Related Experiment Video
Updated: Sep 15, 2025
07:32
Author Spotlight: 3D Movement Assessment of Maxillary Posterior Teeth in Clear Aligner Treatment
Published on: February 23, 2024
1.3K
Deep learning-assisted comparison of different models for predicting maxillary canine impaction on panoramic
Chunmiao Zhang1, Hailin Zhu2, Hu Long3
1State Key Laboratory of Oral Diseases, National Center for Stomatology and National Clinical Research Center for Oral Diseases, Department of Oral Medical Imaging, West China Hospital of Stomatology, Sichuan University, Chengdu, Sichuan, China.
Summary
Two models accurately predict canine impaction using panoramic radiographs and deep learning. The Margot et al. model showed the best performance, though further refinement is needed.
Area of Science:
- Dentistry
- Medical Imaging
- Artificial Intelligence
Background:
- Panoramic radiography is key for predicting maxillary canine impaction.
- Existing prediction models require external validation.
- Deep learning aids automatic landmark detection for improved accuracy.
Purpose of the Study:
- Compare prediction accuracy of existing models for canine impaction.
- Validate models using an automatic landmark detection system.
- Evaluate model performance with metrics like AUC, sensitivity, and specificity.
Main Methods:
- Included patients aged 7-14 with impacted canines on panoramic radiographs.
- Utilized a deep learning system for automatic landmark localization.
- Assessed three models (Arnautska, Alqerban et al., Margot et al.) using diagnostic metrics.
Main Results:
- Margot et al. model demonstrated highest performance (AUC 0.97, sensitivity 95%, specificity 86%).
- Arnautska model showed good accuracy (AUC 0.94, sensitivity 93%, specificity 71%).
- Alqerban et al. model exhibited poor performance (AUC 0.20).
Conclusions:
- Two predictive models show good diagnostic accuracy for canine impaction.
- One model performed suboptimally, requiring further investigation.
- Limitations exist, necessitating refinement of even the most effective models.
