A New Tooth Semantic Segmentation Algorithm of Orthodontics

Abstract

AbstractThe acquisition of three-dimensional tooth model and the accurate segmentation of tooth boundary are of great significance for orthodontics and dental implant diagnosis and subsequent treatment plan. In order to realize the accurate segmentation of a single tooth, an automatic tooth segmentation algorithm proposed. Based on the three-dimensional tooth model, the accurate semantic segmentation of a single tooth realized through the deep learning network of the local fine and global coarse structure of the tooth model. This paper proposes a Rows and Columns Aggregated Network (RCANet) for segmentation. The framework based on graph convolutional networks, which mainly includes two parts, one is the example segmentation network, which used to obtain the general shape and relative position information of teeth. The second is the fine-grained segmentation network that used to learn the fine details of a single tooth, and add a punishment mechanism for the wrong label that further improves the accuracy of tooth segmentation. The results show that the framework can achieve accurate 3D tooth segmentation. The realization of tooth segmentation is of great significance for further clinical application.KeywordsTooth semantic segmentationGraph convolutional networksDeep learningRCANet