Validating the AFA3D forensic method using live CT data: A multiregional comparison of approximated and original...

Facial approximation aids identification of unknown individuals in forensic and anthropological contexts. Digital approximation methods estimate Facial Soft‑Tissue Thickness (FSTT) and facial shape from virtual skulls, which are meant to lower subjectivity. Yet substantial variability persists in predicted outcomes, particularly in the nasal and lower facial regions. AFA3D (Anthropological Facial Approximation in Three Dimensions), developed by Guyomarc'h et al. (2014) from French data, generates facial predictions using statistical shape modelling, FSTT‑based warping, and iterative algorithms. Earlier studies reported moderate error in the mouth and smaller errors in nasal and orbital areas, but its broader performance remains insufficiently evaluated. This study assesses AFA3D by comparing approximated faces with original facial meshes from 40 CT-scans, 10 each from Czech, Slovak, Egyptian, and French samples. Geometric morphometric comparison was conducted using Morphome3cs II. Across samples, 75.9-84.2% of facial surfaces fell within ±2.5 mm deviation. Systematic regional errors were observed in the nose, lips, chin, cheeks, and upper face, with males generally showing greater localised discrepancies than females. These patterns correspond to anatomical regions with limited skeletal constraint and to sex‑linked cranial structural differences, as observed in previous approximation validations. Overall, AFA3D produces predictions with consistent regional error patterns, underscoring the need for more detailed mapping of local deviations, better modelling of posture‑related influences, and continued refinement of automated approximation methods to strengthen forensic reliability.