Quantification of 3D microstructural parameters of trabecular bone is affected by the analysis software.

Abstract

Over the last decades, the use of high-resolution imaging systems to assess bone microstructural parameters has grown immensely. Yet, no standard defining the quantification of these parameters exists. It has been reported that different voxel size and/or segmentation techniques lead to different results. However, the effect of the evaluation software has not been investigated so far. Therefore, the aim of this study was to compare the bone microstructural parameters obtained with two commonly used commercial software packages, namely IPL (Scanco, Switzerland) and CTan (Bruker, Belgium). We hypothesized that even when starting from the same segmented scans, different software packages will report different results. Nineteen trapezia and nineteen distal radii were scanned at two resolutions (20 μm voxel size with microCT and HR-pQCT 60 μm). The scans were segmented using the scanners' default protocol. The segmented images were analyzed twice, once with IPL and once with CTan, to quantify bone volume fraction (BV/TV), trabecular thickness (Tb.Th), trabecular separation (Tb.Sp), trabecular number (Tb.N) and specific bone surface (BS/BV). Only small differences between IPL and CTan were found for BV/TV. For Tb.Th, Tb.Sp and BS/BV high correlations (R2≥0.99) were observed between the two software packages, but important relative offsets were observed. For microCT scans, the offsets were relative constant, e.g., around 15% for Tb.Th. However, for the HR-pQCT scans the mean relative offsets ranged over the different bone samples (e.g., for Tb.Th from 14.5% to 19.8%). For Tb.N, poor correlations (0.43≤R2≤0.81) for all tested cases were observed. We conclude that trabecular bone microstructural parameters obtained with IPL and CTan cannot be directly compared except for BV/TV. For Tb.Th, Tb.Sp and BS/BV, correction factors can be determined, but these depend on both the image voxel size and specific anatomic location. The two software packages did not produce consistent data on Tb.N. The development of a universal standard seems desirable.