Precise tooth design using deep learning-based templates.

Area of Science:

• Biomedical Engineering
• Computer Science
• Dental Materials

Background:

• Traditional computer-aided design (CAD) in prosthodontics faces challenges with time consumption and accuracy in restoring tooth shape.
• Accurate morphological restoration is crucial for functional and aesthetic outcomes in dental prostheses.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for precise, personalized tooth defect restoration in prosthodontic procedures.
• To assess the accuracy of a novel DL-based approach compared to existing methods for morphological restoration.

Main Methods:

• A DL model, ToothDIT, was trained on 80 maxillary incisor models to create an implicit template and predictive neural network.
• Defective teeth were processed, and ToothDIT generated customized templates (DIT) by deforming the implicit template.
• Restorative accuracy was evaluated using metrics like RMS_estimate, width, length, and aspect ratio, comparing DIT with template shape library (TSL) and average tooth template (ATT).

Main Results:

• The DIT approach demonstrated significantly lower RMS_estimate compared to TSL and ATT.
• 2D analysis revealed that DIT exhibited significantly less deviation from natural tooth morphology than TSL and ATT.
• The DL model successfully reconstructed anterior teeth with varying defect degrees, achieving high accuracy.

Conclusions:

• The proposed DL model offers a precise and reliable method for reconstructing defective tooth morphology.
• This approach enhances the accuracy of prostheses design and improves the efficiency of digital restorative workflows.
• Customized templates generated by the DL model can significantly aid dentists and technicians in achieving natural-looking restorations.