Refined Individual Tooth Segmentation from Cone Beam CT Images

Tuo Zheng, Jiaxiang Liu, Yang Feng, Junhui Lv, Zuozhu Liu

Abstract: In the field of dental imaging, precise segmentation of individual teeth from 3D Cone Beam Computed Tomography (CBCT) images remains a challenging task. This paper proposes a novel coarse-to-fine segmentation framework specifically optimized for individual tooth segmentation in CBCT imagery. At its core, in the coarse stage, the framework employs an embedding grouping network that ensures accurate localization and establishes a strong tooth-specific prior. To address the intricacies of tooth morphology, in the fine stage, our architecture designs a Tripartite Fusion Module and a Decouple Body and Edge Module. These modules work in concert to refine the segmentation process, thereby improving the granularity of individual tooth delineation. We have tested our framework on a real-world dataset, and the results demonstrate improvement over existing methods.