Is it possible that the additional slight smoothing introduced by the
spatial normalisation resulted in smaller blobs merging together to give
a bigger blob? Also, the additional smoothing may have increased the
t-scores by a small amount. If there is a diffuse activated area, then
SPM should be more sensitive to identifying it if there is more
smoothing (which I believe relates to the matched-filter theorem).
I would guess that both analyses are correct. It's just that they are
testing the hypothesis at slightly different spatial scales.
These are the only explanations that I can currently think of.
Regards,
-John
| I obtained nearly three times the size of an activated region after
| normalization and couldn't understand why. A similar question was posted
| in Dec. 1999 but I thought I would post another e-mail to ask for help
| because different way of analyzing data may make a difference.
|
| I did statistics on two sets of image files: rsfmri*.img (before
| normalization) and nrsfmri*.img (after normalization). Both were
| processed from the same fMRI data. The original spatial resolution was
| 3.75x3.75x5mm and the smoothing kernal was twice the voxel size. When
| doing normalization, I followed pretty much all the suggested option
| except for the following:
|
| neurological convention
| don't mask brain
| don't mask object
| bounding box: template
| voxel sizes: 4X4X4
|
| I started with transposing the first fMRI volume to the template EPI.img
| and then applied the transformation parameters to all volumes in the
| time series.
|
| No global scaling was done. The same statistical threshold was applied
| on both analyses: P=0.001 (T=3.43) with extent threshold k=20 on
| rsfmri*.img and k=22 on nrsfmri*.img series. The statistics table listed
| smoothness FWHM = [11.9, 11.4, 15.1] mm for rsfmri*.img and FWHM =
| [12.5, 13.2, 16.6] mm for nrsfmri*.img. The voxel sizes only differed by
| 10%: [3.75, 3.75, 5] for rsfmri*.img and [4, 4, 4] for nrsfmri*.img.
|
| In the results, the largest cluster from rsfmri*.img had T=11.65, Z=6.88
| and k=307. The largest cluster from nrsfmri*.img had T=12.18, Z=7.03,
| and k=1167. Both the largest clusters were located on similar anatomic
| structure. It looked like the overall Z score increased in nrsfmri*.img
| and therefore, more voxels were above the same threshold.
|
| My questions are:
| 1) Why was the activated region so much larger in the nrsfmri*.img than
| that in the rsfmri*.img?
|
| 2) Which one is correct?
|
| Your reply and suggestion will be very much appreciated.
|