Shape-preserving Tooth Segmentation from CBCT Images Using Deep Learning with Semantic and Shape Awareness

TL;DR

This work tackles automated tooth segmentation in CBCT images, focusing on preserving anatomical shapes in the presence of interdental adhesions that distort boundaries. It introduces a three-module pipeline that combines semantic awareness via target-tooth-centroid prompting, a centroid-based localization, and shape-aware learning using signed distance maps within a shared encoder multitask framework. The approach yields state-of-the-art results on internal and external datasets, with ablation studies confirming the contributions of centroid prompting, multi-label relations, and shape priors to boundary fidelity and overall segmentation accuracy. Practically, this method offers more accurate, shape-faithful tooth delineation suitable for clinical workflows in implant planning and orthodontics, while highlighting avenues for broader validation and application.

Abstract

Background:Accurate tooth segmentation from cone beam computed tomography (CBCT) images is crucial for digital dentistry but remains challenging in cases of interdental adhesions, which cause severe anatomical shape distortion. Methods: To address this, we propose a deep learning framework that integrates semantic and shape awareness for shape-preserving segmentation. Our method introduces a target-tooth-centroid prompted multi-label learning strategy to model semantic relationships between teeth, reducing shape ambiguity. Additionally, a tooth-shape-aware learning mechanism explicitly enforces morphological constraints to preserve boundary integrity. These components are unified via multi-task learning, jointly optimizing segmentation and shape preservation. Results: Extensive evaluations on internal and external datasets demonstrate that our approach significantly outperforms existing methods. Conclusions: Our approach effectively mitigates shape distortions and providing anatomically faithful tooth boundaries.

Figures (4)

• Figure 1: The architecture of the proposed tooth segmentation network. (A) The tooth region-of-interest (ROI) detection module. (B) The fast clustering based tooth centroid localization module. (C) the multi-task multi-label shape-aware segmentation module.
• Figure 2: Composition of internal and external datasets for method development and evaluation.
• Figure 3: Visual comparison of tooth segmentation results using three different methods. Three typical examples are shown, each example displayed in both 2D (A) and 3D (B) views.
• Figure 4: The qualitative comparison of tooth segmentation results in the ablation study includes the following networks: the baseline B-Net; C-Net, which incorporates the tooth centroid prompt into B-Net; CM-Net, formed by adding multi-label learning to C-Net; and the complete network (CMS-Net), integrated with shape-aware learning. Three representative examples are presented, each visualized in both 2D (A) and 3D (B) views.