Hi Federico,
> In SPM 5 when I specify model at the 1st level I can select an explicit
> mask. I select smwc1_subxx.img as explicit mask. I obtain this image with
> segmentation (mwc1_subxx.img) and after I smooth it with the same FWHM (8
> mm) as EPI
...
> Following this procedure I expect to model only gray matter activations and
> also I expect to have more statistical power when I apply FWE or FDR
I was reluctant to comment on this, since I have no experience of
functional imaging, but it seems no-one else has, so...
I think it sounds broadly reasonable, certainly FWE correction will be
less harsh with a smaller search volume, FDR probably also, though I
think it's harder to predict. I think how much of a good idea it is
might depend on how good your segmentations look (are they from
separate structural MRI? If so, are they well-aligned with the
normalised EPI?).
An important point to note is that the stats masking uses (mask>0) to
decide if a voxel should be included; the smoothed segmentation will
actually have very small but non-zero values out quite far from the
edge of the original segment (overall kernel size, rather than kernel
FWHM) so your mask will probably be a lot bigger than you expect it to
be. You probably want to first threshold and binarize the
segmentation, at a threshold like 0.05, 0.1, or maybe 0.2. E.g. with
imcalc, something like 'i1>0.1'.
Personally, I'd recommend thresholding at several levels and then
having a look with check reg to compare the mask with the original
segmentation, and with the images you intend to use it on.
Best,
Ged.
|