Hi Katherina,
I hope you don't mind me CCing this back to the list -- it seems
better that way, since I might be talking nonsense, and/or someone
else might have some better ideas. Also, on the rare chance that I
might be saying something both right and useful, the list archives
could then be of some help to others later.
> the save button saves T-values according to the whole brain threshold,
> not the ROI correction.
Okay... I've just checked this and it's the same in SPM5 anyway. Well,
I don't know if it works with SPM2, but I've seen that Matthew Brett's
VolCorr program can output t-thresholds for a given alpha-level
small-volume correction.
http://imaging.mrc-cbu.cam.ac.uk/imaging/VolCorr
You could then use imcalc to threshold the spmT at this level, and
multiply it by the ROI mask, giving a resulting mask of voxels that
you are interested in.
> It's not the intersection of two masks that I need.
> I need exactly what I described in the last part: Take only the voxels
> that were significant in the first contrast, but correct their contrast
> values for the whole volume of the ROI.
Ah, so just using the first (ROI-corrected) contrast to mask the
second (also ROI-corrected; and without changing the correction),
right? I think the second contrast's SVC-corrected p-values will be
independent of the first contrast in that case, so I guess you are
just concerned about knowing which of them was significant for the
first and which wasn't (rather than changing their values).
If you just want to find the set of significant voxels, I think you
could analyse both contrasts separately, with SVC over the ROI you are
interested in, and create masks for each of their significant voxels,
as described above, and *then* take the intersection of those masks.
I.e. the voxels that are significant in both contrasts, when each
contrast is corrected for the ROI.
If you want the p-values for just these voxels, and it's too
time-consuming to look through the lists for the separate contrasts,
then I think some slightly fiddly programming would be necessary...
spm_VOI computes the search volume etc. for the ROI, and also the list
of voxel values and coordinates. You could perhaps adapt spm_VOI to
use a second mask (taken from the results of your first contrast) i.e.
in line 112:
j = find(spm_sample_vol(D, XYZ(1,:), XYZ(2,:), XYZ(3,:),0) > 0);
use D2 instead of D, after something like:
Msk2 = spm_get(1,'.img','Image defining voxels of interest');
D2 = spm_vol(Msk2);
This should leave k (and hence S; and Ps(k) if using FDR) coming from
the original mask, but with j being only the voxels that were
significant in contrast 1. The resel count R will be done using the
ROI mask (D), so that should be correct as well. I think this should
be right, but I haven't tested it at all... Hopefully others on the
list might comment on its validity or lack thereof.
Sorry if I still haven't understood you correctly!
Best,
Ged.
|