Additionally, the DXA technicians in this study were highly trained and accustomed to the careful attention to detail required in research studies. This expertise in patient positioning may also partially explain the important result that exactly matching the ROIs in 3D space with co-registration was not required for high correlations between DXA HSA and QCT for the NN region. We did YH25448 mw not foresee this surprising result, as one might intuitively expect that oblique planes caused by improper positioning could result in considerable variation, as well

as variations caused by limiting the determination of the narrowest point to a single 2D projection of a complex 3D object. The fact that the high correlations were seen, albeit with careful positioning, encourages the use of the HSA NN region in clinical studies where co-registration is not possible as a reasonable surrogate for measuring the “true narrow neck” with QCT. This result may also be due to the femoral neck region not having a well-defined weakest location. Physiological remodeling Momelotinib may cause the femoral neck to have a relatively large region which has approximately the same resistance to bending and compression, which would

make the exact placement of the NN region less critical. Previously, Prevrahl et al. [12] have undertaken a DXA QCT study comparing narrow neck region CSMI and reported an r 2 of 0.5, much less that the r 2 of 0.81 with non-registered ROIs and r 2 of 0.88 for co-registered ROIs reported here. The lower correlation found in the Prevahl study may have been due to a combination of different hardware

and algorithms used. Prevahl et al. used a Prodigy (GE/Lunar), and the QCT was performed on a GE9800-Q (GE Healthcare, Inc.) with an Image Analysis QCT phantom and with lower spatial resolution (1 mm × 1 mm × 3 mm voxel). A global threshold was used for the segmentation of the CT data. The algorithm utilized by Prevahl et al. were those contained in the GE/Lunar AHA® software for the DXA and for the QCT, those developed by Lang et al. [31, 32]. Also, careful co-registration was not used. Importantly, the high correlations reported in this study cannot be generalized to other structural measurement software and hardware implementations without further validation. In this study, we chose to only selleck screening library calculate on the QCT dataset that subset of HSA parameters for which highly these accurate QCT results can be obtained. Even the relatively high-resolution QCT used in this in vivo study cannot measure cortical thickness below 1–1.5 mm accurately [33]. Thus, we did not calculate on the QCT dataset parameters such as cortical thickness and buckling ratio where partial volume artifacts, in particular in elderly patients with decreased cortical thicknesses, would have had large effects. As the true cortical thickness and the true cortical BMC are not known, it is also extremely difficult to correct these artifacts in a theoretically rigorous manner.