Machine (Deep) Learning for Orthodontic CAD/CAM Technologies

Accurate tooth segmentation on a 3D dental mesh model is a vital task in computer-aided design and computer-aided manufacturing (CAD/CAM) technologies for orthodontic treatment planning. This task is especially challenging due to the complexity and variation of human teeth. Recently, several pioneering deep neural networks (e.g., PointNet) have been proposed in the computer vision and computer graphics communities to efficiently segment 3D objects in an end-to-end manner. However, these methods do not perform well and produce undesirable results in the specific task of tooth labeling because they cannot explicitly model the fine-grained local geometric context of teeth. In this chapter, we comprehensively review a recently-proposed deep neural network, MeshSegNet or MeshSNet, which is designed for end-to-end tooth segmentation on 3D dental surface meshes captured from intraoral scanners. The entire approach from preprocessing to postprocessing for the task-automated tooth segmentation is discussed in detail. The preprocessing procedures of the MeshSegNet include mesh simplification and data augmentation. The results from MeshSegNet are then refined in two stages during the post-processing phase. Each of these steps is detailed in this chapter, along with a comparison of segmentation results between the original PointNet and MeshSegNet, which highlights the superior performance of MeshSegNet.