Tactile Sensing System and Convolutional Neural Network for Mechanical Property Classification

In this letter, we developed a method to characterize tumors by representing multiple tactile images obtained from a tactile sensing system. We convert multiple tactile images into a single representative image called the tactile profile diagram (TPD). TPD is a pattern image with tactile information about a tumor. TPD is used to estimate the mechanical properties of the tumors. We verified the TPD method by classifying tumor phantoms using a convolutional neural network (CNN) with an accuracy of 83% for depth, 78% for stiffness, and 72% for size. We also applied machine learning algorithms for tumor mechanical property classification with the human dataset. CNN classification accuracy of depth, stiffness, and size was 77%, 69%, and 69%, respectively, for the human data. Consequently, we conclude that the TPD method is useful for inclusion classification with the tactile sensing system in breast cancer applications.