Optimizing Heart Valve Surgery with Model-Free Catheter Control

Abstract

Currently, cardiac catheters for Structural Heart Disease (SHD), are maneuvered manually through the vascular pathway to the chambers of the heart by skilled surgeons. Given the complexity of these maneuvers, we aim at introducing a variable shared autonomy robotic platform for intra-procedural support, by robotizing the commercial MitraClipTM System (MCS). The MCS allows the treatment of mitral regurgitation by percutaneously implanting a clip that grasps the valve leaflets. In light of that, the aim of this paper is to propose a position control strategy that guarantees good trajectory tracking. In the field of control of catheter robots, having a good model is a key point in order to obtain reliable control. A model-based approach on the assumption of a constant curvature (CC) model has been proposed by [1]. The CC model, however, involves simplifying assumptions about catheter shape and external loading, moreover, nonlinearities of the catheter (as dead zones and tendon slack) are usually neglected. In order to include such nonlinearities, the Cosserat Rod model has been exploited by [2]; however, this complicates the model and involves high computational costs which makes the control not feasible in real-time. Modelfree controllers based on machine learning represent a valid alternative to analytical models, considering their potential in model uncertainties that strongly influence soft robot control [3]. In [4] they proposed a formulation for learning the inverse kinematics of a continuum manipulator while integrating the end-effector position feedback. We developed a Neural Network based Inverse Kinematic Controller (IKC) shown in the scheme in Fig. 1. The inputs of the net are the target tip pose, 𝒑¯𝑘+1 at the next time instant, the current servomotors position, 𝒒𝑘 , and the current tip pose 𝒑𝑘 , while the output is the position of the servomotor at the next time instant 𝒒𝑘+1 . Our goal is to build a robust control starting from the state-of-the-art control applied to the MCS presented in [5] by X. Zhang et all and adding to it the control also in the orientation of the tip. Moreover, we characterize the control model proposed, by testing its robustness at different motors’ velocities.