A Deep Reinforcement Learning Approach for Autonomous Guidewire Navigation in Vascular Interventional Surgery

With the increasing incidence of cardiovascular and cerebrovascular diseases, minimally invasive vascular interventional surgery has emerged as a crucial therapeutic option. Vascular interventional surgical robots serve as auxiliary tools for surgeons and effectively shield surgeons from radiation exposure. The integration of artificial intelligence technologies into surgical robots is being extensively explored for autonomous navigation, which holds the promise of alleviating the burden on surgeons and enabling more efficient surgical procedures. An autonomous guidewire navigation system for vascular interventional surgery is proposed in this study, which utilized Deep Reinforcement Learning (DRL) to enhance surgical success rate and efficiency. The system utilized a Deep Q-Network (DQN) algorithm to control guidewire movement, utilizing a discretized state space and an optimized reward function to accelerate the learning process. Through experiments on a cerebral neurovascular model, the system demonstrated a significant improvement in success rates and a reduction in the number of operations required compared to human novices.