Facial soft tissue thickness (FSTT) estimation models-And the strength of correlations between craniometric dimensions and FSTTs

It has been speculated that craniometric dimensions can be used to improve estimations of facial soft tissue thickness (FSTT) in craniofacial identification. Subsequently, linear regression (LR) models have been published, but the practical utility of these models (lower errors than means) has never been tested/demonstrated. Using 71 living subjects measured by B-mode ultrasound, this study calculates and compares standard errors for previously published LR models and untrimmed FSTT means. Correlations between craniometric dimensions and FSTTs were calculated and regression model reproducibility examined by: generating new models using a 61 subject training set; and three-fold cross validation. Published regression models, applied to the above mentioned new individuals of this study, provided substantially worse estimates of ground truth FSTTs than untrimmed arithmetic means (mean S-est = 4.0 mm compared to 2.8 mm, n = 61-71). Correlations between craniometrics and FSTTs were generally small (mean of absolute values = 0.17, raw interval = -0.24 to 0.48) and only two of 15 previously published LR models were reproducible (mr-mr' and g-g')-i.e., contained the same independent variable with no more than one other different independent variable entering the model. Under three-fold cross-validation (training sets of 40-41 individuals), no LR equation was reproduced across all three validation test runs. Basic craniometric dimensions do not appear to generally improve FSTT estimations and relationships between craniometric dimensions and FSTTs are much weaker and less reliable than previously thought. B-mode ultrasound data for adult Australians were pooled herein to provide larger sampled and updated FSTT statistics for this cohort (n = 118-123). (c) 2018 Elsevier B.V. All rights reserved.