Abstract

In this thesis, an image-based force estimation schema for tendon-driven steerable catheters with the application in robot-assisted tissue ablation procedures was proposed and validated.
To this end, initially, an inverse kinematics solution for modeling the catheter deformation and estimating the contact force between the catheter and cardiac tissue was proposed. Next, a tendon-driven catheter was designed and developed in-house. Afterwards, the developed catheter was integrated with a motor current control software for controlling and recording the deformation of the catheter. Furthermore, the performance of the proposed method in contact force estimation was improved by various learning approaches. Validation studies were performed on phantom tissue as well as excised porcine tissue. The results of the validation studies showed that the proposed image-based force estimation models had a maximum root-mean-square error of 0.012 N for the tests on the porcine atrial tissue. Furthermore, adopting machine learning approaches resulted in the accuracy of 97.7%, 97.6%,
98.3%, and 98.8% for the Support vector machine, Random Forest, AdaBoost, and Deep Neural Network, respectively. For yielding optimal models, hyper-parameter searches were performed for all proposed learning models. In summary, the proposed force estimation system did not necessitate the utilization of force sensors and could successfully contribute in operation rooms during cardiac catheterization.