Position Estimation of Zone of Different Young’s Modulus Using Deformation Imaging and Convolutional Neural Network

Abstract

The common method to estimate the position of the region with different stiffness is the sensing force information. In the specific case of the contact condition cannot being satisfied, the force sensor cannot be used for distinguishing the region with different stiffness. The researches on a force estimation have done with contact-condition or assumed that the soft tissue has a homogeneous stiffness. In this research, the finite element method simulation of the nonhomogeneous soft tissue model and the depth map of the soft silicone model deforming experiment is used as a vision input data of the proposed algorithm, which estimates the relative position between the pressing tool and the stiffer region. The convolutional neural network is applied to use spatial information of the deformed area and the boundary condition of the liver-shaped model by using 14 classes. The network performs on a test set which is constructed with both simulation data and experiment data shows 0.82 accuracy, and performs on a test set which is constructed with experiment data shows 0.66 accuracy based on top 1 accuracy. The network for distance regression shows 22.33% error.