Depth Recognition of Hard Inclusions in Tissue Phantoms for Robotic Palpation

Abstract

Medical palpation is an effective diagnosis method in which physicians use tactile sensation to diagnose a patient’s pathology. Robotic palpation is a novel technique that leverages robots to assist medical diagnosis. The problem of tactile information loss in Robot-assisted Minimally Invasive Surgery (RMIS) has limited the development of the robot-assisted surgical system. Meanwhile, surgeons are difficult to acquire some key information about lesions only via visual feedback, such as tumor depth. To address the issue, we propose a tactile perception algorithm on the basis of the CNN-LSTM network, which achieves the depth recognition of hard inclusions in tissue phantoms. The method realizes the classification of twelve depths of hard inclusions. In addition, due to using hypergeometric distribution encoding, the proposed method can exploit the ordinal information of the labels to significantly improve the recognition accuracy rate. The experimental results on 720 real tactile data show that the average recognition rate is 96.45%. Compared with other state-of-the-art methods, the recognition accuracy rate of the proposed algorithm is the highest.