Decentralization of Learning and Trust in the Healthcare: Blockchain-driven Federated Learning for Alzheimer's MRI Image Classification

This paper proposes a framework that combines Federated Learning (FL) and blockchain technologies in applications where sensitive data need to be analyzed. FL allows exchanging machine learning model parameters instead of sensitive data, thus ensuring data privacy preservation. Model parameters are ciphered and stored into the InterPlanetary File System (IPFS). Coordination via a dedicated smart contract allows to efficiently handle the parameters update phases, fortifying data security. We validate our approach using an Alzheimer's MRI image dataset, showing the benefits in terms of practical implementation and classification accuracy.