Learning-based catheter and guidewire-driven autonomous vascular intervention robotic system for reduced repulsive force

Abstract

Manual vascular interventional radiology (VIR) procedures have been performed under radiation exposure conditions, and many commercial master-slave VIR robot systems have recently been developed to overcome this issue. However, master-slave VIR robot systems still have limitations. The operator must reside near the master device and control the slave robot using only the master device. In addition, the operator must simultaneously process the recognition of the surgical tool from the X-ray image while operating the master device. To overcome the limitations of master-slave VIR robot systems, we propose an autonomous VIR robot system with a deep learning algorithm that excludes the master device. The proposed autonomous VIR robot with a deep learning algorithm drives surgical tools to the target blood vessel location while simultaneously performing surgical tool recognition. The proposed autonomous VIR robot system detects the location of the surgical tool based on a supervised learning algorithm, and controls the surgical tools based on a reinforcement-learning algorithm. Experiments are conducted using two types of vascular phantoms to verify the effectiveness of the proposed autonomous VIR robot system. The experimental results of the vascular phantom show a comparison between the master-slave VIR robot system and the proposed autonomous VIR robot system in terms of the repulsive force, task completion time, and success rate during the operation. The proposed autonomous VIR robot system is shown to exhibit a significant reduction in repulsive force and a 96% success ratio based on a vascular phantom.