coverage were acquired. For optimal use in layer-fMRI (https://layerfmri.com/mp2rage), background noise (outside the brain) was minimized with a regularization term lambda = 0.1, fold-over-artifacts were minimized by using an excitation pulse BWTP of 25, and no partial Fourier was applied in the second phase encoding direction. Physiological traces of respiration and heartbeat were recorded for the first 12 participants that underwent the visual task. Further recordings were discontinued because in the thermal noise-dominated regime of 0.8mm voxel size here, we did not observe any improvements by conducting RETROICOR (Li et al. 2000) physiological noise correction (Hall et al. 2017). All fMRI images were motion corrected using SPM12 (UCL, UK) (Penny et al. 2007). Volume realignment and interpolation were performed with a 4th-order spline. In order to minimize effects of variable distortions (non-rigid motion), the motion was estimated in a manually drawn mask. The 8 outermost slices were excluded from the analysis to minimize any residual 3D-EPI related slab fold-over artifacts and reslicing artifacts, when necessary. Raw VASO time series usually consist of interleaved acquisitions of MR signal with and without blood nulling. Thus, the analysis pipeline treats odd and even time points separately for the motion correction and subsequently divides
