DWI preprocessing pipeline

DWI preprocessing

The DWI data were preprocessed using MRtrix3 (Tournier et al., 2019) and FSL (Smith et al., 2004). The images were first denoised using the Marchenko-Pastur PCA method (Veraart et al., 2016, Cordero-Grande et al. 2019) implemented with the MRtrix dwidenoise function. Then, to correct the distortions due to inhomogeneities of the magnetic field, FSL’s topup (Andersson, Skare, and Ashburner 2003) and eddy (Andersson and Sotiropoulos 2016) correction were used. The topup method estimates the susceptibility-induced distortions of the subject’s head from the pairs of images with opposite distortion patterns (because of acquisition with opposite phase-encoding directions - anterior-to-posterior and posterior-to-anterior). This was followed by eddy correction that corrects for eddy current-induced distortions, which are a consequence of rapid switching of the diffusion gradients. No bias field correction was done.

Fiber orientation density estimation and tractography

From this preprocessed data, the response functions (required for fiber orientation density estimation) for each of white matter, grey matter, and cerebro-spinal fluid tissue types were estimated using dwi2response dhollander the MRtrix implementation of the Dhollander algorithm (Dhollander et al., 2019). These derived response functions were then used to estimate the amount of diffusion in three orthogonal directions (known as fiber orientation density estimation) using multi-shell multi-tissue constrained deconvolution method implemented under dwi2fod in MRtrix.

Then to seed the streamlines from the grey matter-white matter interface in the next step, a mask of this grey matter-white matter boundary was first generated using the high-resolution segmented T1 image with the 5tt2gmwmi function in MRtrix. Finally, using this grey matter-white matter boundary mask and the estimated white-matter fiber orientation density, the second-order integration over fiber orientation distributions (iFOD2) method (Tournier et al., 2010) was used to estimate the streamline tracts. For this, the MRtrix function tckgen, was used to generate \(10^{7}\) streamlines with a maximum length of 250 mm and the fiber orientation density amplitude cut-off set at 0.6.

Structural connectivity estimation

These streamlines were then warped into the MNI152 space using ANTs (Avants et al., 2009) image registration described here. The structural connectivity matrix was then calculated for the warped streamlines in MNI space for 400 parcels of the Schaefer atlas (Schaefer et al., 2018) using tck2connectome from MRtrix. Each value in this connectivity matrix was the sum of the contribution (SIFT2 weights (Smith et al., 2015) calculated using :code: tcksift2) of each streamline (between any two given parcels) to the overall fiber orientation density and was normalized by the volume of the two parcels (using parameter -scale_invnondevol with tck2connectome).