MRI Protocol for Pediatric Brain Imaging

For all MR image acquisition, children under 4 years of age were scanned during natural and non-sedated sleep and older children were imaged whilst watching a movie or other video. Our imaging protocol included relaxometry, multi-shell diffusion, resting-state connectivity, and magnetic resonance spectroscopy acquisitions in addition to the anatomical data. As a result, depending on child compliance (sleeping and/or motion), high quality anatomical data were not collected or available for every child at every scan time-point. Following data acquisition, scans were inspected to ensure there were no motion-related artifacts and image blurring and ghosting. T1-weighted anatomical data were acquired on a 3T Siemens Trio scanner with a 12-channel head RF array. T1-weighted magnetization-prepared rapid acquisition gradient echo anatomical data were acquired with an isotropic voxel volume of $1.2\times 1.2\times 1.2{\text{mm}}^3$ , resampled to $0.9\times 0.9\times 0.9{\text{mm}}^3$ . Sequence specific parameters were: TE = 6.9 ms; TR = 16 ms; inversion preparation time = 950 ms; flip angle = 15 degrees; BW = 450 Hz/Pixel. The acquisition matrix and field of view were varied according to child head size in order to maintain a constant voxel volume and spatial resolution across all ages^{41 (link)}. Using a multistep registration procedure^{42 (link)}, a series of age-specific anatomical T1-weighted templates were created corresponding to 3, 6, 9, 12, 15, 18, 21, 24, 30, 36, 42, 48, 60, 72, 84, 96 and 108-month ages. At least 10 females and 10 males were included in each template. An overall study template was then created from these age templates, which was aligned to the MNI152 template^{43 (link)}. Each child’s anatomical T1-weighted image was transformed into MNI space by first aligning to their age-appropriate template and then applying the pre-computed transformation to MNI space, with the calculated individual forward and reverse transformations saved and used for the volumetric analysis described below. All template creation and image alignment were performed using a 3D nonlinear approach^{44 (link)} with cross-correlation and mutual information cost functions. We then applied the Desikan-Killiany-Tourville (DKT) cortical labeling protocol, FreeSurfer’s wmparc and aseg non-cortical (plus white matter) labels through Mindboggle^{45 (link),46 (link)}, resulting in volumetric output from 96 brain regions. Five regions with very small volumes were excluded: left inferior lateral ventricle, left vessel, right inferior lateral ventricle, left basal forebrain, and right basal forebrain.