A Privacy-Preserving and Fault-Tolerant Framework for Real-Time IoMT Analytics via SMPC and Encrypted Graph Neural Networks

Abstract

The Internet of Medical Things (IoMT) enables continuous health monitoring through distributed wearable and implantable devices but faces critical challenges in privacy, latency, trust, and reliability, particularly when aggregating sensitive physiological data across untrusted networks. In this paper, we propose GraphMedSMPC , a novel privacy-preserving framework that integrates threshold-based Secure Multi-Party Computation (SMPC), spectral Graph Neural Networks (GNNs), fog computing, and a custom IoMT-BFT Byzantine fault-tolerant consensus protocol to achieve secure, intelligent, and scalable analytics in dynamic IoMT environments. GraphMedSMPC models patient-device interactions as a dynamic encrypted graph, enabling advanced inference tasks, such as patient risk prediction (92.4% accuracy), epidemic spread modeling (MAE = 0.11 in \( R_0 \) estimation), and anomaly detection (F1-score = 0.91), directly on homomorphically encrypted data using the CKKS scheme under Ring-LWE hardness assumptions, ensuring end-to-end confidentiality without exposing raw health records, even to intermediate fog nodes. A \((k,n)=(6,10)\) threshold SMPC architecture eliminates reliance on trusted authorities, while the IoMT-BFT consensus protocol maintains 98.7% agreement success under up to \( f = 3 \) faulty nodes, satisfying the \( n \geq 3f + 1 \) resilience condition. Fog-layer computation reduces end-to-end latency to 169.8 ms, 58.2% faster than FedCCW, while adaptive compression reduces communication overhead by 84.0% (from 80.4 MB to 12.8 MB per round) compared to TPSRA. Energy consumption is minimized to 0.086 mJ/sample via adaptive sampling and dynamic voltage scaling, extending wearable battery life by over 2 \(\times\) . Security analysis confirms quantum-resistant privacy, with eavesdropping resistance bounded by \( < 2^{-128} \) under standard lattice assumptions. By unifying privacy, scalability, intelligence, and fault tolerance, GraphMedSMPC establishes a new benchmark for decentralized, real-time, and trustworthy healthcare AI systems.