Data quality and automatic landmarking for ~1,000 cranial CT scans from MorphoSource

Abstract

In any statistical analysis there’s a large investment of time, mostly unseen, dedicated to ensuring and consolidating the datasets that will be used to feed the models. In this study, we measured the distance Prosthion-Basion (PR-BA) between the real skulls and the CT models that were scanned at different slice spacing for the Southwest National Primate Research Center (SNPRC) baboon skull collection downloaded from MorphoSource (n=986). Many scans experienced unexpected problems of slice spacing identity, so we checked the automatic value when they were loaded in 3D Slicer [1], and the expected one in the header that was introduced manually. We compared the values with previously-published landmark data from the same CT models [2] and the real measurements. Our results indicate that scans for which we didn’t experience problems loading their DICOM files had a high level of overlap between the real PR-BA distance and the CT-measured distance, no matter the slice spacing. Skulls for which the slice spacing was automatically calculated by the software (and not the value present in the header), the distances exhibited a high level of overlap with the real values. In the latter skulls, when the slice spacing is manually-introduced to fit the header, the CT-measured distance is significantly overestimating the real distance. Because of this mis-match, we also validated the published data to confirm that there was no error in how they calculated the distances between their landmarks.