AI - Based Cyber Attack Detection in Internet of Medical Things

A Review of Edge–Cloud Deployment and On-Device Learning

Abstract

The Internet of Medical Things (IoMT) connects patient monitors, infusion pumps and wearable sensors to hospital networks. This connectivity supports continuous and remote care, but it also means that a cyber attack on a device or gateway can expose patient data or interfere with treatment. Our final-year project aims to detect at-tacks such as denial of service, malware and unauthorised access by analysing IoMT network traffic with artificial intelligence. Choosing a good detection model is only part of the task. The model must also run somewhere (device, gateway, hospital server or cloud), and each choice affects response time, privacy and cost. This paper reviews the twenty studies listed in our project synopsis. They cover five themes: the edge–cloud computing continuum, edge computing for IoT, orchestration and benchmarking, dis-tributed and federated learning, and on-device AI including Tiny ML, knowledge distillation and test-time adaptation. The reviewed work shows that processing near the data source shortens response time, that federated learning allows training without pooling raw data, and that compact models can run and even adapt on very small hardware. However, the studies rarely treat security detection as the workload being placed, and none of the on-device methods targets network traffic. Based on these gaps, we outline a conceptual edge–cloud framework for our project, which has not yet been validated.