Automated reconstruction of missing tooth morphology using a transformer-based implicit neural network: A multi-tooth...

Area of Science:

• Artificial Intelligence in Dentistry
• Computational Dental Anatomy
• Digital Dentistry

Background:

• Automated tooth morphology reconstruction is crucial for digital dental restorations.
• Existing methods may lack accuracy or generalizability across diverse tooth types.

Purpose of the Study:

• Develop a novel transformer-based implicit neural network (INN) model for automated tooth morphology reconstruction.
• Evaluate the model's accuracy and generalizability across different tooth positions (molars, premolars, incisors).

Main Methods:

• Utilized digital full-arch casts of molars, premolars, and incisors.
• Developed a transformer-based INN model with a self-structure enhancement module and multi-view 2D depth maps.
• Assessed performance using chamfer distance, F-score, and volumetric intersection over union; compared reconstructed crowns (GC) with original (OC) and technician-designed (TC) crowns using RMS and dimensional analysis.

Main Results:

• The model trained with 50,000 sampling points demonstrated superior reconstruction accuracy, closely matching natural tooth morphology.
• Central incisors achieved the highest accuracy (CD = 0.0028 × 10^-2, F-score = 0.9670, IoU = 0.9716).
• Reconstructed crowns showed comparable deviations to technician-designed crowns in molars, but higher deviations in premolars and incisors; dimensional analysis revealed close matches to original crowns across all tooth types.

Conclusions:

• The transformer-based model effectively reconstructs missing single-tooth morphology with acceptable accuracy and adaptability.
• The model's high fidelity and dimensional consistency show potential for improving digital dental restoration workflows.
• Further research should focus on expanding datasets, refining the model, and incorporating clinical validations.