Research Of The Reliable Platform Of Distributed Machine Learning For Medical Data Based On Blockchain

Medical data resources have been exploding in recent years.It is of great significance to safely and effectively mine the potentially huge value of learning medical big data,which is of great significance for improving the level of medical research and clinical service quality.In reality,medical data are often controlled by different organizations,and at the same time,those data contain a lot of sensitive information related to the personal privacy of patients,which poses a huge challenge to traditional machine learning using physically distributed islands of medical big data.In response to this problem,this paper conducts research on a trusted distributed medical data learning platform based on blockchain,and on the premise of ensuring data security,protecting data ownership and privacy of sensitive information,fully utilizes the medical data of physically distributed data islands for training a machine learning model,form a mutually beneficial and win-win learning model.The specific main work includes:First,in order to solve the inter-institutional trust problem,a platform system for cross-institution training medical data was studied,an identity management and permission control mechanism based on the hyperledger,a solution for verifying the identity of shared data medical institutions was studied,meanwhile,encryption and certificates were adopted Authentication mechanism to prevent malicious nodes from joining the network.Secondly,in terms of privacy and security of distributed training of medical data,the consensus algorithm called Proof of Information was studied,which allows each data owner to select the node with the right of updating machine learning model by initiating a transaction with flags in each round of iterative training.Only training parameters will be upload with encryption,while the local data used in the model training is not traded.When the preset accuracy or the threshold of the number of iterations is met,the final consensus model is obtained.In addition,in terms of information traceability and model credibility,the chaincode logic was designed to automatically write the operation information that meets the set conditions into the distributed consensus ledger,so that every key operation is resistant to denial and traceability.Finally,a complete distributed medical data machine learning platform system was implemented,and functions such as medical institution registration,distributed medical data machine learning model training,and related information on-chain testing were tested.Experiments were conducted to verify the performance between the model training by local data and training the algorithm Proof of Information,the experiment results show the effectiveness and availability of the blockchain-based trusted distributed medical data learning platform system proposed in this paper.