Hi Muthun,
Once your threshold gets low enough, and particularly
with rmm = 10 (that means you can have voxels that are
almost 3 voxels distant being considered as part of a
single cluster), most clusters will be close enough to
each other to "merge together".
Look at sub-brick 5 in afni as the overlay dataset
(as both overlay and threshold). Then set the
threshold to the 1.965 value that you use in your
3dclust computation. This shows what data can go into
some cluster. If most of the brain area is "active",
then the main cluster will probably cover a significant
portion of the brain (and so be a bit meaningless).
If you simply want to allow for diagonals, and not to
allow a large separation in the other two directions,
maybe it would be worth using -dxyz=1, rmm=1.415, and
vmul = 5.
The former option tell 3dclust to treat all voxels as
1 mm^3. Then rmm should be just over sqrt(2), and
vmul should be effectively 5 voxels.
This is almost identical to how you got your weird
output, except that voxels are now considered to be
1 mm cubes, as opposed to 3.75x3.75x6 mm boxes.
If this still makes too large a cluster, then you
will want to raise the threshold higher.
- rick