Concept for a Reinforcement Learning Approach to Navigate Catheters Through Blood Vessels

Minimally invasive procedures are state of the art in the treatment of cardiovascular diseases. The navigation of catheters is a challenging task for cardiologists. Therefore, new robot-based techniques are emerging to assist them. However, the task of navigation remains with the cardiologist. In this paper, a concept is presented for automating the navigation of a catheter using reinforcement learning on a simulation. The results show that the agent is able to learn how to navigate to a defined target point, which allows the study of more realistic scenarios.