Related Experiment Video
Updated: Jan 15, 2026
10:23
Author Spotlight: Three-Dimensional Cephalometric Landmark Annotation Demonstration on Human Cone Beam Computed Tomography Scans
Published on: September 8, 2023
3.6K
Age Estimation from Lateral Cephalograms Using Deep Learning: A Pilot Study from Early Childhood to Older Adults.
Ryohei Tokinaga1, Yuichi Mine1,2, Yuki Yoshimi3
1Department of Medical Systems Engineering, Graduate School of Biomedical and Health Sciences, Hiroshima University, Hiroshima 734-8553, Japan.
Journal of Clinical Medicine
|October 16, 2025
Summary
Deep learning models can estimate age from lateral cephalograms with approximately 2.5 years error. A male-only model slightly improved accuracy, showing potential for forensic and developmental research.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Automated age estimation from lateral cephalograms is valuable for clinical and forensic applications.
- Current methods require evaluation for accuracy across diverse age groups and sexes.
Purpose of the Study:
- To develop and assess a deep learning model for automated age estimation from lateral cephalograms.
- To investigate if sex-specific training enhances the predictive accuracy of age estimation models.
Main Methods:
- A retrospective study of 600 lateral cephalograms (ages 4-63) was conducted.
- A DenseNet-121 model was trained on mixed-sex, female-only, and male-only datasets.
- Performance was measured using Mean Absolute Error (MAE) and coefficient of determination (R²).
Main Results:
- The mixed-sex model yielded an MAE of 2.50 years (R²=0.84).
- Female-only and male-only models showed MAEs of 3.04 years (R²=0.82) and 2.29 years (R²=0.83), respectively.
- Grad-CAM analysis highlighted sex-specific craniofacial region activations.
Conclusions:
- Deep learning models can estimate age from lateral cephalograms with clinically relevant accuracy (approx. ±2.5 years).
- A male-only model demonstrated a slight improvement in performance metrics.
- Findings support applications in forensic science, growth research, and identification of unknown individuals.
