A spatially-dense regression study of facial form and tissue depth: Towards an interactive tool for craniofacial reconstruction

Forensic Craniofacial Reconstruction (CFR) is an investigative technique used to illicit recognition of a deceased person by reconstructing the most likely face starting from the skull. A key component in most CFR methods are estimates of facial soft tissue depths (TD) at particular points (landmarks) on the skull based on averages from databases of TD recordings. These databases vary in their method of extraction, number and position of landmarks (usually sparse <100), condition of the body, population studied, and sub-categorization of the data. In this work a new dataset is presented in a novel manner based on 156 CT scans using a spatially-dense set (∼7500) of TD recordings to allow for a complete understanding of TD variation interpolating between typical landmarks. Furthermore, to unravel the interplay between soft-tissue layers, skull and facial morphology, TD and Facial Form (FF) are investigated both separately and combined. Using a partial least squares regression (PLSR) analysis, which allows for working with multivariate and spatially-dense data, on metadata of Sex, Age and BMI, different significant patterns on TD and FF variation were found. A similar, but with TD and FF combined, PLSR generated a model useful to report on both, in function of Sex, Age and BMI. In contrast to other datasets and due to the continuous nature of the regression there is no need for data sub-categorization. In further contrast, previous datasets have been presented in tabulated form, which is impractical for spatially-dense data. Instead an interactive tool was built to visualize the regression model in an accessible way for CFR practitioners as well as anatomists. The tool is free to the community and forms a base for data contributions to augment the model and its future use in practice.