Standardization of bone mineral density and microstructure from high-resolution CT-scans of the spine in a multicenter setting.

Abstract

Purpose: Quantitative Computed Tomography (QCT) has not fully addressed the need to reduce intra- and inter-scanner variability for Osteoporosis and bone-related studies, which can lead to inaccuracies when pooling data from different CT manufacturers, models, devices, or protocols. In this context, the aim of this work was to develop ex vivo methods for the standardization of bone mineral density and microstructural parameters.

Materials and methods: Six human vertebral body specimens embedded in poly-methyl methacrylate (PMMA) were scanned ex vivo inside an anthropomorphic abdomen phantom in eight different CT-scanners. We measured 3D trabecular and cortical bone mineral density (Tb.BMD and Ct.BMD at the peeled spongiosa and the vertical cortex, respectively), trabecular separation (Tb.Sp) and cortical thickness (Ct.Th). Standardization of Tb.BMD and Ct.BMD across CT-scanners was conducted by correcting for the influence of PMMA and kernel related differences in the segmented cortical volume. For Tb.BMD and Ct.BMD two CT-scanners, where the majority of the patients were scanned, were used as reference. For Tb.Sp standardization we accounted for the image binarization threshold and used high-resolution peripheral QCT (HR-pQCT) as reference. Cross-calibration factors were obtained for each CT-scanner from which the cross-calibrated measures xTb.BMD, xCt.BMD and xTb.Sp were computed both ex vivo and in vivo. Agreement of the ex vivo measurements with respect to the references was quantified with Lin's concordance correlation coefficient (r_CCC) before and after standardization. For the clinical in vivo part of the study, 152 patients (24M, 128F) undergoing long-term bisphosphonate treatment had their T12 or L1 vertebrae scanned with the same CT-scanners and protocols as for ex vivo. Statistical bone fracture models were conducted before and after cross-calibration to assess the performance of the standardization procedure in vivo.

Results: After cross-calibration the overall ex vivo mean Tb.BMD across CT-scanners was basically maintained, changing only from 119.0 mgHA/cm³ to 119.4 mgHA/cm³. The mean Ct.BMD raised from 420.4 mgHA/cm³ to 441.1 mgHA/cm³. Tb.BMD showed a small variability (SD of means) across centers of 2.7 mgHA/cm³. For Ct.BMD additional kernel related thickness correction reduced this variability from 31.7 mgHA/cm³ to 22.4 mgHA/cm³. Non-standardized Tb.Sp showed a mean of 2.63 mm across CT-scanners, which after standardization was corrected to 1.18 mm. Agreement to the reference measurements was markedly improved after standardization (before: the r_ccc [min, max] for Tb.BMD, Ct.BMD and Tb.Sp was [0.64, 0.92], [0.40, 0.89] and [0.57, 0.99], respectively; after standardization: [0.98, 0.99], [0.96, 0.99] and [0.78, 0.99], respectively). For in vivo, Tb.BMD and Ct.BMD showed a mean (SD of means) across CT-scanners before standardization of 72.3 (7.6) mgHA/cm³ and 352.4 (44.6) mgHA/cm³, respectively and after standardization 72.6 (7.0) mgHA/cm³ and 370.7 (31.0) mgHA/cm³, respectively. Non-standardized Tb.Sp showed a mean (SD of means) of 3.55 (2.42) mm across CT-scanners, which after standardization was corrected to 1.65 mm (0.16) mm. The cross-calibrated xTb.BMD showed a highly statistical significance in prevalent fracture classification (p = 0.0001) similar to Tb.BMD (p = 0.0002). For xCt.BMD a trend was observed in improving fracture prediction, albeit not significant (p = 0.14), compared to Ct.BMD (p = 0.23). xTb.Sp demonstrated improved fracture prediction (p = 0.024) compared to a non-standardized Tb.Sp (p > 0.1).

Conclusion: The improved inter-scanner agreement with corresponding reduced variability underscores the importance of cross-calibration of bone mineral density and microstructural parameters. For the in vivo application of the methods, cross-calibrated Tb.Sp improved fracture prediction in patients, whereas cross-calibrated BMD had no discernible impact, possibly due to the distribution of patients across the participating CT-centers and the already high fracture classification power of Tb.BMD.