A Deep Learning Network For Point Cloud Of Medicine Structure

Great progress has been made in 3D reconstruction of medicine structures through medical images such as CT and Mill in recent years. With the demand for virtual surgery and other technologies, it has become an important research topic that how to segment the reconstructed 3D model. 3D models are generally stored in CAD or MAX formats and need to be transformed into point clouds before they can be segmented. Our work proposed a new deep learning network structure, which can directly process irregular point cloud. Empirically, it showed competitive performance in many 3D tasks such as object classification and semantic segmentation. The trained network was applied to the reconstructed 3D models of medicine structures, and good results were also obtained.