There are two p-values which are automatically returned, an uncorrected p-val and a FWE-corrected one. Both of them, if I understood correctly, should be used between 0.95 and 1 in mrview.
In the latter case, assuming that you set a = 0.05 before the commencement of your experiment. For uncorrected p-values there isn’t really any rule or expectation of a priori threshold to have been set.
I’m wondering whether there are other types of correction, perhaps less conservative than FWE, which I could use.
I’ve at times considered the prospect of implementing some form of false discovery rate correction for the various statistical inference commands. But there’s never been an adequately strong motivation to pursue such. So no, there’s not anything currently implemented that is less conservative than FWE. Indeed my efforts have been going in the other direction, making false positive control more stringent rather than less.
Second question is whether there is the possibility to output the results in tabular form, where you see the stats of the peaks and clusters ,similar to the SPM output for fMRI.
The problem here is that the definition of “cluster” in 3D voxel data does not trivially extend to fixel data. In some circumstances you can get away with mapping fixels to voxels, doing a connected-component analysis, and then projecting those cluster definitions to fixels, but as soon as you have fixels in the same / adjacent voxels belonging to different bundles the logic no longer applies. The
fixelfilter command does have a “
connect” filter, which is intended to do something akin to a connected-component analysis but by utilising the fixel-fixel connectivity matrix, but my suspicion is that despite the intention it’s a bit too simplistic to robustly address this need. You could have a play with it regardless.
Related to this, is there a way to identify the peak coordinates, even if the results are not in MNI space?
If you can select a fixel as the local maximum, then you can map this to a voxel, then extract the voxel coordinate using the
maskdump command. The challenge is the data-driven identification of a local maximum; while you could get the global maximum using
mrthreshold -top 1, identifying a local maximum requires knowledge of what defines “local”, which is again non-trivial for fixel data unlike voxel data.