融合联邦学习与区块链的医疗数据共享方案

Medical Data Sharing Scheme Combined with Federal Learning and Blockchain

随着医疗大数据的发展,医疗数据安全、个人隐私保护等问题日益突出。为在高效利用各个医疗机构医疗数据的同时保护病人的隐私,提出一种将联邦学习与区块链相结合的医疗数据共享与隐私保护方案。使用联邦学习对多源医疗数据进行建模,将训练的模型参数和医疗机构的声誉值存储于区块链上,并利用区块链对贡献高质量数据的医院进行奖励。通过分析数据源质量对联邦学习算法性能的影响,提出一种基于双重主观逻辑模型的声誉值计算算法来改进联邦学习的精确度,使用改进的声誉机制保证在数据共享中筛选数据源的效率,并利用区块链和联邦学习技术,提高共享效率和实现隐私保护。此外,利用Tensorflow搭建分布式平台并对算法性能进行对比分析,实验结果表明,所提方案能够筛选出高质量的数据源,减少边缘节点与恶意节点的交互时间,即使当声誉值在0.5以上时,也能达到0.857的学习精确度。

With the development of medical big data,the security of medical data and personal privacy are becoming increasingly prominent.To efficiently use the medical data of various medical institutions and protect the privacy of patients,this study proposes a medical data-sharing and privacy-protection scheme that combines federated learning and blockchain.First,federated learning models the multi-source medical data and stores the trained model parameters and reputation value of the medical institution in the blockchain,which is used to support the high quality data of hospitals.Based on the analysis of the impact of data-source quality on the performance of the federated learning algorithm,a reputation value calculation algorithm based on a double subjective logic model is proposed to improve the accuracy of federated learning using an improved reputation mechanism to ensure the efficiency of screening data sources in data sharing,improve sharing efficiency,and achieve privacy protection through blockchain and federated learning.In addition,a distributed platform is built using Tensorflow for a comparative analysis of the algorithm performance.The results show that the proposed scheme can screen high-quality data sources,reduce the interaction time between edge and malicious nodes,and achieve a learning accuracy of 0.857,even when restricting the reputation value to be greater than 0.5.