Evaluating classification approaches for population affinity estimation in a contemporary South African CT-derived sample: an automatic landmarking-based approach

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Abstract Population affinity estimation remains an important component of the biological profile, particularly in demographically complex populations such as South Africa, where morphological overlap exists among groups. This study evaluated the performance of geometric morphometric (GMM) and inter-landmark distance (ILD) classification approaches derived from an automatic landmarking-based workflow for population affinity estimation in a contemporary South African sample. A total of 474 cranial computed tomography scans representing recorded Black, Coloured, Indian, and White South Africans were analysed. Eighteen three-dimensional craniofacial landmarks were automatically transferred from a reference template to individual cranial surfaces using rigid and non-rigid registration, and nine ILDs were calculated. Mean intra- and inter-observer landmark errors were 0.746 mm and 1.672 mm, respectively. GMM analyses identified significant craniofacial shape differences among all recorded population groups, with the greatest separation between Black and White South Africans and the greatest overlap involving Coloured and Indian South Africans. Cross-validated GMM-based discriminant function analysis achieved an overall classification accuracy of 90.08%, compared with 58.7% for ILD-based linear discriminant analysis and 52.0% for random forest classification. Nasal breadth contributed most strongly to ILD-based random forest classification. Although classification performance differed among methods, overlap remained evident across analyses. GMM shape variables provided substantially higher classification performance than the selected ILDs, whereas ILD-based models offered interpretable information on the craniofacial dimensions contributing to population variation. These findings reinforce the probabilistic nature of population affinity estimation and the importance of interpreting classification results within the context of continuous human biological variation.