Related Experiment Video
Updated: May 21, 2025
Author Spotlight: Three-Dimensional Cephalometric Landmark Annotation Demonstration on Human Cone Beam Computed Tomography Scans
Published on: September 8, 2023
Classification of skeletal discrepancies by machine learning based on three-dimensional facial scans
1Department of Orthodontics, Peking University School and Hospital of Stomatology, Beijing, China; National Centre for Stomatology, Beijing, China; National Clinical Research Centre for Oral Diseases, Beijing, China; National Engineering Research Centre of Oral Biomaterials and Digital Medical Devices, Beijing, China; Beijing Key Laboratory of Digital Stomatology, Beijing, China; Research Centre of Engineering and Technology for Computerized Dentistry Ministry of Health, Beijing, China.
Machine learning models accurately classify sagittal and vertical skeletal discrepancies using 3D facial scans. These advanced methods show potential for aiding orthodontic diagnosis by analyzing facial morphology.
Area of Science:
- Orthodontics and Dentofacial Orthopedics
- Medical Imaging and Computer-Aided Diagnosis
- Biomedical Engineering
Background:
- Skeletal discrepancies significantly impact facial aesthetics and function.
- Accurate classification of sagittal and vertical discrepancies is crucial for effective orthodontic treatment planning.
- Traditional cephalometric analysis can be time-consuming and may lack comprehensive 3D morphological insights.
Purpose of the Study:
- To apply machine learning (ML) models for classifying sagittal (Class II, Class III) and vertical (hypodivergent, hyperdivergent) skeletal discrepancies.
- To evaluate the performance of different ML algorithms in discriminating these discrepancies using 3D facial scans.
- To analyze facial shape variability associated with skeletal discrepancies using principal component analysis.
Main Methods:
- Utilized 3D facial scans from 435 pre-orthodontic patients.
- Performed cephalometric analysis and 3D facial landmark identification.
- Trained and evaluated three ML models: Random Forest, AdaBoost, and Multi-layer Perceptron, using Area Under the Curve (AUC) and accuracy metrics.
- Employed Principal Component Analysis (PCA) to assess facial morphology variation.
Main Results:
- ML models demonstrated high performance, with AUCs for sagittal classification ranging from 0.91 to 0.95.
- Random Forest achieved the highest sagittal classification accuracy (88.5% for Class II, 95.5% for Class III).
- Multi-layer Perceptron showed the best vertical classification accuracy (78.8% for hypodivergent, 86.2% for hyperdivergent).
- Six principal components explained 94.0% of the facial morphology variation.
Conclusions:
- Machine learning models show significant promise in discriminating sagittal and vertical skeletal discrepancies from 3D facial scans.
- 3D facial soft tissue features are effective indicators for identifying skeletal discrepancies in most orthodontic patients.
- ML-based analysis offers a potentially efficient and accurate adjunct to traditional diagnostic methods in orthodontics.
