Related Experiment Video
Updated: Aug 9, 2026
Reliability of Artificial Intelligence-Based Cone Beam Computed Tomography Integration with Digital Dental Images
Published on: February 23, 2024
First-stage assessment of AI-based anatomical measurement accuracy for the Cameriere European dental age estimation
Zehra Selin Karauğuz1, Halenur Altan2
1Department of Pediatric Dentistry, Faculty of Dentistry, Necmettin Erbakan University Yaka Mah, Beyşehir Avenue, Bağlarbaşı Street No: 4, 42090, Meram, Konya, Türkiye. zehraselinkarauguz@gmail.com.
Aim:
This study represents the first methodological stage of a broader AI-assisted Cameriere European dental age estimation workflow. The aim of this first stage was to evaluate the performance of YOLOv8-based deep learning models in automatically detecting the anatomical landmarks and apical structures required for the Cameriere European method. Rather than directly estimating dental age, the proposed system was designed to automate the measurement-related inputs needed for subsequent Cameriere European-based age calculation.
Methods:
This retrospective first-stage validation study included 4,050 panoramic radiographs of boys and girls aged 5-13 years. Two YOLOv8-based models were developed to automate the anatomical measurement components of the Cameriere European method: a YOLOv8x-pose model for open-apex landmark detection and tooth-length reference point localization, and a YOLOv8x-seg model for closed-apex segmentation. Owing to model-specific anatomical eligibility criteria, 3,796 images were used for the pose model and 2,971 images for the segmentation model. Image annotations were performed using CranioCatch software according to a standardized annotation protocol. Model outputs were compared with manual reference annotations to evaluate landmark detection, measurement-related localization, and apical segmentation performance.
Results:
In the YOLOv8x-pose model, mAP_0.5 = 0.963 and mAP_0.5:0.95 = 0.842 were achieved; recall was 0.928 and precision was 0.918. Error metrics were MAE = 0.0032, RMSE = 0.0045, SMAPE = 2.09%, and the coefficient of determination R² = 0.9992.
Conclusion:
YOLOv8-based pose and segmentation models demonstrated technical feasibility for automating anatomical measurement extraction required for the Cameriere European dental age estimation method. Because dental age was not calculated in this first-stage analysis, the findings should be interpreted as measurement-level validation rather than complete dental age-estimation accuracy. Further validation is required to integrate AI-derived measurements into the Cameriere European formula and to compare AI-assisted dental age estimates with manual Cameriere-based assessment and chronological age.
