Co-registration Periosteal and endosteal bone surfaces of the QCT datasets were segmented using the Medical Image Analysis Framework software package developed at the University of Erlangen [17]. A tetrahedral mesh model with third-order Bernstein Palbociclib manufacturer polynomial density functions was then calculated from the segmented QCT volume [18, 19]. The meshed QCT

volume was co-registered to the four DXA images using a general purpose 2D–3D deformable body registration algorithm [20–23]. A rigid registration allowing rotations and translations but not deformations was used. The 2D–3D registration algorithm used a fast GPU-based algorithm [24] to produce digitally reconstructed fan beam radiographic projections (DRRs) of the meshed volume at each angle that a DXA image was obtained. Each of the four DRRs was compared to the corresponding DXA image using mutual information. The sum of the mutual information of these image pairs served as a cost function. An optimization routine using simulated annealing (a robust method that avoids being trapped in local minima [25]) was used to determine the correct transform for the three translational and rotational parameters of the QCT meshed volume to co-register MAPK Inhibitor Library purchase it with the DXA images. The inverse of this transform was used to place a 1 mm plane at the center of the HSA NN and IT ROIs (which were defined

on the standard hip PA DXA image), onto the QCT dataset. This plane is the 2D slice on which the QCT parameters are calculated. The procedure of co-registration ensured that anatomically equivalent regions were measured by HSA and QCT. Because many of the QCT scans did not extend far enough below the lesser trochanter into the femoral shaft to allow a comparison to the HSA shaft ROI,

the comparison at the shaft ROI was not attempted. Calculation of parameters on the QCT dataset Cross-sectional area (CSA) in square centimeters was defined in accordance with the traditional GPX6 HSA definition as the area of the slice filled with bone. In this definition, the area of each pixel is weighted by the amount of bone in the pixel. Cross-sectional moment of inertia (CSMI) in quartic centimeters is defined around a given axis. In DXA HSA, CSMI is calculated and averaged over line profiles along the u direction in Fig. 1. The center line profile of HSA is a projection of the 2D slice in the PA image. CSMIHSA can therefore only be calculated around an axis perpendicular to the PA image (v in Fig. 1). However, QCT is not restricted by the directionality of the PA image, and one is free to choose the axis around which CSMI is calculated. Let (u, v, w) define an ortho-normal coordinate system centered at the center of mass (COM) of the 2D slice, ρ(u, v) be the volumetric bone density in milligrams per cubic centimeter per voxel in the slice, and ρ NIST = 1,850 mg/cm3.