Convolutional Neural Network-Based Age Prediction from Cephalometric Images and Analysis of Site-Specific
Toshiro Emori1, Ryo Hamanaka2, Runa Yamaguchi-Higuchi1
1Department of Orthodontics, Nagasaki University Hospital, 1-7-1 Sakamoto, Nagasaki 852-8588, Japan.
Journal of Clinical Medicine
|July 28, 2026
Summary
This study developed an artificial intelligence (AI) model using convolutional neural networks (CNNs) to predict chronological age from lateral cephalograms, accurately identifying craniofacial growth patterns in younger patients.
Area of Science:
- Orthodontics and Craniofacial Biology
- Artificial Intelligence in Medicine
- Medical Imaging Analysis
Background:
- Accurate craniofacial growth prediction is crucial for orthodontic diagnosis and treatment planning.
- Current methods rely heavily on subjective clinical experience for assessing jaw growth.
- This study explores AI's potential to objectively analyze growth-related morphological features.
Purpose of the Study:
- To develop and evaluate a convolutional neural network (CNN) model for predicting chronological age from lateral cephalograms.
- To determine if AI can autonomously learn and identify growth-related craniofacial features.
- To identify the most informative anatomical regions for growth prediction using AI.
Main Methods:
- Retrospective analysis of 2114 lateral cephalograms from patients aged 5-30 years with malocclusion.
- Training a ResNet50-based CNN model on entire cephalograms and five regional subsets (mandible, maxilla, cervical vertebrae, frontal, cranial base).
- Performance evaluation using Mean Absolute Error (MAE), Pearson's correlation coefficient (r), and coefficient of determination (R²), with age trend analysis in growth and post-growth groups.
Main Results:
- The CNN model achieved high accuracy in predicting chronological age, especially in patients under 20 years old (MAE 1.16 years, r 0.952, R² 0.884).
- Model performance decreased significantly in the older age group (20-30 years).
- The mandible, cervical vertebrae, and maxilla were identified as the most informative regions for growth prediction.
Conclusions:
- The developed CNN model accurately predicts chronological age and captures age-associated craniofacial features related to growth, particularly in younger individuals.
- The mandible and cervical vertebrae regions are highly informative for AI-driven growth assessment, aligning with clinical practices.
- This AI approach shows promise as a foundation for future craniofacial growth prediction studies, with potential for transfer learning applications pending further validation.
