Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

High-Fidelity 3D Tooth Reconstruction by Fusing Intraoral Scans and CBCT Data via a Deep Implicit Representation

Yi Zhu, Razmig Kechichian, Raphaël Richert, Satoshi Ikehata, Sébastien Valette

TL;DR

The paper tackles the challenge of building complete, high-fidelity 3D tooth models by fusing the crown detail from Intraoral Scans (IOS) with root morphology from Cone-Beam Computed Tomography (CBCT). It introduces an automated pipeline that creates a hybrid proxy by aligning and merging IOS crowns with CBCT roots, then refines this proxy using a class-specific DeepSDF to project the result onto a learned manifold of plausible tooth shapes. Key contributions include robust two-stage registration, a principled hybrid proxy construction with a root-extraction threshold, and implicit refinement with a DeepSDF trained on ToothFairy3 to generate seamless, watertight surfaces. Quantitative results show that Fused-SDF reduces crown error (CD $0.082$ mm) and maintains root accuracy (HD95 $0.243$ mm) compared to CBCT-only baselines, demonstrating improved accuracy and clinical viability for digital dentistry applications. Overall, the approach enables accurate, patient-specific, complete 3D tooth models suitable for restoration design, surgical planning, and biomechanical simulation.

Abstract

High-fidelity 3D tooth models are essential for digital dentistry, but must capture both the detailed crown and the complete root. Clinical imaging modalities are limited: Cone-Beam Computed Tomography (CBCT) captures the root but has a noisy, low-resolution crown, while Intraoral Scanners (IOS) provide a high-fidelity crown but no root information. A naive fusion of these sources results in unnatural seams and artifacts. We propose a novel, fully-automated pipeline that fuses CBCT and IOS data using a deep implicit representation. Our method first segments and robustly registers the tooth instances, then creates a hybrid proxy mesh combining the IOS crown and the CBCT root. The core of our approach is to use this noisy proxy to guide a class-specific DeepSDF network. This optimization process projects the input onto a learned manifold of ideal tooth shapes, generating a seamless, watertight, and anatomically coherent model. Qualitative and quantitative evaluations show our method uniquely preserves both the high-fidelity crown from IOS and the patient-specific root morphology from CBCT, overcoming the limitations of each modality and naive stitching.

High-Fidelity 3D Tooth Reconstruction by Fusing Intraoral Scans and CBCT Data via a Deep Implicit Representation

TL;DR

The paper tackles the challenge of building complete, high-fidelity 3D tooth models by fusing the crown detail from Intraoral Scans (IOS) with root morphology from Cone-Beam Computed Tomography (CBCT). It introduces an automated pipeline that creates a hybrid proxy by aligning and merging IOS crowns with CBCT roots, then refines this proxy using a class-specific DeepSDF to project the result onto a learned manifold of plausible tooth shapes. Key contributions include robust two-stage registration, a principled hybrid proxy construction with a root-extraction threshold, and implicit refinement with a DeepSDF trained on ToothFairy3 to generate seamless, watertight surfaces. Quantitative results show that Fused-SDF reduces crown error (CD mm) and maintains root accuracy (HD95 mm) compared to CBCT-only baselines, demonstrating improved accuracy and clinical viability for digital dentistry applications. Overall, the approach enables accurate, patient-specific, complete 3D tooth models suitable for restoration design, surgical planning, and biomechanical simulation.

Abstract

High-fidelity 3D tooth models are essential for digital dentistry, but must capture both the detailed crown and the complete root. Clinical imaging modalities are limited: Cone-Beam Computed Tomography (CBCT) captures the root but has a noisy, low-resolution crown, while Intraoral Scanners (IOS) provide a high-fidelity crown but no root information. A naive fusion of these sources results in unnatural seams and artifacts. We propose a novel, fully-automated pipeline that fuses CBCT and IOS data using a deep implicit representation. Our method first segments and robustly registers the tooth instances, then creates a hybrid proxy mesh combining the IOS crown and the CBCT root. The core of our approach is to use this noisy proxy to guide a class-specific DeepSDF network. This optimization process projects the input onto a learned manifold of ideal tooth shapes, generating a seamless, watertight, and anatomically coherent model. Qualitative and quantitative evaluations show our method uniquely preserves both the high-fidelity crown from IOS and the patient-specific root morphology from CBCT, overcoming the limitations of each modality and naive stitching.
Paper Structure (9 sections, 4 figures, 1 table)

This paper contains 9 sections, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Method
  3. Data Preparation and Instance Extraction
  4. Robust Two-Stage Mesh Registration and Fusion
  5. Implicit Refinement with a Class-Specific DeepSDF
  6. Experiments and Results
  7. Qualitative Comparison
  8. Quantitative Analysis
  9. Conclusion

Figures (4)

  • Figure 1: Illustration of data. (a) Incisors segmented from a CBCT scan, showing complete root anatomy. (b) Full tooth mesh obtained by extracting segmented CBCT surface. (c) IOS mesh, with relatively more precise crown details, but no root information.
  • Figure 2: A naive method for fusing a lower-resolution CBCT root (a, grey) with a high-resolution IOS crown (b, orange) via direct stitching (d). (c) shows the alignment of the two parts. The final model (d) exhibits an unnatural seam, geometric discontinuities, and surface noise at the junction.
  • Figure 3: Two-column pipeline with an additional segmentation stage. Thumbnails to the right of each box show representative outputs. Abbreviations: $R$ = CBCT segmentation; $C$ = IOS segmentation; $T$ = two-step registration (RANSAC followed by ICP) aligning IOS to CBCT; $H$ = hybrid proxy from naive fusion of crown $C$ with root region $R_{\text{root}}$; $S$ = final implicit surface reconstructed by DeepSDF.
  • Figure 4: Qualitative comparison and error maps. Colors represent one-sided surface distance from the Naive-Fusion mesh: white = low error, red = high error. (a) Naive-Fusion mesh. (b) CBCT-SDF. (c) Fused-SDF. (d) Error map (CBCT-SDF). (e) Error map (Fused-SDF). (f) Legend.