Evaluation of an Autonomous Navigation Method for Vascular Interventional Surgery in Virtual Environment

Abstract

With the increasing use of vascular interventions, catheter navigation in complex vessels has become even more critical. Vascular intervention surgeries also require more precise manipulation and a high level of automation. In this paper, a reinforcement learning-based navigation model was designed and implemented in virtual environment. The task is to insert the catheter and guidewire into the aortic arch automatically. The whole experiment was carried out in the virtual environment based on SOFA engine. Firstly, the model of vessel and catheter/guidewire was established. Secondly, a reinforcement learning method (asynchronous advantage actor-critic) was used to test the performance of catheter autonomous navigation in vessels. Finally, the results were analyzed and compared with manual operation. The results demonstrate that the automatic insertion has shorter operation time and less contact force.