3D Tooth Segmentation and Labeling Using Deep Convolutional Neural Networks
3D Tooth Segmentation and Labeling Using Deep Convolutional Neural Networks
Authors: Xiaojie Xu, Chang Liu, and Youyi Zheng
Summary
This study introduces a fully automated deep learning-based framework for segmenting and labeling 3D dental meshes. The authors developed a two-level hierarchical convolutional neural network (CNN) to perform tooth-gingiva and inter-tooth labeling, achieving an accuracy of 99.06% for upper dental models and 98.79% for lower dental models. The approach integrates a boundary-aware mesh simplification method, graph-based label optimization, and an improved fuzzy clustering refinement to preserve tooth boundaries while maintaining computational efficiency. The network was trained on 1,000 manually labeled dental meshes and validated on 150 test cases, demonstrating robustness to noise, missing teeth, and dental accessories. The resulting framework provides an efficient and accurate alternative to traditional geometry-based methods and is directly applicable to orthodontic CAD systems.
DOI: https://doi.org/10.1109/TVCG.2018.2839685
Key Words
AI in Dentistry, Deep Learning, 3D CNN, Dental Mesh Segmentation, Label Optimization, Boundary-aware Simplification, Orthodontic CAD
Extracted Data
- Year: 2018
- Modality: 3D dental mesh (STL models)
- Dataset: 1,000
- Dataset Split: Train = 1,000 / Validation = 50 / Test = 150
- Network Architecture: Two-level hierarchical CNN (TGCNN + TTCNN)
- Metrics: Accuracy (area-based) = 99.06% (upper), 98.79% (lower); mean boundary error = 0.0848 mm
- AP – Strategy: Not reported
- AP – Professional Qty: Multiple experts annotated dental meshes
- AP – Supervisor Presence: No information
- AP – Experience Level: No information
- AP – Expertise Area: Orthodontics
- AP – Tool or System: Custom mesh annotation system (proprietary)
- Task: Segmentation
- Project Objective: To develop a robust, automated method for precise segmentation and labeling of teeth and gingiva
Clinical Relevance
- Scientific impact: This work marked a milestone in the use of deep learning for 3D dental applications, overcoming the limitations of geometry-based segmentation through a hierarchical CNN design.
- Technical innovation: The combination of boundary-aware simplification and fuzzy boundary refinement achieved high accuracy while maintaining efficiency.
- Clinical application: This framework supports automatic tooth labeling for orthodontic CAD workflows and 3D treatment planning, illustrating how AI enhances digital orthodontics.
Related Video
▶️ Watch on YouTube
