A Deep Learning‐Based Human–Robot Collaborative Navigation Framework for Vascular Interventional Surgery

BACKGROUND: Autonomous robotic surgery has demonstrated its potential for the optimal outcomes. However, vascular interventional surgery (VIS) with flexible catheters and soft tissue raises challenges to autonomous execution of preplanned tasks due to indeterminately changed surgical state. METHODS: We present a novel end-to-end deep learning-based framework of human-robot collaborative navigation for VIS. A surgical Generative Adversarial Networks is employed for real-time local path planning of the catheter tip under variable-vascular-contour environment. A CNNs-based action estimator is proposed for nonlinear mapping from the tip's path to the end's action. A human-robot trust-based shared control model is established for surgical navigation. RESULTS: The networks are trained by a self-built dataset and experiments are conducted under catheterization room environment. The results show the catheter's action decision accuracy achieves 93.75%. The surgical effectiveness and safety are improved with the proposed method. CONCLUSION: This work provides a way to achieve autonomous VIS.