Elliptical Fourier analysis of lateral skull profiles as a tool to aid skeletal identification

Abstract

n n n When dealing with the identification of skeletonised human remains, osteological features can beused to reduce the pool of individuals to whom the remains may belong by estimating biologicalfactors such as ancestry and sex. While pre-existing quantitative methods for estimating ancestryand sex can achieve considerable accuracy, they typically use sparsely located landmarks, riskingthe loss of meaningful information between the data points. These methods also require substantiallevels of manual input, which is time consuming and often requires specialised recordingequipment. This also acts to restrict use of analytical methods to the laboratory. The overarchingaim of this thesis is to determine if morphology of the skull, specifically its outline in normalateralis as viewed in a 2D photograph, can be used to accurately estimate the ancestry and sex ofskeletons.n n n When capturing lateral 2D views of 3D skulls, anatomical characters that are positioned morelaterally act to obscure morphologies of interest due to the camera viewpoint. To evaluate the levelof information retained in 2D lateral views, the 'anterior profile line' visible as the skull margin wastraced in a 3D view. This also resulted in a reworking of the pre-existing craniometric terminologyused to describe skulls in 2D views (also extended herein to faces).n n n A standardised photography protocol was developed to acquire images of skulls in normabasalis, frontalis, lateralis, and verticalis, and applied to adult skulls of known sex from differentpopulations sourced from: the Pretoria Bone Collection, the Hamann-Todd Human OsteologicalCollection, the WM Bass Donated Skeletal Collection, the Robert J Terry Anatomical SkeletalCollection, the Chiba Bone Collection, and the Khon Kaen Osteological Collection. Photographswere taken in a manner that permitted the profile line to always be in focus in a lateral view, with asingle setup using: a full-frame Canon 6D camera body fitted with a 100 mm macro lens, and with aset subject-to-camera distance of 1.20 m.n n n To visualise the average morphology of each sex and ancestry group in photographic resolution(i.e., centroids of data), means were calculated of anterior and lateral skull photographs usingPsychomorph. These depictions of group means serve as texture mapped versions of groupcentroids for the skull outlines, and thus provide high-quality visual representations for groupcomparisons with an unknown skull.n n n Skull and cranial outlines were specifically extracted from the photographs and analysed withelliptical Fourier analysis (EFA) using 40 harmonics on partial Procrustes aligned coordinates. Bothsize-retained and size-normalised methods were employed. EFA coefficients were subject toprincipal component (PC) analysis to reduce the data and to assess variations in size and shape ofthe outlines. The PCs were statistically significant overall by ancestry and sex, for skull and cranialoutlines both with and without size information (p l 0.01). Linear discriminant function analysis was used to derive group classification models, and using five-fold cross-validation to estimateclassification accuracy.n n n Using skull outlines, size-retained data correctly classified ancestry and sex 73 % of the timewith all seven groups included in the comparison. This accuracy decreased to 60 % with sizenormalisation. Cranial outlines for the seven group comparison correctly classified 73 % and 62 %with size-included and size-normalised analyses respectively. A four-way assessment of ancestryand sex for Black and White females and males using size-retained skull outlines achieved 67-89 %correct classifications, with an overall accuracy of 78 %. For comparison, analysis in FORDISC 3.1using 14 measurements visible in norma lateralis achieved 66-80 % accuracy (74 % overall); andstepwise selection of the eight best performing variables achieved 70-84 % correct classifications(79 % overall).n n n Accuracy of out-of-group ancestry and sex estimation was assessed using nine DNA-identifiedskulls from the Defense POW/MIA Accounting Agency (DPAA) Laboratory. When compared toall seven groups, eight of the nine identified skulls (89 %) correctly classified as White male withsize retained, with one misclassification as White female (for both skull and cranial comparisons).These results indicate that skull and cranial outlines captured in norma lateralis contain enoughmorphological information to enable correct ancestry and sex estimation most of the time. Theportable and user-friendly nature of the EFA protocol make this method viable for use at skeletalrecovery sites. This would help to mitigate risk of erroneous repatriation of unknown humanremains, which is especially pertinent where large distances and expenses are involved. This sexand ancestry estimation method has been scripted in R and is freely available as SkullProfiler atCRANIOFACIALidentification.com.