SafeRPlan: Safe deep reinforcement learning for intraoperative planning of pedicle screw placement

Yunke Ao, Hooman Esfandiari, Fabio Carrillo, Christoph J. Laux, Yarden As, Ruixuan Li, Kaat Van Assche, Ayoob Davoodi, Nicola Alessandro Cavalcanti, Mazda Farshad, Benjamin F. Grewe, Emmanuel B. Vander Poorten, Andreas Krause, Philipp Fürnstahl

Abstract: Highlights•SafeRPlan enables online screw trajectory planning in robotic surgery.•A distance-based safety filter greatly improves the safety of the procedure.•SafeRPlan deals with noisy partial data and outperforms registration approaches.•A domain randomization approach to surface reconstruction improves generalizability.