1Department of Exercise Science, High Point University, 2Department of Anthropology, University of Nevada, Reno, 3Department of Anatomy, Des Moines University, 4Department of Anatomy, University of Pretoria, 5Forensic Anthropology Program, Washburn University
March 29, 2019 , CC Ballroom BC
The utility of CT scan data for creating skeletal reference databases has gained recognition, but requires protocol standardization to ensure accuracy and comparability between specimens. Here, we investigate the impact of scan processing protocols and landmark placement on error in virtual data collection.
MSCT postmortem scans of subadult individuals were obtained from the University of New Mexico Health Sciences Center, Office of the Medical Investigator. Four researchers segmented the pelves of four randomly-sampled individuals and each performed four trials on a fifth individual. Deviation analyses in Geomagic Studio were used to visualize discrepancies and calculate pairwise comparisons of the root-mean-square error (RMSE) and the average deviation distances. Resultant RMSE values ranged from 0.13-0.78mm (mean=0.33mm), with an average deviation of 0.17mm between model vertices; although certain regions (e.g., ASIS) were prone to larger deviations. The effects of threshold value selection and smoothing protocol on overall deviation and variance at specific anatomical locations are discussed.
To examine error from landmark placement, three researchers each collected 32 pelvic landmarks on a set of 30 segmented pelves in Amira. Error was assessed using distances from mean landmark placement and a Procrustes MANOVA. The overall average error was 2.4mm. Mean error at individual landmarks ranges from 0.9-7.4mm; landmarks associated with pelvic height exhibited the highest error. The MANOVA demonstrated significantly more variation among individuals than within landmark placement (p=0.27).
Altogether, results suggest that inter- and intra-observer errors associated with CT scan segmentation and landmark placement are generally acceptable, but can improve with further protocol standardization.
This work was funded by the National Institute of Justice grants 2015-DN-BX-K409 and 2017-DN-BX-K0144.