Federated Learning-Based Secure Electronic Health Record Sharing Scheme in Medical Informatics

Abstract

Medical Cyber-Physical Systems support the mobility of electronic health records data for clinical research to accelerate new scientific discoveries. Artificial Intelligence improves medical informatics, but current centralized data training and insecure data storage management techniques expose private medical data to unauthorized foreign entities. In this paper, a Federated Learning-based Electronic Health Record sharing scheme is proposed for Medical Informatics to preserve patient data privacy. A decentralized Federated Learning-based Convolutional Neural Network model trains data locally in the hospital and stores results in a private InterPlanetary File System. A secondary global model is trained at the research center using the local models. Private IPFS secures all medical data stored locally in the hospital. The novelty of this study resides in securing valuable hospital biomedical data useful for clinical research organizations. Blockchain and smart contracts enable patients to negotiate with external entities for rewards in exchange for their data. Evaluation results demonstrate that the decentralized CNN model performs better in accuracy, sensitivity, and specificity, similar to the traditional centralized model. The performance of the Private IPFS exceeds the Blockchain-based IPFS based on file upload and download time. The scheme is suitable for promoting a secure and privacy-friendly environment for sharing data with clinical research centers for biomedical research.