Autonomous Guidewire Navigation in Vascular Interventional Surgery Using Deep Reinforcement Learning

In vascular interventional therapy, digital subtraction angiography (DSA) is used to guide a guidewire to lesions for therapeutic intervention. However, the complexity of human vascular structures and the guidewire’s mechanical characteristics pose significant challenges to traditional control methods in robotic surgery systems. Recent advancements in deep reinforcement learning (DRL) offer a promising alternative, enabling autonomous robotic navigation without exhaustive environmental modeling. This paper introduces an innovative autonomous guidewire navigation method using the Proximal Policy Optimization (PPO) algorithm. We developed a prototype of a vascular interventional surgery robot capable of precise guidewire manipulation. Our method utilizes a UNet++ network for real-time guidewire segmentation and a B-spline curve for navigation path optimization, eliminating the need for invasive sensors. Experiments on a simulated human vascular model showed that the DRL algorithm effectively navigates flexible guidewire, achieving target accuracy within 10 steps with a success rate above 94%.