Artificial intelligence -based modeling and comparative evaluation of craniofacial soft tissue and subcutaneous fat...

Craniofacial soft tissue thickness (CFSTT) is a critical parameter in forensic facial reconstruction, serving as a link between skeletal morphology and facial appearance. However, most existing studies rely on mean CFSTT values without considering the contribution of subcutaneous fat layer thickness, limiting the accuracy of reconstruction models.

This study aims to develop an artificial intelligence (AI)-assisted, CT-based framework for the comparative evaluation of craniofacial soft tissue thickness and subcutaneous FLT in a Northwestern Indian population, with emphasis on sexual dimorphism, age-related variation, and bilateral symmetry.

A retrospective cross-sectional analysis was conducted on CT scans of 1972 individuals aged 18-80 years. Measurements of CFSTT and FLT were obtained across 73 standardized craniofacial landmarks. Statistical analysis included descriptive statistics, independent t-tests, ANOVA, and Pearson correlation. AI-based predictive modeling was performed using regression and machine learning algorithms to evaluate relationships between variables and improve prediction accuracy.

CFSTT values were consistently higher than FLT across all anatomical landmarks, indicating the contribution of both adipose and non-adipose components. A strong positive correlation between CFSTT and FLT was observed at the landmark level (R² ≈ 0.785), while a weaker association was noted at the age-group level (r ≈ 0.143). Significant sexual dimorphism was identified, with males exhibiting higher values than females, although effect sizes were small. Age-related analysis demonstrated an increase-peak-decline pattern, with maximum values observed in middle adulthood. Bilateral symmetry analysis revealed extremely high correlation between right and left sides (R² = 0.992). AI-assisted models demonstrated strong within-dataset performance, with R² values exceeding 0.90 for the best-performing models, and showed limited systematic bias.

The integration of CFSTT and FLT provides a population-specific framework for characterizing craniofacial soft tissue architecture. AI-assisted models demonstrated strong within-dataset performance, particularly the Random Forest model; however, independent external validation is required to establish model generalizability. The resulting database provides reference data that may support future research in population-specific craniofacial modeling and forensic facial approximation.