Original title: TSegFormer: 3D Tooth Segmentation in Intraoral Scans with Geometry Guided Transformer
Authors: Huimin Xiong, Kunle Li, Kaiyuan Tan, Yang Feng, Joey Tianyi Zhou, Jin Hao, Haochao Ying, Jian Wu, Zuozhu Liu
In digital dentistry, Intraoral Scanners (IOS) offer detailed 3D dental information, but accurate tooth segmentation remains a challenge. Existing methods struggle at complex boundaries and vary in effectiveness across patients. Enter TSegFormer: a solution proposed in this article. It employs a multi-task 3D transformer architecture, capturing local and global dependencies among teeth and the gingiva in IOS point clouds. To refine boundaries seamlessly, they introduce a geometry-guided loss based on novel point curvature, eliminating the need for time-consuming post-processing. Additionally, they assemble an extensive dataset of 16,000 IOSs, the largest known. Results showcase TSegFormer’s consistent outperformance of existing methods, validated through thorough analysis, visualizations, and real-world clinical tests. TSegFormer not only advances accuracy but also streamlines clinical applicability. The article makes their code available, providing a valuable resource for further research.
Original article: https://arxiv.org/abs/2311.13234