Testing the use of three-dimensional surface rendering and mesh-to-mesh comparisons as a method of pair-matching commingled human skeletal remains
Date
2024
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
When dealing with mass graves, natural disasters, or mass fatalities, forensic anthropology faces a complex scenario that necessitates specific methods for sorting, matching, and identification. The main aim of dealing with commingled human remains is to sort the remains to identify the minimum number of individuals (MNI). Traditional sorting methods, such as visual-pair-matching, articulation, process of elimination, and taphonomy, are welldocumented but subjective and rely heavily on the experience and knowledge of the forensic anthropologist.
The study aimed to evaluate the effectiveness three-dimensional surface renderings of bone and the subsequent comparison of mesh-to-mesh values as a more objective and repeatable method for pair-matching commingled human remains. The Structure-from-Motion (SfM) technique was used to create and render 171 three-dimensional mesh model samples using Agisoft Metashape. These models were used to test the effectiveness of a new digital pairmatching method, the mesh-to-mesh value comparison (MVC) method against a South African populated sample. To generate the mesh-to-mesh values, different softwares were used: Viewbox 4.1, which is proprietary and uses a Trimmed Iterative Closest Point (ICP) algorithm to run the program, and Meshlab 2022.12, an open-source software that does not require a license and uses a general ICP algorithm to run the program. Each software was programmed to generate a root mean square value, which was used as the mesh-to-mesh value required for comparison.
Analysis was performed using two techniques: LCV (Lowest Common Value) mesh-to-mesh comparison using Microsoft Excel and ROC curve analysis performed using MedCalc. The LCV mesh-to-mesh comparison requires the user to filter, sort, and eliminate values based on observation using specific commands on Microsoft Excel until the lowest agreed upon value is found across left-right and right-left. The ROC curve analysis on MedCalc follows the principle stated by Delong et al. (1988), which works with a scoring system. The closer the threshold value is to the top-left corner of the graph, the higher the accuracy of the test being performed.
For the complete cadaveric humeri and femora analysis, both elements performed moderately well with the LCV mesh-to-mesh comparison method, while the ROC curve analysis produced considerably higher results, comparatively. For the complete cadaveric bones, the femur performed better with both softwares when compared to the humerus. Mesh-to-mesh values iv obtained through Meshlab and analysed with the ROC curve method generated significantly higher results and were more adept at correctly identifying and distinguishing differences between matches and non-matches. Differences in mesh-to-mesh values between sexes and different population groups were observed.
In the simulated fragmented remains, the distal mesh fragment performed the best with 100% sensitivity for Viewbox, and the shaft performed the best using Meshlab with sensitivity results greater than 79%. For the fragmented remains, the distal mesh fragment and the shaft mesh fragments generated significantly high results when using the values obtained through Viewbox and statistically analysed with the ROC curve method.
The mesh-to-mesh value comparison method was applied to a real-world forensic commingled assemblage with high fragmentation and taphonomically altered bones. The Structure-fromMotion technique was effective in capturing distinct details on bones, aiding in 3D modelling. However, the method failed to identify true positive matches in the humeri bone sample, as there were no individuals matched with visual pair matching in the original case.
In conclusion, the mesh-to-mesh value comparison method has the potential to improve the accuracy and objectivity of forensic anthropology in handling commingled human remains. By utilizing advanced technologies like SfM, ROC curve analysis, and real-world forensic commingled assemblages, this method can be applied to enhance the accuracy and reliability of forensic anthropological research.
Description
A research report submitted in partial fulfilment of the requirement for the degree of Master of Science in Medicine to the Faculty of Health Sciences, University of the Witwatersrand, School of Anatomical Sciences, Johannesburg, 2023
Keywords
Human skeletal, Commingled human remains