关键词: 颅缝早闭, 轨迹提取, 深度学习, 深度相机, 点云处理

Abstract: To standardize the plan generation of craniosynostosis surgery, we propose a skull cutting trajectory generation method which combines deep learning, stereo vision and point cloud processing technology to establish cutting plan template library and generate cutting trajectory for new cases. The proposed method,for the first time,uses the deep learning to segment the external surface of skull. First, it takes the advantage of Mask R-CNN to detect and segment the surgical cutting area. Then, a simplified contour extraction algorithm is explored to extract the surgical cutting trajectory. After that,the point cloud processing technology is used to map the surgical cutting trajectory to three-dimensional coordinates and realize the automatic extraction of cutting trajectory. Finally,a template library with typical cases is established and the cutting plans of new cases are automatically generated by template matching method. The experiment shows that the proposed trajectory extraction method can detect the skull surgical cutting trajectory accurately and efficiently and conduct three-dimensional mapping of trajectory coordinate.The depth measurement error of point cloud is less than 3 mm, which meets the clinical available standard. Using the template matching method,the new case's cutting trajectory can be generated effectively, which conforms to surgical plan of senior doctors.

Key words: craniosynostosis, trajectory extraction, deep learning, depth camera, point cloud processing