Comparisons of Bone Density Measurements between Quantitative Computed Tomography and Magnetic Resonance Ideal Imaging

Introduction-Quantitative computed tomography (QCT) has been commonly used to acquire subject-specific, volumetric bone densities in vivo. With QCT, bone is modeled as a specimen composed of water and mineral, and the x-ray attenuation coefficients (i.e., Hounsfield number) after being corrected with scan phantoms, is used to calculate the mineral density of each voxel.[1] With QCT, a heterogeneous bone finite element (FE) model can be generated by assigning 3-dimensional density measures to FE bony mesh. However, as articular cartilage exhibits little signal on CT, such an approach becomes challenging when structuring cartilage contact problems (e.g., patellofemoral joint). As such, we utilize fat-water chemical shift imaging (IDEAL MRI) [2-3] to estimate bone density in vivo with the assistance of a calcium hydroxyapatite (CHA) calibration phantom. The assumption is that 1) bone consists of water, fat and mineral, and 2) bone mineral density is negatively associated with the porosity of bone (i.e., space occupied by water and fat molecules). Thus, bone densities are hypothesized to be negatively corrected with their in-phase (IP: water+fat) values acquired from IDEAL protocol. The purposes of this study are 1) to investigate the relationship between CHA densities and MR signal intensities, and 2) to correlate the bone density measurements between QCT and IDEAL. Methods-Five 40ml phantoms with various mass of CHA (0 g, 4g, 8g, 12g, and 16g) and distilled water, which were equivalent to CHA density 0 g/ml, 0.1 g/ml, 0.2 g/ml, 0.3 g/ml, and 0.4 g/ml, were constituted for MR scan on a 3T GE scanner with a single-channel head coil. To examine the phantoms in a suspension state, a spoiled-gradient-echo pulse sequence (TR=10ms, TE=4.4ms, flip angle=5o, slice thickness=2 mm, FOV= 150*150 mm, matrix=128*128, NEX=12) was performed immediately after completely mixing CHA and water. As there was no fat in the CHA phantoms, TE was arbitrary. The image signal intensity of each CHA phantom was defined as average signal intensity within each phantom sample on 4 slices (Fig. 1). The signal intensities were plotted against the CHA densities and Pearson correlation coefficient was calculated. The sample with CHA density 0.4g/ml was then utilized as a calibration phantom for quantifying CHA equivalent density in vivo during an IDEAL scan. A spoiled-gradient-echo IDEAL pulse sequence was performed on a female’s knee joint (40 year, 1.67 m, and 57 kg) with an 8-element knee coil: TR= 20.2 ms, TE= {1.68 2.67 3.65 4.63 5.62 6.61} ms, slice thickness= 2 mm, FOV= 160*160 mm, matrix= 256*256, BW= 125 kHz. The CHA equivalent density (ρCHA) of each voxel within patella was calculated using the following equation from IDEAL reconstructed IP imaging (Fig. 2):