Hi Woogul,
First: Yes, the FWHM can be smaller than a voxel dimension.
The 3-D gaussian curve determines how much neighboring voxels
will contribute to the smoothed center voxel, and even though
the curve has gone below half-max in the 5mm direction, those
neighbors will still contribute (just less than in the other
major directions).
It does not seem correct to me to use either a different
smoothing level or voxel size in the AlphaSim computation.
---
Second: Yes, using just "-nx -ny ..." means not using any mask
at all, which is not a good idea.
When running AlphaSim, you pay an unnecessary penalty if you
do not provide at least a brain mask, because the program is
allowed to cluster outside the brain, and since that will lead
to finding more clusters, the estimated probability of having
a "false positive" cluster will be higher than it should be.
If you are working in Talairach space, consider creating a
mask from the TT_N27+tlrc dataset (resampled to match yours).
Note that you should run AlphaSim at the voxel resolution that
you wish to apply the results to. If your datasets are 3x3x3
in +tlrc space, run AlphaSim at that resolution.
- rick