This paper establishes a method to realize semi-automatic or automatic labeling of multi-dimensional data based on small samples and weak labeling. This method could effectively assist dentist in segmentation of different tissues. Based on the U-net combined with the Generative Adversarial Networks method, segmentation can be realized on multi-dimensional data. It also includes three-dimensional mesh reconstruction of the segmented tissue, smooth the boundary, and the result data can be used to aid clinical diagnosis and print. The result of segmentation can reflect the structural distribution of different tissues, and effectively build a mechanical model based on Cone-beam computed tomography systems (CBCT) datasets.

segmentation,Generative Adversarial Networks,annotation,CBCT