Kinematics Skill Modeling of Cardiac Catheterization via Deep Learning Method

According to advances in robotic surgery, the importance of data-driven techniques that incorporate deep learning methods is expanding quickly, with a focus on objective surgical skill evaluation. Unlike traditional evaluation where surgeons’ skills are evaluated in the real surgery room, capturing users’ motion kinematics can be used as input for an AI model to assess their skills. For this study, a simulated mechanical setup has been provided for the trainees, focusing on cardiac catheterization procedures. This setup allows users to engage in hands-on practice while simultaneously capturing their hand movements for further evaluation. Trainees have the opportunity to engage in extensive practice on a mechanical setup as a pre-operation procedure, enabling them to develop a deeper familiarity and understanding. The task is to pass the tip of a commercial catheter through curves and level intersections on a plastic transparent blood vessel phantom. The objective is to guide the catheter’s tip from the vessel entry point to the designated ablation target. By conducting various experiments involving both novices and experts, a deep recurrent neural network was employed to extract a skill model by solving a binary classification task. The trained model demonstrated a remarkable $\mathbf{9 2. 3} \%$ accuracy in effectively discerning between the maneuvers performed by novices and experts, indicating the successful implementation of the proposed methodology.