Harnessing Federated Learning for Secure Data Sharing in Healthcare Systems

Background: The digitization of healthcare data has made significant progress in medical research and personalized medicine. Nonetheless, conventional centralized data-sharing structures present primary obstacles to information privacy and security due to laws like the Health Insurance Portability and Accountability Act (HIPAA). Federated Learning (FL) has been proposed as a potential solution that can enable collaborative learning with decentralized datasets without the requirement of data centralization.Objective: This study examines how well Federated Learning performs in meeting both the requirements of secure data exchange between healthcare organizations and achieving high model accuracy without violating any privacy compliance regulations.Methods: A Federated Learning framework was implemented with a neural network model using federated learning on an electronic health records (EHR) database collected from multiple hospitals. Its accuracy was compared to a traditional centralized model across various states, while the people also consider its convergence speed and data leakage risks. We incorporated differential privacy mechanisms in order to improve the security of data sets as well as prevent malicious attacks.Results: The FL model achieved accuracy comparable to the centralized model, with only a marginal reduction. Furthermore, the integration of differential privacy significantly reduced the risk of data breaches, providing robust protection against adversarial attacks.Conclusion: The FL model reported a performance difference which was only slightly reduced as compared with the centralized model. Moreover, this privacy compliance overcomes the risk of data breaches by integrating with differential privacy & securing against adversarial attacks.