| I noticed too this enhanced activation in the analysis of the normalized
| images compared to the analysis of the unnormalized images. I was thinking
| that the reason might be that when you normalize the images you do also an
| additional resampling, i.e. a kind of averaging process of the signal across
| neighboring voxels that practically increases the ratio signal /noise. In
| fact the enhanced activation is present even using uncorrected p-values in
| which the resels count should not be involved. Am I guessing wrong
| (somebody more expert than me would like to comment on this...)?
When you resample a voxel using trilinear interpolation, you are effectively
taking a weighted average of the closest eight voxels in the original image,
where the weights depend on the distance away from these voxels. In the
worst case, when you sample a point right in the middle of these voxels, then
the resulting value is the average of the 8 voxel values. This introduces
smoothing.
If the image noise contains higher spatial frequencies than the true
physiological signal, then smoothing may reduce the noise more than it
reduces the signal. A higher signal to noise ratio should produce higher
t values.
Best regards,
-John
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|