Related Experiment Video
Updated: Mar 15, 2026
10:23
Author Spotlight: Three-Dimensional Cephalometric Landmark Annotation Demonstration on Human Cone Beam Computed Tomography Scans
Published on: September 8, 2023
3.8K
3D exemplar-based random walks for tooth segmentation from cone-beam computed tomography images
Yuru Pei1, Xingsheng Ai1, Hongbin Zha1
1Department of Machine Intelligence, School of EECS, Peking University, Beijing 100871, China.
Medical Physics
|September 3, 2016
Summary
This study introduces a novel semisupervised method for segmenting teeth in cone-beam computed tomography (CBCT) images, achieving high accuracy and efficiency for dental applications.
Area of Science:
- Medical Imaging
- Computer-Aided Diagnosis
- Dental Technology
Background:
- Tooth segmentation from CBCT images is crucial for patient-specific dental geometries.
- Challenges include low image quality and structural ambiguities.
- Existing methods struggle with accuracy and efficiency.
Purpose of the Study:
- To present a semisupervised tooth segmentation method using adaptive 3D shape constraints.
- To improve the accuracy and reliability of tooth segmentation from CBCT data.
Main Methods:
- A 3D exemplar-based random walk method integrating semisupervised label propagation and regularization.
- Iterative refinement using 3D exemplar registration and soft constraints.
- Combines shape-based and appearance-based probabilities for voxel labeling.
Main Results:
- Achieved high accuracy in segmenting anterior teeth (DSC up to 98%), premolars (DSC 98%), and molars (DSC 95%).
- Demonstrated low mean surface deviation (MSD) for all tooth types.
- Automatic segmentation completed in an average of 1.18 minutes.
Conclusions:
- The proposed technique offers efficient and reliable tooth segmentation from CBCT images.
- Enables practical clinical application for maxillofacial and orthodontic treatments.
- Facilitates pre- and interoperative use of dental morphologies.
