Image Processing Framework for Accuracy Measurements in Real-Time Endovascular Simulation

Abstract

Training surgeons to navigate catheters and guidewires through blood vessels is challenging, and simulators are increasingly used to practice these procedures safely. However, no standard method currently exists to objectively measure how realistically a simulator reproduces the behavior of real surgical instruments. This thesis presents an image-based framework for the quantitative comparison of physical and simulated catheter configurations, combining camera-based image acquisition, centerline extraction, geometric alignment, and shape comparison metrics. Physical experiments were performed by imaging catheters and guidewires both on flat surfaces and inside a vascular phantom (Physical SIM). The same instrument configurations were then reproduced in the Mentice simulator (VIST), allowing a direct comparison between the physical and virtual setups. The framework was further extended with an automated optimization module that searches for simulator stiffness parameters that best reproduce the observed physical catheter behavior using Bayesian optimization. Validation experiments demonstrated high accuracy in catheter length estimation, achieving a mean absolute error of 0.117 cm and a coefficient of determination exceeding R2 > 0.99. It also remained consistent regardless of the catheter shape. The alignment procedure showed that the extracted centerlines converge to a close geometric correspondence after registration. Optimization experiments further showed that simulator stiffness parameters could be tuned to reproduce physical catheter configurations with high geometric similarity, although discrepancies remained for certain catheter-guidewire combinations. A direct comparison between the physical and virtual environments was additionally limited by fundamental differences in anatomy representation. The results demonstrate that the proposed framework enables quantitative and repeatable evaluation of endovascular simulator realism. Among the evaluated metrics, RMS error and curvature analysis were found to be the most informative: RMS error for localizing spatial deviations along the catheter shaft, and curvature analysis for capturing mechanical differences in instrument bending behavior. The framework provides a foundation for future validation and calibration of catheter mechanics in anatomically realistic simulation environments.