T-BFL model based on two-dimensional trust and blockchain-federated learning for medical data sharing: T-bfl model based on two-dimensional trust and blockchain

The data sharing model that integrates federated learning and blockchain is an effective solution for protecting medical data privacy. However, in this sharing mode, there is no pre-established identity trust relationship between users, and the behavior of nodes is not fully considered, which still poses security risks during collaborative training. In response to the above issues, this article proposes a medical data sharing model T-BFL based on two-dimensional trust and blockchain joint learning, which guarantees the trustworthiness of users in sharing through "identity+behavior" two-dimensional trust. In terms of identity trust, we have designed an identity trusted registration scheme based on blockchain oracle to ensure the credibility of registered users’ identities. In terms of behavioral trust, in order to comprehensively assess credibility, we adopt Bayesian theory to eliminate uncertain interactions in the training process on the basis of the subjective logic trust model, add a trust penalty mechanism for participant behavioral misdeeds, take into account the time-decaying nature of the historical trustworthiness and cross-domain trust in healthcare alliances, and improve the direct trust value based on the induced ordered weighted average IWOA operator, and introduce a weighted path’s trust decay model, thus improving the validity and adaptability of behavioral trustworthiness calculation results. Experimentally, we verify the effectiveness of the identity trusted user registration scheme based on the blockchain prediction machine, and validate the robustness and adaptability of the model. The improved behavioral trust calculation model can better identify malicious nodes, and the accuracy of the shared model is higher than 90% under different datasets, and the comparative analysis shows that T-BFL has certain advantages.