A Review of Privacy Protection Technologies in Medical IoT: a Federated Learning Perspective

Abstract

The rapid development of the Internet of Medical Things (IoMT) has driven the transformation of smart healthcare while also triggering the generation and accumulation of massive amounts of sensitive health data, leading to increasingly severe privacy and security risks. The core contradiction is that the construction of high-performance machine learning models requires collaborative training of multi-party data, while privacy regulations such as GDPR and HIPAA strictly require data localization, thus forming a "data island" dilemma. This paper aims to systematically sort out privacy protection technologies in the IoMT field and focus on analyzing the solution path of Federated Learning (FL), a distributed machine learning paradigm. FL relies on the "data does not move, model moves" mechanism, which only exchanges encrypted model parameter updates instead of original data, and can achieve multi-party collaborative modeling while maintaining patient privacy. Based on existing research, this paper points out that FL can achieve performance comparable to centralized learning in medical scenarios (such as medical imaging diagnosis)[2], but its widespread application is still subject to challenges such as considerable communication overhead and system heterogeneity. This paper analyzes that although FL is an effective means to address the privacy and utility trade-off problem, its long-term development urgently needs to be integrated with emerging technologies. For example, integrating blockchain can build a decentralized and auditable trust mechanism[1], thereby shaping a more secure, efficient, and scalable privacy protection framework for Healthcare 5.0. This review clarifies the current state of technology and challenges, and looks forward to future trends, in order to provide useful references for researchers and practitioners.