Federated Deep Reinforcement Learning for Privacy-Preserving Robotic-Assisted Surgery

TL;DR

This work tackles privacy challenges in deploying reinforcement learning for robotic-assisted surgery by proposing a Federated Deep Reinforcement Learning (FDRL) framework that train models across multiple hospitals without sharing raw patient data. It combines differential privacy, secure aggregation, and homomorphic encryption with a dynamic policy adaptation mechanism that selects patient-specific policies in real time, aiming to preserve privacy while maintaining high surgical precision. Empirical results on synthetic data show a ~60% reduction in privacy leakage compared with centralized training and precision within ~1.5% of a centralized baseline, even under substantial data heterogeneity. The approach advances privacy-preserving, adaptive AI for clinical robotics and points to future work in real-world validation, hardware acceleration, and formal verification for safety-critical deployment.

Abstract

The integration of Reinforcement Learning (RL) into robotic-assisted surgery (RAS) holds significant promise for advancing surgical precision, adaptability, and autonomous decision-making. However, the development of robust RL models in clinical settings is hindered by key challenges, including stringent patient data privacy regulations, limited access to diverse surgical datasets, and high procedural variability. To address these limitations, this paper presents a Federated Deep Reinforcement Learning (FDRL) framework that enables decentralized training of RL models across multiple healthcare institutions without exposing sensitive patient information. A central innovation of the proposed framework is its dynamic policy adaptation mechanism, which allows surgical robots to select and tailor patient-specific policies in real-time, thereby ensuring personalized and Optimised interventions. To uphold rigorous privacy standards while facilitating collaborative learning, the FDRL framework incorporates secure aggregation, differential privacy, and homomorphic encryption techniques. Experimental results demonstrate a 60\% reduction in privacy leakage compared to conventional methods, with surgical precision maintained within a 1.5\% margin of a centralized baseline. This work establishes a foundational approach for adaptive, secure, and patient-centric AI-driven surgical robotics, offering a pathway toward clinical translation and scalable deployment across diverse healthcare environments.

Figures (7)

• Figure 1: A FDRL framework for surgical robotics with privacy-preserving techniques. RL policies $P_N$, are trained across multiple hospitals without direct data sharing and securely aggregated.
• Figure 2: Federated Deep Reinforcement Learning architecture for privacy-preserving robotic-assisted surgery. The framework consists of decentralised RL training at hospital nodes using differential privacy (DP) and homomorphic encryption (HE), secure aggregation at the federated server, and adaptive policy selection at the robotic decision-making layer.
• Figure 3: Performance comparison of Federated versus Centralised RL: (a) PLR over 50 global rounds; (b) KL divergence between local and global policy distributions; (c) OPE as a weighted combination of PLR, KL divergence, and DP decay, demonstrating superior privacy–utility trade-off in Federated RL.
• Figure 4: Impact of Differential Privacy on Model Accuracy: A Trade-Off Analysis.
• Figure 5: Comparison of FL and CL across key evaluation metrics: Task-Specific Accuracy, Surgical Risk Mitigation, and Overall Privacy Effectiveness (OPE).