Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Multi-Stage Framework for 3D Individual Tooth Segmentation in Dental CBCT

Chunshi Wang, Bin Zhao, Shuxue Ding

TL;DR

The paper tackles the challenge of accurate 3D tooth segmentation from CBCT under data scarcity and cross-device domain shifts. It introduces a multi-stage SSL framework that first leverages a 2D nnU-Net to generate pseudo-labels and then applies domain-adaptive training with Improved-UniMatch aided by Fourier Transform Augment, along with self-adaptive thresholding and Alpha Dropout. On STS-3D data, the method ranks third and achieves a mean Dice of 0.8801, IoU of 0.8425, and a composite score of 0.8859 on validation, outperforming several semi-supervised baselines. The approach reduces labeling requirements and improves cross-domain robustness, with potential to streamline clinical workflows in orthodontics and dental implant planning.

Abstract

Cone beam computed tomography (CBCT) is a common way of diagnosing dental related diseases. Accurate segmentation of 3D tooth is of importance for the treatment. Although deep learning based methods have achieved convincing results in medical image processing, they need a large of annotated data for network training, making it very time-consuming in data collection and annotation. Besides, domain shift widely existing in the distribution of data acquired by different devices impacts severely the model generalization. To resolve the problem, we propose a multi-stage framework for 3D tooth segmentation in dental CBCT, which achieves the third place in the "Semi-supervised Teeth Segmentation" 3D (STS-3D) challenge. The experiments on validation set compared with other semi-supervised segmentation methods further indicate the validity of our approach.

A Multi-Stage Framework for 3D Individual Tooth Segmentation in Dental CBCT

TL;DR

The paper tackles the challenge of accurate 3D tooth segmentation from CBCT under data scarcity and cross-device domain shifts. It introduces a multi-stage SSL framework that first leverages a 2D nnU-Net to generate pseudo-labels and then applies domain-adaptive training with Improved-UniMatch aided by Fourier Transform Augment, along with self-adaptive thresholding and Alpha Dropout. On STS-3D data, the method ranks third and achieves a mean Dice of 0.8801, IoU of 0.8425, and a composite score of 0.8859 on validation, outperforming several semi-supervised baselines. The approach reduces labeling requirements and improves cross-domain robustness, with potential to streamline clinical workflows in orthodontics and dental implant planning.

Abstract

Cone beam computed tomography (CBCT) is a common way of diagnosing dental related diseases. Accurate segmentation of 3D tooth is of importance for the treatment. Although deep learning based methods have achieved convincing results in medical image processing, they need a large of annotated data for network training, making it very time-consuming in data collection and annotation. Besides, domain shift widely existing in the distribution of data acquired by different devices impacts severely the model generalization. To resolve the problem, we propose a multi-stage framework for 3D tooth segmentation in dental CBCT, which achieves the third place in the "Semi-supervised Teeth Segmentation" 3D (STS-3D) challenge. The experiments on validation set compared with other semi-supervised segmentation methods further indicate the validity of our approach.
Paper Structure (11 sections, 8 equations, 4 figures, 3 tables)

This paper contains 11 sections, 8 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Materials and Method
  3. Research subjects
  4. Multi-stage framework
  5. Evaluation metrics
  6. Experiment
  7. Data preprocessing
  8. Implementation Details
  9. Results
  10. Conclusion
  11. Authors’ Contributions

Figures (4)

  • Figure 1: The architecture of our proposed multi-stage framework. Best viewed in color.
  • Figure 2: The flowchart of FTA.
  • Figure 3: Six examples showing segmentation results obtained from our method. The four rows from top to bottom denotes axial, coronal, sagittal and 3D views. Best viewed in color.
  • Figure 4: Comparison of segmentation performance. (a-b) CBCT slices and their corresponding ground truth. (c-f) are the segmentation results of CPS chen2021semi, FixMatch sohn2020fixmatch, MeanTeacher tarvainen2017mean and our proposed method, respectively.