Hello AFNI Rock Stars!
I'd love to ring in the New Year with a deeper understanding of the use of 3dmerge for clustering the results of a group analysis. My group analysis involved running a 2-way ANOVA on the results of 4 healthy control subjects (S1 thru S4) in an overt picture naming task, as part of a pilot study with 3 stroke patients with aphasia. (I know, I know, not enough subjects... had 9 controls and 9 patients in follow-up study). There were 2 types of pictures: nonsense drawings called "pseudo" pictures, and drawings of real objects called "real" pictures. I should mention here that I'm only interested in the Real Pictures. Here's my script:

3dANOVA2 -type 3 -alevels 2 -blevels 4 \
-dset 1 1 S1_Betas+tlrc'[0]' \
-dset 1 2 S2_Betas+tlrc'[0]' \
-dset 1 3 S3_Betas+tlrc'[0]' \
-dset 1 4 S4_Betas+tlrc'[0]' \
-dset 2 1 S1_Betas+tlrc'[1]' \
-dset 2 2 S2_Betas+tlrc'[1]' \
-dset 2 3 S3_Betas+tlrc'[1]' \
-dset 2 4 S4_Betas+tlrc'[1]' \
-fa StimEffect \
-amean 1 Pseudo \
-amean 2 Real \
-bucket ANOVA2_Betas

The resulting maps look great, but they're in 1mm resolution, even though the functional results for each individual subject respected scanning resolution, i.e. 3.75 x 3.75 x 6mm.
For Multiple Testing Correction, I followed Gang's instructions on his page for resampling the Tlrc-ed group mask to original voxel size:

3dresample -dxyz 3.75 3.75 6 -rmode 'NN' -inset Full_Mask_Controls+tlrc -prefix ControlsMask

I obtained the Standard Dev as follows:

3dcalc -a ANOVA2_Betas+tlrc'[4]' -b ANOVA2_Betas+tlrc'[5]' -expr 'a/b' -prefix StdDev2

where sub-brick [4] refers to the mean for Real Pictures and sub-brick [5] to the t-stat for Real Pictures (once again, I only care about the Real Pictures).

Then I resampled it in this way:

3dresample -dxyz 3.75 3.75 6 -rmode 'NN' -inset StdDev2+tlrc -prefix StdDev2_Resampled

Based on the output of these operations, each in original voxel size, spatial correlation was obtained with this script:

3dFWHMx -dset StdDev2_Resampled+tlrc -mask ControlsMask+tlrc -out StdDev2_fwhm.txt

The spatial correlation parameters were then plugged into my AlphaSim script, as such:

AlphaSim -mask ControlsMask+tlrc \
-iter 1000 -rmm 8.1 -pthr .001 -fwhmx 9 -fwhmy 7 -fwhmz 9 -out Controls2_CorrectedValue001.out

At an alpha-level of .058, I get a cluster size of 7. So far, so good, at least I hope!

The final step is where I get really, really confused. If I want a minimum cluster size of 7 voxels with dimensions 3.75 x 3.75 x 6mm, does my rmm need to be at least 6? Here's what I ran:

3dmerge -1thresh 3.3 -1clust 6.1 7 -1dindex 4 -1tindex 5 -prefix Cor001_Controls2 ANOVA2_Betas+tlrc

The resulting file is in 1mm resolution, just like my ANOVA2_Betas+tlrc bucket file. It looks almost *exactly* the same as the output of the same script, with rmm of 1.1 instead of 6.1:

3dmerge -1thresh 3.3 -1clust 1.1 7 -1dindex 4 -1tindex 5 -prefix Cor001_Controls2_1.1 ANOVA2_Betas+tlrc

HUH? Why does the rmm value in the -1clust option seem to make so little difference? Have I done this right? And most importantly, does it make sense to run 3dmaxima on the output maps of my clustering script? Here's how I found almost 200 maxima within the clustered map:

3dmaxima -input Cor001_Controls2+tlrc -thresh 2 -min_dist 7 > Maxima2.1D

Although the results do make sense, I feel like I'm totally gropin' in the dark here, guys. Any advice or suggestions would be enormously appreciated.
Thanks all, and Happy New Year! AFNI blessings to us all in 2008!
-Whitney