Preserving Privacy and Security of Electronic Health Records using Blockchain-based Federated Learning (BFL) Framework

The digitization of healthcare data into Electronic Health Records (EHRs) are a crucial component of Machine Learning(ML) for healthcare applications, providing patients with quality health care facilities. EHR contains the sensitive information of individuals, which cannot be shared outside the hospital due to data privacy. Also, ML models developed from a single source lead to biased predictions. Two of the most challenging roadblocks to AI (Artificial Intelligence) advancements are data silos and privacy issues. The paper presents a Blockchain-based Federated Learning(BFL) framework to overcome this issue. This work considers one server and two Hospitals or clients in the FL environment. FL is trained with the CKD(Chronic Kidney Disease) dataset using a fundamental Sequential Deep Learning(DL) model. The model is trained without moving the data from the local premise; instead, the modal is sent to the clients to get the data trained locally, thereby preserving privacy. Also, storing the EHR and model parameters on the blockchain provides security. The Federated server aggregates the local parameters during each communication round. It updates the global model, and the global parameters are sent to the local site to train their local models. The outcome of this work is a global model developed with an accuracy of 92.5 % and a local model with 88.33% for the prediction of CKD. Hence, the BFL framework preserves privacy and security and improves the accuracy of the model trained from different sources.