3D Single-Tooth Reconstruction via Curriculum Learning and Topology-aware Generation

Area of Science:

• Digital Dentistry
• Computer Graphics
• Medical Imaging

Background:

• Automated single-tooth reconstruction is crucial for digital dentistry.
• Current point cloud completion methods lack anatomical accuracy and spatial constraints for clinical use.

Purpose of the Study:

• To propose Tooth-STR, a hierarchical generative framework for synthesizing high-fidelity, collision-free tooth models from partial intraoral scans.
• To address limitations in existing methods by incorporating specialized anatomical priors and spatial constraints.

Main Methods:

• Tooth-STR employs a Topology-Aware Coarse Generation (TACG) module using differential edge convolutions.
• A Curriculum Anatomical Constraint (C^2) strategy with phased, differentiable boundary penalties ensures spatial regularization.
• The framework balances geometric exploration with strict spatial constraints.

Main Results:

• Tooth-STR significantly improves reconstruction performance across diverse tooth categories.
• The method generates anatomically high-fidelity and collision-free tooth models.
• Experiments were conducted on a newly constructed Tooth-STR dataset.

Conclusions:

• Tooth-STR offers a robust solution for automated single-tooth reconstruction in digital dentistry.
• The framework's hierarchical and constraint-based approach enhances clinical applicability.
• This method advances the synthesis of accurate dental models from incomplete scan data.