fMRI Data Acquisition and Pre-processing

A 3T Siemens Prisma with a 32-channel head coil was used to collect 0.8 mm isotropic T1-weighted and T2-weighted scans from all participants. Resting state fMRI images were acquired on the same day using multi-band gradient echo EPI (multi-band accel. factor = 6). The scans had high spatial (2.4 mm × 2.4 mm × 2.4 mm) and temporal (TR = 800ms) resolution [repetition time (TR) = 800 ms, echo time (TE) = 33 ms and flip angle = 52°]. A 2.4 mm isotropic spin echo field map was also collected during fMRI acquisition to correct for any distortion in the fMRI data.

We collected six resting-state fMRI scans that were 5 min long, with AP/PA phase encoding directions (60 axial slices each). Volumetric navigator sequences were used to collect T1- and T2-weighted sequences that were corrected for motion by repeating scans (47). During the resting scans, subjects focused their attention on a visual crosshair.

The imaging data was preprocessed using the HCP's preprocessing pipelines (v4.0.0) (43, 48–51). The structural preprocessing pipelines were used to generate subcortical segmentations and cortical surfaces. Following structural pre-processing, the functional pre-processing pipelines corrected for EPI distortion, registered the fMRI data to structural MRI, and then brought the cortical time series from the volume dimension to the surface. The denoising pipelines then registered the fMRI data to the structural MRI data and corrected for motion and distortions within fMRI data by mapping it onto a CIFTI grayordinate space and removing spatially specific noise. The MSMAll areal-feature-based cross-subject surface registration pipeline was then applied to align the individual subject's cortical regions to the HCP's multi-modal parcellation. This process is more accurate than using cortical folding alone. Finally, temporal ICA (50, 52) was used to clean global noise from the MSMAll aligned rs-fMRI data. For this process, weighted regression (43) of group spatial ICA components from a much larger HCP-Young Adult 1,071-subject dataset with an existing temporal ICA decomposition was applied and the resulting concatenated individual subject time series were unmixed using the previously computed temporal ICA unmixing matrix. The noise temporal ICA individual subject component timeseries from this larger dataset were then non-aggressively regressed out from the subject's timeseries (see Supplementary Methods for more information on pre-processing methods). DVARS excursions were used to quantify patient movement (53) and revealed no statistically significant difference between the two patient groups.