Machine learning-assisted three-dimensional mapping of the facial soft tissues in Saudi Arabian adults with skeletal class I relationship

OBJECTIVES: This study aimed to evaluate facial soft-tissue features in adults with skeletal Class I relationships using three-dimensional (3D) imaging and machine learning to identify sex-specific anthropometric patterns and provide reference data for orthodontic assessments. MATERIALS AND METHODS: A total of 94 Saudi Arabian adults (50 men, mean age 25 ± 2.1 years; 44 women, mean age 23 ± 1.9 years) with Class I skeletal and dental relationships and body mass index (BMI) <25 kg/m² were included in this cross-sectional study. Standardized 3D facial photographs were obtained using a radiation-free system (ProMax 3D ProFace, Planmeca Oy, Helsinki, Finland). Sex differences were evaluated using conventional statistical tests. An L1-regularized logistic regression model was applied for feature selection and sex classification. Model performance was assessed using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC). RESULTS: Significant sexual dimorphism was observed in facial regions. Men exhibited greater vertical and transverse dimensions, particularly in the mandible, middle face, and ocular region. Middle facial width was significantly greater in men than in women (135.82 ± 9.43 mm vs. 129.03 ± 7.13 mm; P < 0.001). Women showed higher vermilion height-to-mouth width and upper face height-to-mandibular width ratios. The machine learning model classified sex with high accuracy (91.3% ± 6.2%), precision (96.0% ± 8.0%), recall (85.5% ± 11.9%), F1-score (89.7% ± 7.0%), and AUC (0.91 ± 0.07). CONCLUSIONS: This study demonstrates that combining 3D facial imaging with machine learning effectively captures sexual dimorphism in a Saudi Arabian adult population. The findings provide clinically relevant reference data for personalized orthodontic assessment and treatment planning and support forensic applications.