Testing regression and mean model approaches to facial soft-tissue thickness estimation

Abstract

Average facial soft-tissue thickness (FSTT) databanks are continuously developed and applied within craniofacial identification. This study considered and tested a subject-specific regression model alternative for estimating the FSTT values for oral midline landmarks using skeletal projection measurements. Measurements were taken from cone-beam computed tomography scans of 100 South African individuals (60 male, 40 female; Mage = 35 years). Regression equations incorporating sex categories were generated. This significantly improved the goodness-of-fit ( r2-value). Validation tests compared the constructed regression models with mean FSTT data collected from this study, existing South African FSTT data, a universal total weighted mean approach with pooled demographic data and collection techniques and a regression model approach that uses bizygomatic width and maximum cranial breadth dimensions. The generated regression equations demonstrated individualised results, presenting a total mean inaccuracy (TMI) of 1.53 mm using dental projection measurements and 1.55 mm using cemento-enamel junction projection measurements. These slightly outperformed most tested mean models (TMI ranged from 1.42 to 4.43 mm), and substantially outperformed the pre-existing regression model approach (TMI = 5.12 mm). The newly devised regressions offer a subject-specific solution to FSTT estimation within a South African population. A continued development in sample size and validation testing may help substantiate its application within craniofacial identification.

Global facial soft tissue thicknesses for craniofacial identification (2023): a review of 140 years of data since Welcker’s first study

“…As few new subadults studies were published, only adult studies that reported means and sample size for ≥ 3 landmarks with clear landmark definitions were evaluated. This resulted in 53 new FSTT studies contributing to the 2023 T-Table update, including 33 FSTT studies published between 2018 and 2022 [ 9 , 13 , 124 – 154 ], and an additional 20 FSTT studies published pre-2018 [ 104 – 123 ] (Table 1 ). These new data include all six main FSTT measurement methods used, so far, for data collection (Table 2 ).…”

“…As facial approximations are undertaken in the blind, this maneuver tends to inject a degree of subjectivity into the methods. In contrast to FSTT means, regression approaches attempt to tailor estimations more precisely to individuals, typically via craniometric dimensions [ 8 , 9 , 13 , 14 ]. However, just like means, these estimates still retain errors that can sometimes be large.…”

“…However, just like means, these estimates still retain errors that can sometimes be large. In particular, the strength of craniometric correlations is weak, and generally have only been described for samples of small size, limiting their utility [ 9 , 13 , 14 ]. Consequently, arithmetic means continue to hold foundational value for craniofacial identification casework.…”

“…studies per T-Table 2008 [ 15 ] [ 3 , 8 , 18 – 22 , 24 , 32 – 72 ]; Fisher and Moorman (year unknown) cited in [ 73 ]; Köstler (1940), Bankowski (1958), Weining (1958), Weiber (1940); and Helmer (1984) cited in [ 74 ]. 55 2013 [ 16 ] 2008 + [ 23 , 25 , 75 – 90 ] 55 + 18 new Total = 73 2018 [ 17 ] 2008 + 2013 + [ 91 – 103 ] 73 + 13 new Total = 86 2023 2008 + 2013 + 2018 + [ 9 , 13 , 104 – 154 ]* 86 + 53 new* Total = 139 *New study additions reported herein …”

The development and testing of Thai facial soft tissue thickness data in three-dimensional computerized forensic facial reconstruction

“…For example, this comparison has been performed in a South African FSTT study looking specifically at the mouth-validation testing identified that a population specific approach offered improved results to the total weighted means, when estimating actual soft tissue depth measurements on a hold-out sample. 49…”