Just to update on this issue, in the end, we did find that the fov for the single patient cut off the cerebellum, whereas for controls it was fully present. So in the end, we had to workaround this issue by making a custom brainmask with this part of the cerebellum masked out, and we added a seed in the cerebellum that we would regress as a covariate of nuisance, in order to ensure that this part of the brain would be neither used as a seed nor as a target. This allowed us to get correct results in the end, since the super high z values in this buggy regions were hiding everything else.