Related Experiment Video
Updated: Jun 26, 2026
Midface Hypoplasia and Cranial Base Morphology in Syndromic Craniosynostosis: A Comparative Analysis Study Using a Predictive Regression Model
Published on: November 4, 2025
Predicting Perceived Profile Attractiveness From Cephalometric Measurements Using Machine Learning
Zeynab Pirayesh1, Fatemeh Sohrabniya2, Seyed AmirHossein Ourang3
1Department of Orthodontics and Dentofacial Orthopedics, School of Dentistry, Zanjan University of Medical Sciences, Zanjan, Iran; Topic Group Oral Health, ITU/WHO/WIPO Global Initiative on Artificial Intelligence for Health, Geneva, Switzerland.
Objectives:
Orthodontic treatment planning requires careful consideration of facial aesthetics. This study aimed to compare machine learning (ML) models to predict perceived attractiveness based on lateral cephalometric measurements and to identify parameters that significantly impact perceived attractiveness.
Methods:
Lateral cephalometric radiographs and corresponding facial photographs of 400 patients were utilized in this study. Twenty-six raters, including 20 laypeople and 6 domain experts (art, maxillofacial surgery, and orthodontics), evaluated each photograph using a 1 to 5 Visual Analogue Scale (VAS), and cephalometric analyses using the Dolphin software served as the reference standard. Five ML models, linear regression, Support Vector Machine, XGBoost Regressor, Random Forest, and artificial neural network, were trained to predict attractiveness scores based on the reference standard, and their performance on a test set (n = 59) was evaluated using root mean square error (RMSE). Subsequently, feature importance analysis was conducted to identify cephalometric indices that had the largest impact on VAS scores.
Results:
Experts consistently rated facial attractiveness higher than laypeople, especially in age groups >15 years. Furthermore, female raters rated female profiles significantly higher and male profiles significantly lower than male raters, with differences evident in >15 age groups. Among the 5 ML models, Random Forest demonstrated the best performance (RMSE = 0.50). The feature importance analysis revealed that soft tissue parameters, especially those related to facial convexity, maxillary and mandibular prognathism, and vertical proportions, were most predictive of attractiveness scores.
Conclusion:
ML models showed promising results in predicting facial attractiveness based on cephalometric measurements and highlighted the correlation of facial aesthetics with soft tissue parameters. The proposed ML-based approach can aid clinicians in the objective assessment of treatment outcomes.
