Hi, Clément
Re. your #1: Ummmm, what was the the thing that helped with runtime? In my "#1", I just said you should *not* need to resample the data... I didn't actually suggest anything there. (Was it checking about OMP_NUM_THREADS, which I mentioned below?)
Re. your #2: OK, great, you should definitely be able to take advantage of that number of CPUs here for the application of the NL warps.
Re. your #3: Oooooh, right, yes.... wow, sorry. Indeed, mALFF is scaled by the whole brain total. To be honest, I don't know that mALFF would be the first thing I would pick for comparisons. I think I put it there because someone asked me about it, but is there any reason that is particularly meaningful to scale like that? I would have thought fALFF might be the most useful. Note that perhaps you could include the "scale" block in your afni_proc.py processing, so that the time series are scaled to have the interpretation of "local % BOLD change" by voxelwise scaling. This is typically recommended/used in task studies for benefitting comparisons and interpretation (see here: [
pubmed.ncbi.nlm.nih.gov]), but I think it could also be useful in RS-FMRI studies helpfully, too.
Note that the "-mask_epi_anat yes" option should help make a more reasonable mask. Though, if your datasets have varied coverage across your group (some not whole brain, or some missing subcortex, for example, then you might get inherent variability with mALFF, anyways).
Re. your #4b: OK, as long as you have a good reason for it. Because bandpassing really uses up a lot of degrees of freedom in the data.
Re. more about the degree of freedom usage and papers about perhaps not needing it, see this section of the AP help:
[
afni.nimh.nih.gov]
--pt
