreover, this algorithm determines the optimal offloading strategy based on the global model while considering trustworthiness, the heterogeneous nature of tasks, resource availability, and latency. Simulation results validate the efficiency of the TLTO approach in securing task offloading while improving performance in dynamic IoMT environments. This approach provides a scalable and resilient solution to the stringent requirements of privacy-preserving, latency-sensitive healthcare applications.