ENHANCING SAFETY IN INTRAOPERATIVE ORTHOPAEDIC SURGICAL PLANNING THROUGH UNCERTAINTY QUANTIFICATION

Robotic-assisted surgical systems have demonstrated the potential to enhance the efficiency and precision of orthopedic procedures. However, most state-of-the-art systems still rely on pre-installed markers on patients and offer limited autonomy in decision-making. The emerging trend toward markerless, image-based autonomous robotic surgery systems presents new possibilities but also introduces significant challenges. These challenges include strict safety requirements, restricted visibility, and sensor noise, which complicate planning and execution. While advanced techniques such as deep reinforcement learning (DRL) and model predictive control (MPC) have excelled in handling complex decision-making tasks in other robotics domains, their adoption in robotic orthopedic surgery has been limited. This is primarily due to the difficulty of providing safety guarantees in dynamic surgical environments and the lack of corresponding simulation platforms to facilitate development. To address these challenges, we propose novel uncertainty-based safety mechanisms that integrate seamlessly with existing DRL and MPC frameworks. First, we developed an uncertainty-aware, distance-based safety filter to enhance the safety performance of DRL policies for surgical planning. Additionally, we introduced a method to compute probabilistic reachable sets (PRS) for linear systems under general sub-Gaussian noise, enabling closed-loop safety guarantees in image-based MPC frameworks. To support and validate our approach, we created realistic simulators for ultrasound-guided spinal surgery, as is shown in Figure 1.c. These simulators serve as testbeds for training DRL agents and benchmarking the safety performance of various techniques. Simulation results (illustrated in Figure 1.a and Figure 1.b) demonstrate the potential of our proposed methods to improve the safety and reliability of autonomous robotic surgery, paving the way for safer and more effective orthopedic interventions.