Kyra E. Stull

Accuracy and reliability of measurements obtained from computed tomography 3D volume rendered images

Forensic pathologists commonly use computed tomography (CT) images to assist in determining the cause and manner of death as well as for mass disaster operations. Even though the design of the CT machine does not inherently produce distortion, most techniques within anthropology rely on metric variables, thus concern exists regarding the accuracy of CT images reflecting an objects true dimensions. Numerous researchers have attempted to validate the use of CT images, however the comparisons have only been conducted on limited elements and/or comparisons were between measurements taken from a dry element and measurements taken from the 3D-CT image of the same dry element. A full-body CT scan was performed prior to autopsy at the Office of the Chief Medical Examiner for the State of Maryland. Following autopsy, the remains were processed to remove all soft tissues and the skeletal elements were subject to an additional CT scan. Percent differences and Bland-Altman plots were used to assess the accuracy between osteometric variables obtained from the dry skeletal elements and from CT images with and without soft tissues. An additional seven crania were scanned, measured by three observers, and the reliability was evaluated by technical error of measurement (TEM) and relative technical error of measurement (%TEM). Average percent differences between the measurements obtained from the three data sources ranged from 1.4% to 2.9%. Bland-Altman plots illustrated the two sets of measurements were generally within 2mm for each comparison between data sources. Intra-observer TEM and %TEM for three observers and all craniometric variables ranged between 0.46mm and 0.77mm and 0.56% and 1.06%, respectively. The three-way inter-observer TEM and %TEM for craniometric variables was 2.6mm and 2.26%, respectively. Variables that yielded high error rates were orbital height, orbital breadth, inter-orbital breadth and parietal chord. Overall, minimal differences were found among the data sources and high accuracy was noted between the observers, which prove CT images are an acceptable source to collect osteometric variables.

Ancestry estimation in South Africa using craniometrics and geometric morphometrics

Population history and positive assortative mating directs gene flow in such a way that biological differences are recognized among groups. In turn, forensic anthropologists quantify biological differences to estimate ancestry. Some anthropologists argue that highly admixed population groups, such as South African coloureds, cannot achieve acceptable accuracies because within group variance is too large. Whereas ancestry estimation in South Africa has been limited to craniometric data from South African blacks and whites, the current study integrates craniometric and geometric morphometric data from the three largest South African groups. Crania from 377 South African individuals (black=158, white=112, and coloured=107) comprised the sample. Standard measurements were collected and the coordinate data were subjected to Generalized Procrustes Analysis (GPA), which resulted in size-free shape variables (ProCoords). A principal component analysis was used to combine the shape variation captured in the ProCoords (ProCoords PC). Linear discriminant analysis (LDA), using equal priors, stepwise variable selection and leave-one-out cross-validation, was conducted on the ProCoords, the ProCoords PCs, and the traditional craniometric data. The LDA using 18 stepwise selected ProCoords resulted in the highest cross-validated accuracy (89%). Utilization of geometric morphometric data emphasized that the relative location of cranial landmarks was more discriminating than simple linear distances. Regardless of high levels of genetic admixture, South African coloureds are a homogeneous group and morphologically distinct from other contemporaneous South African populations. Furthermore, the present study demonstrated a correspondence between peer-reported race and morphological differences in the crania of black, white, and coloured South Africans.

Using multivariate adaptive regression splines to estimate subadult age from diaphyseal dimensions

Subadult age estimation is considered the most accurate parameter estimated in a subadult biological profile, even though the methods are deficient and the samples from which they are based are inappropriate. The current study addresses the problems that plague subadult age estimation and creates age estimation models from diaphyseal dimensions of modern children. The sample included 1,310 males and females between the ages of birth and 12 years. Eighteen diaphyseal length and breadth measurements were obtained from Lodox Statscan radiographic images generated at two institutions in Cape Town, South Africa, between 2007 and 2012. Univariate and multivariate age estimation models were created using multivariate adaptive regression splines. k-fold cross-validated 95% prediction intervals (PIs) were created for each model, and the precision of each model was assessed. The diaphyseal length models generated the narrowest PIs (2 months to 6 years) for all univariate models. The majority of multivariate models had PIs that ranged from 3 months to 5 and 6 years. Mean bias approximated 0 for each model, but most models lost precision after 10 years of age. Univariate diaphyseal length models are recommended for younger children, whereas multivariate models are recommended for older children where the inclusion of more variables minimized the size of the PIs. If diaphyseal lengths are not available, multivariate breadth models are recommended. The present study provides applicable age estimation formulae and explores the advantages and disadvantages of different subadult age estimation models using diaphyseal dimensions. Am J Phys Anthropol 154:376-386, 2014.

Sex estimation from the long bones of modern South Africans

Best scientific practice for sex estimation incorporates accurate techniques that employ appropriate standards and population- and period-specific data. Single measurements provide accurate sex estimations, but multiple measurements and multivariate techniques offer greater validity to biological profile assessments. Appropriate, modern standards for sex estimation are limited to the cranium in South Africans (SA), which warrants the examination of the potential for sex estimation using the postcrania of socially defined SA blacks, whites and coloureds through multivariate models and advanced statistical techniques. A total of 39 standard osteometric measurements were taken from the postcrania of 360 socially defined SA blacks, whites and coloureds (equal sex and ancestry). Univariate and multivariate models were evaluated. Multivariate models, with cross-validation and equal priors, were explored with linear and flexible discriminant analysis (LDA and FDA, respectively). Classification accuracies associated with univariate models ranged from 56 to 89%, whereas multivariate classification accuracies using bone models (i.e. all measurements from one element) ranged from 75 to 91%. The highest correct classifications were achieved with multivariate subsets (i.e. combinations of measurements from different bones) and ranged from 90 to 98%. Overall, FDA and LDA yielded similar accuracy rates. Postcranial bones achieve comparable classification accuracies to the pelvis and higher accuracies than metric or morphological techniques using the cranium. While LDA is the most commonly used classification statistic in biological anthropology, FDA provides a good alternative for classification.

Evidence of fatal skeletal injuries on Malapa Hominins 1 and 2

Malapa is one of the richest early hominin sites in Africa and the discovery site of the hominin species, Australopithecus sediba. The holotype and paratype (Malapa Hominin 1 and 2, or MH1 and MH2, respectively) skeletons are among the most complete in the early hominin record. Dating to approximately two million years BP, MH1 and MH2 are hypothesized to have fallen into a natural pit trap. All fractures evident on MH1 and MH2 skeletons were evaluated and separated based on wet and dry bone fracture morphology/characteristics. Most observed fractures are post-depositional, but those in the right upper limb of the adult hominin strongly indicate active resistance to an impact, while those in the juvenile hominin mandible are consistent with a blow to the face. The presence of skeletal trauma independently supports the falling hypothesis and supplies the first evidence for the manner of death of an australopith in the fossil record that is not attributed to predation or natural death.

Measuring distortion of skeletal elements in Lodox Statscan-generated images

Due to a scarcity of available skeletal material, anthropologists and other practitioners face difficulties with either the creation or validation of techniques used to estimate a biological profile in subadults. To address this problem, radiographic images of living individuals are often used in lieu of dry skeletal elements. However, radiographic images suffer from distortion. Some problems with metric analyses when using radiographic images may be addressed with the Lodox Statscan, an X‐ray machine that claims to produce minimal distortion along the scan‐axis due to a linear slot‐scanning design. The purpose of this research was to measure the distortion of skeletal elements in radiographic images generated from a Lodox Statscan. Skeletal elements subject to multiple imaging variables that affect distortion were radiographed, measured, and then compared to the dry bone measurements through multiple approaches. An 85% percent agreement within a ±1 mm range and a 97% agreement within a ±2 mm range was obtained. Percent difference results demonstrate that slot‐axis measurements incurred more distortion than scan‐axis measurements (11.8% and 2.7%, respectively). Inclusion of foam results in 4.5% more error than when foam is not included in the image. Angled scan‐axis measurements also incurred more distortion than either nonangled slot‐ and scan‐axis measurements. A Bland‐Altmanplot reveals an overall agreement between the radiographic and dry bonemeasurements, with most measurements falling within the upper and lower limits. Similar measurement error is found in Statscan radiographic and dry bone measurements; therefore, the Statscan offers a radiographic venue to collect metric data. Clin. Anat. 26:780–786, 2013.

Evaluating the Accuracy of Cranial Indices in Ancestry Estimation Among South African Groups

Historically, population differences were quantified using cranial indices. Even though the application of indices is associated with numerous statistical and methodological problems, the use of cranial indices to estimate ancestry persists as demonstrated by its inclusion in several recent papers and conference presentations. The purpose of this study was to classify 207 South African crania and compare the results of five standard cranial indices to linear discriminant analysis (LDA). New sectioning points were created to contend with low classification accuracies (40–79%) and possible secular trends. Although the accuracies of the new sectioning points increased (66–87%), the accuracies associated with the stepwise LDA were higher (84%) and could classify the crania into one of the three South African groups. The results of the study demonstrate that indices cannot compete with multivariate techniques and should not be used in forensic anthropological analyses for ancestry estimation.

Population differences in the postcrania of modern South Africans and the implications for ancestry estimation

The cranium is widely recognized as the most important skeletal element to use when evaluating population differences and estimating ancestry. However, the cranium is not always intact or available for analysis, which emphasizes the need for postcranial alternatives. The purpose of this study was to quantify postcraniometric differences among South Africans that can be used to estimate ancestry. Thirty-nine standard measurements from 11 postcranial bones were collected from 360 modern black, white and coloured South Africans; the sex and ancestry distribution were equal. Group differences were explored with analysis of variance (ANOVA) and Tukeys honestly significant difference (HSD) test. Linear and flexible discriminant analysis (LDA and FDA, respectively) were conducted with bone models as well as numerous multivariate subsets to identify the model and method that yielded the highest correct classifications. Leave-one-out (LDA) and k-fold (k=10; FDA) cross-validation with equal priors were used for all models. ANOVA and Tukeys HSD results reveal statistically significant differences between at least two of the three groups for the majority of the variables, with varying degrees of group overlap. Bone models, which consisted of all measurements per bone, resulted in low accuracies that ranged from 46% to 63% (LDA) and 41% to 66% (FDA). In contrast, the multivariate subsets, which consisted of different variable combinations from all elements, achieved accuracies as high as 85% (LDA) and 87% (FDA). Thus, when using a multivariate approach, the postcranial skeleton can distinguish among three modern South African groups with high accuracy.

Sex estimation of infants between birth and one year through discriminant analysis of the humerus and femur.

Estimation of sex when investigating subadult skeletal remains is largely problematic because of unreliable and inaccurate results. Despite the limitations encountered with skeletal material, the medical literature clearly demonstrates differences between males and females in utero that persist through life. The current study investigates sexual dimorphism in the long bones of the humerus and femur for individuals between birth and 1 year of age. A radiographic sample amassed from Erie County Medical Examiners office includes 85 femoral and 45 humeral images for analysis in relation to sex. Measurements for lengths and breadths were collected through tpsDig software. Discriminant analysis proved to be the most successful method, with error rates of 3% when utilizing maximum breadth at midshaft of the femur and 11% with humerus maximum distal breadth. This research demonstrates that it is possible to correctly classify sex of unknown subadult remains when comparing them to a known sample.

Subadult sex estimation from diaphyseal dimensions: STULL et al.

OBJECTIVES Many studies on subadult sex estimation focus on elements that express sexually dimorphic features in adults. In contrast, diaphyseal dimensions have been shown to display sex-specific differences prior to adolescence. The current study evaluates the use of diaphyseal dimensions in subadult sex estimation. MATERIALS AND METHODS Eighteen postcranial measurements from six long bones were collected on Lodox Statscan radiographic images of 1,310 modern South African children between birth and 12 years of age. Linear (LDA) and flexible discriminant analysis (FDA) and logistic regression were employed with single and multiple variable models with age both included and excluded from the model. Bootstrapped cross-validation was employed because some of the multiple variable subsets had small sample sizes. Each of the bootstrapped accuracies has an associated 95% confidence interval demonstrating the ranges in classification. RESULTS Classification methods utilizing multiple variables achieved the highest bootstrapped classification accuracies (70% to 93%). The inclusion of age in the models did not consistently increase or decrease the classification accuracies. Proximal and distal breadth measurements were consistently recognized as important measurements in model creation. FDA yielded the highest overall accuracies, but the logistic regression presented with overall smaller bootstrapped 95% confidence intervals. DISCUSSION Quantifiable sex differences were discovered in the appendicular skeleton of children between birth and 12 years of age. The high classification accuracies were likely due to using numerous predictor variables from multiple skeletal elements, which were optimized for classification using FDA. To facilitate application, a graphical user interface, KidStats, was developed.