The effect of microcomputed tomography scanning and reconstruction voxel size on the accuracy of stereological measurements in human cancellous bone.

Stereological parameters have been used as an approximation for the architecture of trabecular bone. Structural indices such as bone volume fraction (BV/TV), trabecular number (Tb.N), trabecular thickness (Tb.Th), trabecular separation (Tb.Sp), bone surface-to-volume ratio (BS/BV), degree of anisotropy (MIL1/MIL3), and connectivity density (-Euler/Vol) have been widely studied to investigate pathological conditions in bone. Due to its high resolution and nondestructiveness, microcomputed tomography (micro-CT) has been utilized to take precise three-dimensional (3D) images of trabecular microstructures. However, spatial limitations for applying micro-CT-based analyses to large specimens, such as whole vertebral bodies, require using larger scanning and reconstruction voxel sizes. In this study, combinations of three different scanning and reconstruction voxel size were used to represent best possible voxel size (21 microm; best in our scanner for the specimen size used) relative to other voxel sizes used in this study, commonly used intermediate voxel sizes (50 microm), and those applicable to scans of whole human vertebral bodies (110 microm) in order to examine the effect of scanning and reconstruction voxel size on stereological measures for human cancellous bone. The error in stereological parameters calculated using combinations of large voxel sizes compared to the gold standard (best possible case) ranged from 0.1% to 102%. The signed magnitude of the error in other cases relative to the gold standard was a function of either scanning or reconstruction voxel size or both (r2=0.55-0.95). For most of the structural indices, the results from analysis of images with larger voxel sizes were correlated with those from the gold standard (r2=0.55-0.99) except for Tb.N at 110/110 microm, MIL1/MIL3 at larger than 110 microm reconstruction voxel size, and -Euler/Vol at any combination of voxel sizes. Overall, it was observed that resampling a high resolution image at lower resolutions (corresponding to increasing reconstruction voxel size in this study) had different effects on the calculated parameters than scanning at the same low resolution (corresponding to increasing scanning voxel size in this study). Our results show that investigations of image resolution should include actual scans at the resolution of interest rather than simply coarsening of high-resolution images as is customarily done.