> I am trying to determine the proper threshold to exclude non-gray matter voxels
> from the analysis. It seems that the default 0.8 threshold may be too liberal
> and may be including too many white matter voxels, in turn inflating the search
> volume and decreasing the statistical power. On the other hand, I don't want to
> make the threshold too stringent and exclude critical areas that have relatively
> low counts (e.g., amygdala).
>
> Is it correct that I could use the Check Registration function to compare a
> voxel in a specific subject's image with the mask.img that is created during the
> analysis to confirm whether that voxel in a particular region was included or
> excluded from the analysis ?
Dear sg,
Yes, the checkreg function should show you equivalent voxels.
However, it might be dangerous to change the gray matter threshold to do the
thing that you want. I find the 'gray matter' term rather misleading, as I
have only seen it remove scalp and some ventricle signal at its usual settings.
Increasing the threshold will run a large risk of excluding voxels at the edge
of the brain and ventricles, especially after smoothing. A better idea might be
to use a version of the gray matter mask in the apriori directory as an explicit
mask (easiest done if the images are normalized, and pretty difficult for FMRI).
However, there is not a straightforward relationship between number of voxels
and the stringency of the corrected statistics. Because the surface area of a
gray matter mask will be quite large, the random field corrections may be more
severe than you were expecting - see Keith Worsley's 1996 HBM paper.
Good luck,
Matthew
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|