RFLPV: A robust federated learning scheme with privacy preservation and verifiable aggregation in IoMT

Abstract

With the rapid development of the Internet of Medical Things (IoMT), medical institutions are accumulating vast amounts of medical data and aiming to utilize this data to train high-quality deep-learning models for assisting doctors in diagnosis. However, due to the highly sensitive nature of medical data, data fusion has always been problematic. Recently, Federated Learning (FL) has gained significant attention as it allows for model training without exposing raw data. Nevertheless, existing research demonstrates that FL still brings the risk of privacy breaches. Furthermore, within the FL framework, a malicious server could potentially provide forged aggregated results, leading to disastrous consequences for the medical system. In this paper, we propose the RFLPV, a robust FL scheme with privacy preservation and verifiable aggregation to enable collaboration among medical institutions and facilitate the fusion of medical data. Specifically, we preserve gradient privacy by adding pairwise masks. Additionally, we design an efficient verification mechanism to tackle the threat posed by a malicious server. Furthermore, we develop a customized local training strategy to accommodate the heterogeneity of medical data. Finally, we demonstrate the superior performance of RFLPV using a real medical dataset.