Empowering Surgical Embodied Intelligence: A Blockchain-based Federated Learning Framework for Collaborative Video Analysis

Abstract Automated surgical video analysis serves as the critical foundation for surgical embodied intelligence. However, privacy regulations and data-sharing restrictions pose significant challenges to its collaborative development in computer-assisted surgery. While federated learning (FL) offers a decentralized approach, it remains vulnerable to privacy breaches and single-point failures due to its reliance on a central server. Here, we propose a blockchain-based federated learning (BFL) framework for secure and collaborative surgical video analysis. BFL integrates blockchain’s decentralized consensus and encryption mechanisms with collaborative learning, eliminating the need for a central coordinator. The framework employs cross-validation guided parameter encryption and dynamic weight aggregation to safeguard data privacy, and adopts a hybrid model architecture ''shared + private'' to balance global generalization with local personalization. We apply BFL to distributed surgical video analysis environments involving both medical center and perspective heterogeneity, using data collected from multiple medical centers. Experimental results show that BFL outperforms traditional federated learning methods in terms of both accuracy and robustness. Overall, by establishing a trustworthy model for collaborative medical AI, our framework facilitates secure data sharing across medical centers, empowering surgical embodied intelligence with the visual perception essential for future autonomous surgical assistance.