Consistency Assessment for R2 Measurements Obtained with Different Techniques at 7 Tesla

Introduction R2 relaxometry has been established as an important technique in various quantitative MR approaches [1, 2]. In practice, transverse relaxation of H nucleus usually deviates from the ideal mono-exponential decay due to the influences of various factors such as B0 inhomogeneity [3] and slice profile imperfection [4]. Free induction decay (FID) distortion causes discrepancy between R2 values obtained with different measurement methods. This problem is becoming more and more important with the growing availability of high and ultra-high field scanners in recent years. In this study, we investigated the consistency between R2 values measured with three different methods in a group of volunteers at 7 Tesla. Materials and Methods Axial brain images of four healthy volunteers (3 male and 1 female, 38±15 years old) were acquired on an Achieva 7T whole body scanner (Philips, Cleveland, OH) with a 16-channel head coil (Nova, Wilmington, MA). High temporal resolution FID measurements were obtained with a multi-echo (ME) 2D gradient echo (GRE) sequence with the following scanning parameters: field-of-view (FOV) 220 × 180 mm, 20 slices with no gap, voxel size 0.86 × 0.86 × 3.00 mm, TR/FA = 1988 ms/90°, 31 TE’s ranging from 1.7 ms to 40.3 ms (ΔTE = 1.3 ms). Another data set with higher spatial resolution was also acquired with a 3D dual echo (DE) turbo field echo (TFE) sequence with the following scanning parameters: FOV 200 × 100 mm, 32 slices, no gap, voxel size 0.50 × 0.50 × 1.60 mm, TR/FA/TE1/TE2 = 13.3 ms/8°/2.3 ms/10.5 ms. For each volunteer, twenty cubic regions-of-interest (ROIs) were defined in low brain white matter by an experienced radiology reader. Each ROI covers 3 × 3 × 1 voxels in the ME images and 5 × 5 × 2 voxels in the DE images so that its size and volume were comparable between those data sets. ROI center positions were matched by rigid-body registration between the two data sets. The ROI’s were picked to represent brain regions with different levels of cross-slice B0-inhomogeneity (quantified by the maximum cross-slice difference in local resonance frequencies, γΔB0/2π). Eighty ROIs were thus grouped by heterogeneity level into three groups: low (γΔB0/2π ≤ 15 Hz, 27 ROI’s), moderate (15 Hz < γΔB0/2π ≤ 50 Hz, 35 ROI’s) and high (γΔB0/2π > 50 Hz, 18 ROI’s), for better illustration and analysis of the comparison results. Both the ME and DE data sets were fitted to the classical mono-exponential decay model. While R2 measurement with 2D GRE sequences is prone to FID distortion caused by cross-slice B0 inhomogeneity, 3D sequences are less sensitive to B0 variations along the slice selection direction. In order to take this factor into account, high temporal resolution FID curves in the 2D ME data set were also fitted to a model that corrects for the quadratic background field inhomogeneity effect [5]. The B0 information used by the quadratic correction model was collected with a B0 map sequence described in [5]. ROI-averaged R2 values obtained with the three measurement methods were compared using graphical and statistical techniques. Results R2 values obtained from the 3D DE sequence were plotted against mono-exponential (Fig. 1a) and quadratic correction (Fig. 1b) R2 values from the 2D ME sequence. The three groups of ROIs corresponding to low, moderate, and high levels of cross-slice B0 inhomogeneity were represented by dots, crosses, and triangles, separately. It is clearly demonstrated that the mono-exponential 2D ME R2 values are systematically larger than the 3D DE R2 values in regions with moderate and high inhomogeneity (Fig. 1a). After quadratic correction for the cross-slice B0 inhomogeneity, the 2D ME sequence can generate R2 values that are overall consistent with those from the 3D DE sequence in all three groups of ROIs (Fig. 1b). This observation is also supported by the paired student’s t-test result, which shows that the mono-exponential 2D ME R2 values are significantly different from the 3D DE R2 values (p < 0.01), while the difference between 3D DE and quadratic correction 2D ME R2 values are not statistically significant (p = 0.84). Moreover, there are several data points scattered far away from the line of unity in the data cloud in Fig. 1b. It suggests that the observed consistency is a population/group property. Substantial differences may exist in individual measurements of R2.