The impact of ct-data segmentation variation on the morphology of osteological structure

CT (computed tomography) scans have become indispensable tools for gross anatomy teaching and research [1-5]. Computational methods can create high-resolution 3D models of anatomical structures for education, research, and clinical applications [6-9]. However, data processing has a large influence on 3D model generation. Understanding how these differences in processing alter morphology, and whether such disparities impact the conclusions one may draw from the data, is imperative for interpreting radiology ground truth. Failure to account for these differences can lead to erroneous decisions regarding joint repair, joint replacement, and prosthetics design. In this work, we investigate how segmentation algorithms influence the morphology of 3D models of osteological structures (femurs) from human cadaveric CT scans. We measure dissimilarity in 3D model morphology resulting from multiple different segmentation protocols. As CT scanderived 3D models become more commonplace in gross anatomical research, it is critical to fully understand proper segmentation approaches and how much variation is acceptable for 3D anatomical models derived from radiological imaging.