[Zurück]

E. Dall´Ara, P. Varga, D. H. Pahr, P.K. Zysset:
"A Calibration Methodology of QCT BMD for Human Vertebral Body with Registered Micro-CT Images";
Medical Physics, 38 (2011), 5; S. 2602 - 2608.

Purpose: The accuracy of QCT-based homogenized finite element (FE) models is strongly related to the accuracy of the prediction of bone volume fraction (BV/TV) from bone mineral density (BMD). The goal of this study was to establish a calibration methodology to relate the BMD computed with QCT with the BV/TV computed with micro-CT (μCT) over a wide range of bone mineral densities and to investigate the effect of region size in which BMD and BV/TV are computed.
Methods: Six human vertebral bodies were dissected from the spine of six donors and scanned submerged in water with QCT (voxel size: 0.391 × 0.391 × 0.450 mm3) and μCT (isotropic voxel size: 0.0183 mm3). The μCT images were segmented with a single level threshold. Afterward, QCT-grayscale, μCT-grayscale, and μCT-segmented images were registered. Two isotropic grids of 1.230 mm (small) and 4.920 mm (large) were superimposed on every image, and QCTBMD was compared both with μCTBMD and μCTBV/TV for each grid cell.
Results: The ranges of QCTBMD for large and small regions were 9-559 mg/cm3 and −90 to 1006 mg/cm3, respectively. QCTBMD was found to overestimate μCTBMD. No significant differences were found between the QCTBMD-μCTBV/TV regression parameters of the two grid sizes. However, the R2 was higher, and the standard error of the estimate (SEE) was lower for large regions when compared to small regions. For the pooled data, an extrapolated QCTBMD value equal to 1062 mg/cm3 was found to correspond to 100% μCTBV/TV.
Conclusions: A calibration method was defined to evaluate BV/TV from QCTBMD values for cortical and trabecular bone in vitro. The QCTBMD-μCTBV/TV calibration was found to be dependent on the scanned vertebral section but not on the size of the regions. However, the higher SEE computed for small regions suggests that the deleterious effect of QCT image noise on FE modelling increases with decreasing voxel size.

http://dx.doi.org/10.1118/1.3582946