An automated computer-assisted approximation of the nose in South Africans from CBCT (Cone Beam Computed Tomography) scans

Abstract

Each year in the Gauteng province of South Africa, approximately 1300 bodies are incinerated without a known identity (Bloom, 2015; Krüger et al., 2018). Because of various socio-economic reasons, identification is not always possible with conventional methods such as DNA comparisons and fingerprints. Therefore, more creative methods, including facial reconstruction, have been implemented to assist in the identification of unknown persons from their skeletal remains in South Africa. The aim of this thesis was to provide an automated computer-assisted method, independent of any forensic artistic interpretations, to create accurate statistical models for predicting nasal soft-tissue shape from information about the underlying skull substrate. The acquisition and extraction of the relevant anatomical structures (hard- and soft-tissue) were performed using an automatic dense landmarking procedure and analysed by geometric morphometrics. In this research, a validation of the precision of the automatic placement of landmarks, demonstrated its utilisation as a convenient prerequisite for geometric morphometric based shape analysis of the nasal complex. The automatic landmark positioning on hard- and soft-tissue 3D surfaces offered increased objectivity and the possibility of standardisation. In addition to reducing measurement errors in landmark placements, automatic landmarking, achieved a better precision for facial approximation, enabling the possibility to include more samples and populations with ease. A detailed study of the influence of factors (ancestry, sex, ageing and allometry) on the variability of the mid-facial skeleton among two South African ancestral groups were performed, revealing their statistically significant influences on the overall shape variation of the nose. Ancestry was found to be a very important factor in shape variation within the sample emphasising ancestral-specific differences. In addition, the expression of sexual dimorphism and effect of aging appeared to be different on distinct elements of the shape of the mid-facial region. From the findings, the two South African groups differed significantly regarding hard- and soft-tissue nasal complex morphology and their correlations, emphasising the importance of considering ancestry, sex and age as factors in the process of approximating the nose and highlighting the need for population specific accurate and reliable 3D statistical nose prediction methods. This study provided accurate statistical models using Partial Least Squared Regression (PLSR) algorithms which were optimised by including additional information such as ancestry, sex and age. Age and sex appeared to be important factors to be considered as additional information in order to improve the quality of the prediction. The predictions were based on a sample of 200 specimens resulting in an error when using the landmark-to-landmark distances on non-trained data, ranging between 2.139 mm and 2.833 mm for black South Africans at the tip of the nose and the alae, while they ranged from 2.575 mm to 2.859 mm for white South Africans. This research is the first attempt at a computer-assisted facial approximation of the nose with an automatic landmarking approach for the development of valid and reliable South African population specific standards using Cone Beam Computer-Tomography scans.