Reinforcement learning driven edge-cloud coordination for secure and energy efficient IoMT

The Internet of Medical Things (IoMT) enables sophisticated medical devices, but it also poses significant challenges in terms of data privacy, real-time processing, and energy efficiency for edge devices with limited resources. In this paper, we propose a hierarchical framework for intelligent and secure IoMT-based healthcare monitoring. At the sensor nodes, Federated Variational Mode Decomposition (VMD) is used to decompose physiological signals and locally extract high-fidelity features, ensuring data privacy. To overcome the computational limitations of microcontroller- based sensor nodes, a SparseBonsai neural network is designed for real-time classification of medical signals on the sensor nodes. A centralized orchestration layer, controlled by a Proximal Policy Optimization (PPO) reinforcement learning agent, makes dynamic decisions on whether to queue data for low-latency processing at the edge server or offload to the cloud, depending on data severity, network conditions, and battery level. To further improve energy efficiency, an advanced Sha-Dragon (Shannon-Entropy Dragonfly) optimization algorithm is proposed for resource and transmission power allocation in the IoT network. For security, a dual-layer approach is adopted: ASCON v1.2 lightweight authenticated encryption is used to secure node-to-edge communications, and a WireGuard VPN with ChaCha20-Poly1305 encryption protects data in transit to the cloud. Experimental validation on a Raspberry Pi 5 testbed with a cloud-connected laptop shows that the proposed system achieves a significant reduction in latency for critical alerts and improves the battery life of IoT nodes (8.5 days) compared to the conventional non-adaptive offloading approach. The results confirm the effectiveness of the proposed framework to facilitate energy-efficient, privacy-preserving, and real-time healthcare monitoring in IoMT.