Machine Learning-Based Surgical State Perception and Collaborative Control for a Vascular Interventional Robot

Abstract

In robot-assisted vascular interventional surg- ery (VIS), surgeons often need to operate outside the operating room to avoid exposure to X-ray. However, it greatly changes the operating ways of surgeons, which affects judgment and operation safety. In this paper, a novel VIS robot system was developed to predict guidewire insertion states and operate collaboratively. To assist the surgeons in perceiving the insertion state, an insertion multi-states prediction model based on softmax logistic regression was proposed. Combined with the prediction model, a human-machine collaborative control strategy was designed, which allows surgeons to perceive the insertion states based on not only the force feedback constructed by the master side but also the prediction results from the slave side. Moreover, a human-machine trust evaluation model and a master-slave collaborative mapping model were proposed for improving safety and efficiency of surgery. To verify the effectiveness of these models, the evaluation experiments in the blood vessel model were carried out. It was indicated by the experiment results that the guidewire insertion states can be predicted by the prediction model in different environments, and the overall accuracy is 93%. The master-slave mapping ratio can be adjusted by the collaborative control strategy automatically to adapt to different surgical conditions. The experimental results showed the usability of the robot-assisted VIS system with the novel force-based perception method.