Related Experiment Video
Updated: Jan 8, 2026
Digital Hybrid Model Preparation for Virtual Planning of Reconstructive Dentoalveolar Surgical Procedures
Published on: August 5, 2021
Automated reconstruction of missing tooth morphology using a transformer-based implicit neural network: A multi-tooth
Yiqing Wang1, Yuze Shi2, Nan Li2
1Department of Prosthodontics, Peking University School and Hospital of Stomatology, National Center of Stomatology, National Clinical Research Center for Oral Diseases, National Engineering Research Center of Oral Biomaterials and Digital Medical Devices, Beijing Key Laboratory of Digital Stomatology, Research Center of Engineering and Technology for Computerized Dentistry, Ministry of Health, NMPA Key Laboratory for Dental Materials, Beijing, PR China.
Purpose:
This study aimed to develop a novel transformer-based model for automated tooth morphology reconstruction and evaluate its accuracy and generalizability across multiple tooth positions.
Materials And Methods:
Digital full-arch casts with intact target and adjacent teeth were collected, comprising 500 first molars, 600 first premolars, and 700 central incisors after data augmentation. A transformer-based implicit neural network (INN) model was developed by incorporating a self-structure enhancement module and multi-view 2D depth maps. The model was trained with either 12,000 or 50,000 sampling points. Performance was assessed using chamfer distance (CD), F-score, and volumetric intersection over union (IoU). Reconstructed generated crowns (GC) were compared with original crowns (OC) and technician-designed crowns (TC) in terms of 3D morphological deviations, measured by the root mean square (RMS, mm), and dimensional differences. Statistical analysis was performed using a linear mixed-effects model, repeated measures ANOVA or nonparametric tests (α = 0.05).
Results:
The model trained with 50,000 sampling points exhibited superior reconstruction performance, with high similarity to natural tooth morphology. Central incisors showed the best accuracy (CD = 0.0028 × 10-2, F-score = 0.9670, and IoU = 0.9716). In molars, GC presented comparable surface deviations to TC, with no significant difference. For premolars, GC exhibited higher deviations compared to TC (0.2255 ± 0.0285 mm vs. 0.1557 ± 0.0422 mm, p = 0.002). Similarly, in the incisor, GC exhibited higher deviations compared to TC (0.2155 ± 0.0272 mm vs. 0.1643 ± 0.0295 mm, p = 0.014). In dimensional analysis, GC achieved a close match to OC across all tooth types (p > 0.05). At the same time, TC showed significantly greater mesiodistal width in molars and inciso-gingival height in incisors.
Conclusions:
The proposed transformer-based model effectively achieved automated reconstruction of missing single-tooth morphology with acceptable accuracy and adaptability across different tooth positions. Its high fidelity and dimensional consistency highlight its potential for improving efficiency in digital dental restoration workflows. Further studies are warranted to expand dataset diversity, refine the model architecture, and incorporate clinical and functional validations.
