Process by which images were prepared for analysis: 1) Unified segment and bias-correction. Images were segmented into gray matter, white matter, and cerebrospinal fluid, and
field inhomogeneity-corrected images were produced with SPM8’s “unified segment” module (49). From the next step forward, image preparation steps were performed on gray and white matter segmented images separately.
2) DARTEL import. Gray and white matter segmented images were imported into Diffeomorphic Anatomical Registration Through Exponentiated
Lie Algebra (DARTEL), a component of SPM8 that determines an average-shaped template from all provided images and calculates
high dimensional spatial flow fields between each image and the template (30). During import, images were rigidly aligned and resampled to 1.5 mm cubic voxels. 3) Within-subject DARTEL. For each subject, DARTEL was used to calculate flow fields between Time 1 and Time 2 segmented images
and a subject-specific gray or white matter template, which can be considered an image halfway between Time 1 and Time 2.
We refer to the within-subject flow fields as “A1” for the warp between Time 1 and subject template and “A2” as the warp between
Time 2 and subject template. 4) Between-subject DARTEL. DARTEL was used to calculate flow fields from each subject template to a simultaneously calculated
group template, an image representing all 100 subjects. We refer to the warp parameters between subject template and group
template as “B.” 5) A 12-parameter affine transformation from the group template to Montreal Neurologic Institute (MNI) template was calculated
for ease of interpretation of coordinate results. We refer to the affine transform from group template to MNI space as “C.”
6) Within-subject flow fields (A1 and A2) were applied, respectively, to inhomogeneity-corrected whole brain Time 1 and Time
2 images (produced in Step 1). We averaged the nonzero voxels of the resulting coregistered pair of images. 7) Each subject’s mean image was segmented into gray matter and white matter tissue with SPM8’s unified segment module. 8) A composition of warps from subject template space to MNI space was calculated with SPM’s deformations utility: [subject
template to group template] o [group template to MNI], or [B o C]. Composing warps so that they may be applied simultaneously prevents errors that would be introduced by resampling the
images multiple times. 9) Composed warps [B o C] were applied to the gray and white matter images produced in Step 7, resulting in segmented images in MNI space. Since
the spatial normalization information from subject to MNI space came from the subject-specific template, each time point contributed
to the normalization, avoiding potential bias caused by applying normalization parameters of a single time point to both time
points. 10) A composition of warps from each time point to MNI space was calculated with the deformations utility: [A1 o B o C] and [A2 o B o C]. 11) The MNI-registered segments were then “modulated” by (multiplied by the Jacobian determinant of) the warps from step 10
to preserve quantitative volume. The influence of independent normalization of each time point was minimized by applying Time
1 and Time 2 Jacobian determinants to the same segments. This resulted in MNI-registered Time 1 and Time 2 gray and white
matter segmented images whose intensities correspond to units of volume. 12) MNI-registered Time 1 and Time 2 images were smoothed with a Gaussian kernel 8-mm full-width at half-maximum. 13) Time 1 segment images were subtracted from Time 2 segment images to create images representing change in gray or white matter
volume over time. These different images were entered into statistical models to relate brain volume changes over time to
variables of interest such as hypoglycemia and hyperglycemia exposure. ImCalc, image calculator.