To assess the within- and between-subject variation of the myelin water fraction (MWF); intra- and extracellular water fraction (IEWF), and intra- and extracellular water geometric mean T2 time (IEW-gmT2) in a whole cerebrum MWI sequence¹.

To compute (1) the sample size necessary to detect predefined group differences, and (2) necessary group differences given predefined sample sizes based on our between-subject variability estimates.

For between-subject testing, a 3D GRASE dataset was acquired from 31 healthy volunteers (age 27.7 years $$$\pm$$$ 5 months) on a Philips Achieva 3T scanner with a 32-channel head coil and in-coil AC/DC conversion (dStream). Scan settings: ETL=32, TE = 10ms, TR = 1s, EPI factor 3, voxel size 1 x 1 x 2.5 mm³, SENSE factor 2. For within-subject testing, the GRASE sequence from a different subject (age 26.8) was acquired 4 times on a Philips Achieva 3T. A high-resolution 3DTFE T1-weighted image was obtained from all participants. Voxel per voxel the intensity decay curve was transformed into the underlying T2 distribution, using non-negative least squares with regularization constraint and accounting for stimulated echoes². From the T2 distribution, MWF and IEWF were calculated as the relative T2 fraction between 10 and 40 ms, and 40-200 ms, respectively³. IEW-gmT2 was the geometric mean T2 time between 40 and 200 ms. The TE=10ms image of each participant was affine registered to the T1-weighted image and normalized to MNI space using affine and diffeomorphic registration. The inverse transformation was applied to regions of interest (ROI) from the Johns Hopkins University (JHU) white matter atlas⁴. In each subject’s MWI space, mean MWF, IEWF and IEW-gmT2 values were obtained in the genu, body and splenium of the corpus callosum, fornix column and body, external capsules, anterior and posterior limb of the internal capsules, and anterior corona radiata. Furthermore, a total cerebral white matter mask was obtained from segmentation of the T1 image using SPM8. In each ROI, the mean (μ) and standard deviation (σ) across participants and across repeated scans were obtained. For within-subject and between-subject variability, the coefficient of variation (CoV) was computed as CoV = σ/μ. Using between-subject variability, the sample size (N) was estimated, necessary to detect a deviation (Δ), from the mean, assuming a two-sample t-test with 90% power (Z_β = 1.28) and 95% confidence level (Z_α =1.96) through the formula $$$N = \Bigg[\frac{(Z_{\alpha}+Z_{\beta})\sigma}{\Delta}\Bigg]^{2}$$$. Inversely, the detectable deviation from the mean was computed, given predefined sample sizes.Table 1 summarizes the within-subject variability of white matter ROIs. For MWF, the intra-subject standard deviation varied between 1.9 and 21.3% of the mean, depending on the region of interest. Lowest CoV (hence highest repeatability) was found for the total white matter ROI. Table 2 estimates the necessary sample size and detectable difference given between-subject variabilities. For studies in which two groups of 20 subjects are compared, the minimum detectable group difference is between 7% (e.g. splenium, PLIC, total WM) and 18% (fornix body) of MWF mean; between 0.9% (external capsules, total WM) and 2.6% (fornix body) of the IEWF mean and between 1% (external capsules, total WM) and 1.9% (body of CC) of the IEWgmT2 mean. Figure 1 illustrates the assessed regions and mean MWF across participants of the between-subject test.Previous studies that assessed (single-slice) white matter intra-subject variability reported a scan-rescan difference between 0.6 and 23% with intra-subject CoV between 4.4 and 25%⁵. Meyers et al. found scan-rescan CoV of 19%⁶, with regional CoV in the range of 10%-23%. The intra-subject variability found using the whole cerebrum 3D GRASE sequence is therefore comparable with these previous results.Within-subject repeatability of the whole cerebrum 3D GRASE sequence is similar to reported values in the literature based on single slice acquisition. Between-subject variability was assessed and used to obtain tables of sample size and detectable differences for assessment of MWF, IEWF and IEWgmT2. These tables can be used in power analyses for future studies using whole cerebrum MWI.