Federated Edge Artificial Intelligence Framework for Privacy-Preserving Internet of Medical Things

Abstract

The Internet of Medical Things (IoMT) has been quickly adopted in the intelligent healthcare industry, introducing more and more medical data to be analysed at the network edge, which requires medical information analytics to be secure, privacy-preserving, and low-latency. When deployed in the heterogeneous edge environment, conventional federated learning approaches, however, tend to have high communication overhead, long inference time, long training time and insufficient privacy protection. To solve these problems, this paper presents the Federated Edge Artificial Intelligence (FedEdgeMed-AI) framework for Privacy-Preserving Internet of Medical Things, which incorporates edge intelligence, Federated Model Aggregation, Adaptive Privacy-Preserving Optimization, and Resource Management for distributed healthcare applications. The proposed framework is tested against the MIMIC-IV dataset in EdgeCloudSim environment and compared to FedAvg, FedProx and Blockchain-Assisted Federated Learning (BAFL). Experimental results show that FedEdgeMed-AI can reach an accuracy of 98.94%, a privacy preservation rate of 99.12%, and reduce training time to 31.48 min, response time to 84.67 ms, inference latency to 18.93 ms, and communication cost to 42.15 MB. Compared to the currently best model, the FedEdgeMed-AI model not only achieves a higher privacy preservation rate of 4.2%, but also boosts accuracy by 3.6%, shortens training time by 28.4%, response time by 32.7%, inference latency by 35.1%, and communication cost by 38.9%. These results show that the proposed framework is an effective and scalable solution for secure, privacy-preserving and real-time IoMT-enabled intelligent healthcare applications, which is suitable for next-generation applications in the medical field at the edge.