Predicting Perceived Profile Attractiveness From Cephalometric Measurements Using Machine Learning.

Area of Science:

• Orthodontics
• Aesthetic Dentistry
• Machine Learning in Healthcare

Background:

• Facial aesthetics are crucial in orthodontic treatment planning.
• Objective assessment of attractiveness is challenging.
• Lateral cephalometric measurements offer quantifiable data for analysis.

Purpose of the Study:

• To compare machine learning (ML) models for predicting perceived facial attractiveness.
• To identify key cephalometric parameters influencing attractiveness scores.
• To evaluate the utility of ML in objective aesthetic assessment for orthodontics.

Main Methods:

• Utilized lateral cephalometric radiographs and facial photographs from 400 patients.
• Employed 26 raters (laypeople and experts) to score attractiveness via Visual Analogue Scale (VAS).
• Trained five ML models (Linear Regression, SVM, XGBoost, Random Forest, ANN) on cephalometric data, evaluating performance with RMSE.

Main Results:

• Random Forest model achieved the best performance (RMSE = 0.50).
• Soft tissue parameters, including facial convexity, maxillary/mandibular prognathism, and vertical proportions, were most predictive.
• Experts rated attractiveness higher than laypeople; rater gender influenced ratings in specific age groups.

Conclusions:

• ML models demonstrate potential for predicting facial attractiveness from cephalometric data.
• Facial aesthetics are significantly correlated with specific soft tissue cephalometric parameters.
• This ML-based approach can assist clinicians in objectively evaluating orthodontic treatment outcomes.