A validation study of newly established ear morphology prediction guidelines in facial approximation

Forensic facial approximation and superimposition methods rely on understanding soft and hard tissue anatomical correlations. Guyomarc'h and Stephan investigated eight previously published ear prediction methods, finding that computed tomography scans of 78 living adults did not support any of these recommendations. However, correlations between ear features, sex, age, and facial height led to the development of four validated prediction methods, though accuracy remains a challenge due to significant error. Utilising an independent sample, this study aimed to test these existing protocols and trail newly proposed regression equations. This study utilised 37 computed tomography scans from the New Mexico Decedent Database (deceased) and the University of Science and Technology Hospital Yemen database (living). Observations were performed in 3D Slicer. Skull placement and linear measurement definitions were standardised via Python scripting. All statistical tests were performed using SPSS. Following an assessment of normal distribution, potential sexual dimorphism, asymmetry, and mean differences in population affinities were examined using independent t-tests for normally distributed variables and Mann-Whitney U tests for non-parametric variables. Correlations between measurements and chronological ages were determined using Pearson’s r and Spearman’s correlation. Stepwise multiple regression was used to devise new prediction protocols. This project largely supported Guyomarc’h and Stephan’s findings. In contrast, guidelines proposing only slight differences between the height of the nose and height of the ear (c.2 mm) was found to be unreliable (explains only 13.47% of variation); no correlation was found between the supramastoid crest and auricle (p=0.149), and that lobe morphology correlated (r=0.487) with mastoid lateral angle. Comparably high standard errors of the estimates for regression equations established in this study (especially for ear height) were comparable to those of Guyomarc’h and Stephan. Future research should focus on expanding ear approximation guidelines, potentially incorporating additional landmarks, investigating interlandmark distances and semilandmarks, or adopting a geometric-morphometric approach to enhance reliability. Furthermore, the wider implementation of deep learning in forensic anthropology could prove beneficial in identifying associations between hard and soft tissue features in the auriculotemporal region, especially when established relationships are deemed unreliable with limited alternative options.