Figure 1. Overview of CSF Pseudo-diffusion Spatial Statistics ($\text{C}\mathrm{\Psi }\text{SS}$).

At low b-values, CSF retains it signal and signal changes are sensitive to incoherent flow. For $\text{C}\mathrm{\Psi }\text{SS}$, the following steps were performed: (1) Diffusion tensor model was applied to intermediate/high b-value images to obtain FA and mean diffusivity. These were used for 2-tissue segmentation (Atropos, part of ANTs) to generate CSF partial volume maps and pseudo-T1-weighted images. (2) Low b-value diffusion-weighted images (multi-low b-value subset: b: 0–350 s/mm² [7 non-b0 volumes]; b100 subset: b = 100 s/mm², 16 directions, 3 b0 volumes) were used to generate low b-value $\text{M}\mathrm{\Psi }$ maps, measuring pseudorandom flow magnitude. (3) $\text{M}\mathrm{\Psi }$ and pseudo-T1 images were utilized to create study-specific templates with antsMultivariateTemplateConstruction2.sh (ANTs). Subsequently, $\text{M}\mathrm{\Psi }$ and CSF partial volume maps were registered to the template space. A final CSF mask was generated by retaining voxels with CSF fraction > 0.7 in more than 65% subjects. Voxels with a CSF fraction < 0.7 were filled using surrounding voxel average. Spatial smoothing (3 mm FWHM) was applied within the CSF mask using 3dBlurInMask (part of AFNI).

Abbreviations: CSF, cerebrospinal fluid; FA, fractional anisotropy; FWHM, full width half maximum; $\text{M}\mathrm{\Psi }$, mean pseudo-diffusivity.