Privacy Utility Optimized Adaptive Framework in Medical Cyber Physical Systems

Abstract

Medical Cyber-Physical Systems (MCPS) play a pivotal role in modern healthcare by enabling real-time monitoring, diagnosis, and treatment through interconnected devices. However, balancing data privacy and data utility in MCPS remains a critical challenge. This paper proposes a Privacy-Utility Optimized Adaptive Framework (PUAF) that dynamically adjusts privacy levels while preserving data utility. The framework integrates Federated Learning (FL) with Adaptive Differential Privacy (ADP) to prevent raw data exposure while enabling accurate model training. A Reinforcement Learning (RL) agent optimizes the privacy-utility trade-off by dynamically adjusting privacy parameters based on real-time system conditions, including data sensitivity, model accuracy, and processing latency. The PUAF employs homomorphic encryption and secure aggregation techniques to protect sensitive patient data during model updates. Experimental results demonstrate that the framework achieves higher model accuracy while enforcing robust privacy constraints, outperforming static privacy-preserving methods. The solution is scalable, low-latency, and applicable to real-world healthcare environments, such as remote patient monitoring and AI-driven diagnostics. By effectively balancing privacy protection and data utility, the proposed PUAF enhances privacy-compliant analytics and improves decision-making accuracy in MCPS, making it a practical solution for next-generation healthcare systems.