In order to solve the data sharing problem, which is one of the biggest challenges in building AI models in the medical field, this paper proposes a global healthcare modeling framework without directly sharing datasets from different continents (Europe, North America, and Asia). We choose blood glucose management as a research model to verify its effectiveness, and implement blockchain-based federated learning to meet the privacy and safety requirements of healthcare data, and reward honest participation and punish malicious activities through an on-chain incentive mechanism. Experimental results show that the proposed framework is effective, efficient, and privacy-preserving, and has higher prediction accuracy than models trained with limited personal data, and achieves similar or slightly better results than centralized training in certain scenarios. This study opens the way for international healthcare project collaboration where additional data is crucial to reduce bias and benefit humanity.
