3D Tooth Segmentation and Labeling Using Deep Convolutional Neural Networks.

Abstract in brief

• The paper introduces a new method for 3D dental model segmentation using deep convolutional neural networks (CNNs), addressing limitations of traditional geometry-based techniques that struggle with complex and damaged teeth and cannot easily label individual teeth.
• The approach involves extracting geometric features from each face of the 3D mesh, then training a hierarchical CNN to classify these faces into different dental parts, such as teeth and gums, with additional steps like graph-based label optimization and boundary refinement to boost accuracy.
• The proposed method achieves a high accuracy of over 99%, is more efficient thanks to a boundary-aware tooth simplification process, and is robust against surface artifacts like air bubbles or dental accessories, making it suitable for orthodontic CAD applications.]