Teeth-GS: Gaussian Splatting Diffusion with enamel reflectance prior for single-image tooth crown reconstruction

Area of Science:

• Computer Vision
• Digital Dentistry
• Computational Imaging

Background:

• High-fidelity 3D tooth crown reconstruction is crucial for digital dentistry.
• Intra-oral scanners have limitations in cost, hardware, and capturing fine occlusal details under clinical lighting.
• Current single RGB image methods lack physical grounding, hindering accurate surface topography recovery.

Purpose of the Study:

• To develop a cost-effective, high-precision 3D tooth crown reconstruction method from single RGB images.
• To improve the recovery of fine-scale surface topographies, including occlusal fissures and pits.
• To enable robust surface normal estimation by integrating physical principles of light-enamel interaction.

Main Methods:

• Proposed Teeth-GS: a Gaussian Splatting Diffusion framework incorporating an anisotropic reflectance prior inspired by the Huygens-Fresnel principle.
• Utilized a Gaussian Markov Random Field (GMRF) guided diffusion network for physically consistent enamel scattering properties.
• Implemented a reflectance-guided 3D Gaussian Splatting module for converting predicted normals into explicit Gaussian primitives with hybrid illumination mapping.

Main Results:

• Achieved robust surface normal estimation even in challenging regions with strong specularities and textureless surfaces.
• Generated compact and anatomically consistent 3D tooth crown representations decoupling geometry from appearance.
• Demonstrated significant outperformance over state-of-the-art methods on geometric and photometric benchmarks using intra-oral scan and clinical RGB datasets.

Conclusions:

• Teeth-GS offers a viable, high-precision solution for 3D tooth crown reconstruction from single RGB images.
• The physics-informed approach enhances the recovery of intricate dental micro-geometry.
• The method shows promise for applications in remote dental screening and digital monitoring.