Decentralized and Personalized Federated Learning Framework for Privacy Preservation Using IPFS Model Storage Layer in Healthcare

In the quickly changing healthcare environment, this paper presents a novel framework comprising Decentralized and Personalized Federated Learning (DCPFL), which leverages decentralized model storage using Inter-Planetary File System (IPFS). By addressing data privacy and model personalization, we established a consortium blockchain for model integrity and used IPFS for efficient parameter storage. Our innovative synchronization mechanism enhances collaboration between institutions using gossip protocol while safeguarding patient data by training locally using RMSProp, laying the foundation for safer, secure, efficient and personalized healthcare solutions.