All of the GRASE (T2w images acquired at different echo times) images werecoregistered to the T1w TFE images followed by the reslicing the images to 1 mm3resolution using “reslice_img.m”.
The skull stripping of all theimages was then performed using a two-step procedure, whereby subject-specificbrain masks were generated in SPM12, which were then manually refined using theROIEditor toolbox in MRIStudio. The coregistered and skull-stripped mean images for each subject were then normalized to the “JHU_MNI_SS_T1_ss” template (55) in MontrealNeurological Institute (MNI) coordinate space (56). This step wasimplemented using the DiffeoMap toolbox in MRIStudio to carry out a12-parameter affine (linear) transformation with Automated Image Registration(AIR), followed by high-dimensional, non-linear warping with the largedeformation diffeomorphic metric mapping (LDDMM) algorithm (57). The LDDMManalysis was performed with cascading elasticity(i.e., alpha values of 0.01, 0.
005, and 0.002) to allow increasingly pliabledeformations, as previously reported (58). Each subjects’images were warped to normalized International Consortium for BrainMapping (ICBM) space (56) by applying theoverall Kimap (linear affine + non-linearLDDMM) transformation, as previously described (58).
TheMWF maps were then computed voxel-by-voxel basis using a regularizednon-negative least squares algorithm to extract individual T2 components frommulti-exponential T2 decay curves (21). In addition,extended phase graph algorithm was used to compensate stimulated echoes due toB1 heterogeneities (21,59).MWF was calculated as the ratio of T2 signal from 10-40 ms to thetotal signal (29).