Automatic Path Planning for Navigated Pedicle Screw Surgery Based on Deep Neural Network

摘要

In this paper, we aim to achieve intelligently automatic planning for navigated pedicle screw surgery. To this end, deep neural network is introduced into the procedure of path planning. First, the surgical path (line) is represented as two control points: the entry point and the direction indication point. Then, features on CT image of spine are extracted utilizing a designed multi-layer 3D convolutional neural network, a 3D U-Net-liked segmentation network is designed to rebuild vertebral body (VB) of spine and a fully connection network is connected to learn two control points on the segmented VB. Extensive experimental results show that our network can perform a pixel-wise multi-class segmentation and quickly locate the landmarks on surgical path in the navigated spine surgery with accepted accuracy. This work provides a reference for robot-assisted surgical path planning.