Inverse Reinforcement Learning algorithm for intra-vascular and intra-cardiac catheter's navigation in Minimally Invasive Surgery

Structural Intervention Cardiology (SIC) is a miniinvasive intervention with a catheter based approach for cardiac surgery. Although SIC procedures are becoming increasingly popular, procedures are not ergonomic and technically demanding and, at the same time, high precision and accuracy in reaching target locations inside the human body are necessary for the success these procedures. Thus, there is therefore a need to develop a robust path planner framework to improve the accuracy in target reaching while minimizing interaction with anatomical structures. In this work a pre-operative path-planning method able to guide the catheter from the peripheral access to the desired target position with the needed orientation is proposed. The method exploits an Inverse Reinforcement Learning algorithm based on a combination of Behavioral Cloning (BC) and Generative Adversarial Imitation Learning (GAIL). The method was in-silico tested performing 50 intra-vascular and 70 intra-cardiac paths where the ratio between attempts in which the catheter reaches the target and total number of attempts, computation time, the difference between desired pose and the reached one were considered as validation metrics. Results show that the proposed method computes optimal path enabling the catheter to reach the target with an average error in position below 2 mm in the intra-vascular phase and below 1 mm in position and 6° in orientation in the intra-cardiac phase