Trevor K Stephens1, John J O'Neill2, Nathan J Kong2, Mark V Mazzeo3, Jack E Norfleet3, Robert M Sweet4, Timothy M Kowalewski2. 1. Department of Mechanical Engineering, University of Minnesota, Minneapolis, MN, USA. steph594@umn.edu. 2. Department of Mechanical Engineering, University of Minnesota, Minneapolis, MN, USA. 3. Simulation and Training Technology Center, Army Research Laboratory, Orlando, FL, USA. 4. Department of Urology, University of Washington, Seattle, WA, USA.
Abstract
PURPOSE: This work presents an estimation technique as well as corresponding conditions which are necessary to produce an accurate estimate of grip force and jaw angle on a da Vinci surgical tool using back-end sensors alone. METHODS: This work utilizes an artificial neural network as the regression estimator on a dataset acquired from custom hardware on the proximal and distal ends. Through a series of experiments, we test the effect of estimation accuracy due to change in operating frequency, using the opposite jaw, and using different tools. A case study is then presented comparing our estimation technique with direct measurements of material response curves on two synthetic tissue surrogates. RESULTS: We establish the following criteria as necessary to produce an accurate estimate: operate within training frequency bounds, use the same side jaw, and use the same tool. Under these criteria, an average root mean square error of 1.04 mN m in grip force and 0.17 degrees in jaw angle is achieved. Additionally, applying these criteria in the case study resulted in direct measurements which fell within the 95% confidence bands of our estimation technique. CONCLUSION: Our estimation technique, along with important training criteria, is presented herein to further improve the literature pertaining to grip force estimation. We propose the training criteria to begin establishing bounds on the applicability of estimation techniques used for grip force estimation for eventual translation into clinical practice.
PURPOSE: This work presents an estimation technique as well as corresponding conditions which are necessary to produce an accurate estimate of grip force and jaw angle on a da Vinci surgical tool using back-end sensors alone. METHODS: This work utilizes an artificial neural network as the regression estimator on a dataset acquired from custom hardware on the proximal and distal ends. Through a series of experiments, we test the effect of estimation accuracy due to change in operating frequency, using the opposite jaw, and using different tools. A case study is then presented comparing our estimation technique with direct measurements of material response curves on two synthetic tissue surrogates. RESULTS: We establish the following criteria as necessary to produce an accurate estimate: operate within training frequency bounds, use the same side jaw, and use the same tool. Under these criteria, an average root mean square error of 1.04 mN m in grip force and 0.17 degrees in jaw angle is achieved. Additionally, applying these criteria in the case study resulted in direct measurements which fell within the 95% confidence bands of our estimation technique. CONCLUSION: Our estimation technique, along with important training criteria, is presented herein to further improve the literature pertaining to grip force estimation. We propose the training criteria to begin establishing bounds on the applicability of estimation techniques used for grip force estimation for eventual translation into clinical practice.
Keywords:
Artificial neural network; Grip force estimation; Surgical robotics
Authors: Wael Othman; Kojo E Vandyck; Carlos Abril; Juan S Barajas-Gamboa; Juan P Pantoja; Matthew Kroh; Mohammad A Qasaimeh Journal: IEEE J Transl Eng Health Med Date: 2022-06-08