Force Estimation on Steerable Catheters through Learning-from-Simulation with ex-vivo Validation

Abstract

Monitoring and control of the contact force at the tip of soft flexural robots is of high application need, e.g., the tip force on radiofrequency ablation (RFA) catheters. In this study, a real-time tip force estimation method based on image-based shape-sensing and learning-from-simulation is provided. To this end, a generalized image-based shape-sensing technique for flexural robots was developed using the Bezier spline interpolation method. Afterward, the deflection of a commercial catheter subjected to a series of tip forces was simulated using nonlinear finite element modeling. Next, two independent data-driven models, i.e., artificial neural network (ANN) and support vector regression (SVR), were trained with a dataset with the Bezier spline control points as the inputs and tip forces as the output. For validation, the trained models were used for real-time tip force estimation while the catheter was pressed against porcine atrial tissue. The test was performed using a universal testing machine that recorded the ground-truth contact force. The comparison showed that the ANN model had a mean-absolute-error of 0.0217±0.0191 N, while the SVR model exhibited a mean absolute error of 0.0178 ± 0.0121 N and a correlation coefficient of 0.991. Moreover, the proposed method showed a minimum computational refresh rate of 646 Hz (ANN) and 917 Hz (SVR) during the validation experiment. The performance of the proposed method was in compliance with the clinical requirements of RFA therapy.