Different cluster size for positive zscores

AFNI version info (afni -ver): Version AFNI_23.2.08 'Marcus Didius Severus Julianus'


I was trying to get the voxels which were active in one condition vs all. For that I got the beta coefficients from the bold signals, blurred those beta coefficients and ran a 3dttest across participants using the blurred beta coefficients. I used the -Clustsim flag in 3dtest++ for cluster correction.

I noticed something odd and wanted to check why that might be happening. After running the 3dttest++ command I looked into the ttest_results.CSimA.NN1_bisided.1D to get the cluter size for p=0.005 and alpha=0.05. It showed the cluster size as 26 in the table. But when I got to the GUI and select only the postive voxels (Pos? option), the report after clusterize shows the cluster size as 35 for alpha=0.05. I am not sure why the cluster size change if I select only the positive voxels.

I am also attaching the image with the Pos on.

The 3dtest++ command that I run:

3dttest++ -ClustSim -prefix ttest.results/ttest_results -mask ../../subjects_data/Functional_data/Whole_brain_mask_stan_re+tlrc -setA blur.stats_s103+tlrc blur.stats_s105+tlrc blur.stats_s107+tlrc blur.stats_s108+tlrc blur.stats_s109+tlrc blur.stats_s110+tlrc blur.stats_s112+tlrc blur.stats_s113+tlrc blur.stats_s114+tlrc blur.stats_s115+tlrc blur.stats_s116+tlrc blur.stats_s117+tlrc blur.stats_s118+tlrc blur.stats_s119+tlrc blur.stats_s120+tlrc blur.stats_s121+tlrc blur.stats_s122+tlrc blur.stats_s123+tlrc blur.stats_s126+tlrc blur.stats_s128+tlrc


Hmm, if you are really doing "bisided" or "2sided" thresholding, then you should not have the "Pos" button on. Turning on the "Pos" means you are shifting into "1sided" mode. If you de-select Pos, I believe the Cluster Report window will show that "0.05 -> 26", in line with the *.1D file.


Hello, thanks for the reply! I am selecting the option for bisided and yes, when I de-select the Pos, the cluster report window shows "0.05 -> 26" which is inline with the *.1D file. Though I was wondering how will I get just the positive clusters? From my understanding, bisided option clusters the positive and negative clusters separately, but how would I get a mask of the clusters with positive z-scores. Should I use 1-sided threshold for the Pos option?


Well, personally, I would show all (=both positive and negative) clusters, because those are what are found at the given significance and generally appropriate 2-sided testing. You can use a colorbar to visually distinguish positive from negative (e.g., with hot and cold coloration).

To only show the positive you can still use the cluster size parameter calculated with your chosen NN, sidedness, p-value and alpha-value; in your case, this was 26 voxels. You can threshold with Pos, and just know that those are the positive parts of the full testing.

One additional and important note is that, actually, results are really much more than just the small amount of stuff that survives (multiple layers of) thresholding. More full results visualization and results reporting greatly improves understanding and intepretation, as well as reproducibility and meta-analyses. We made the case for it here:

... and I think it would be a major improvement for neuroimaging as a whole to move in this direction. You can still apply clustering and transparent thresholding, as noted above. This would likely be easier and more understandable with both positive- and negative-sided results shown simultaneously. It's easy enough to pick out the positive vs negative with coloration (again, hot vs cold). However, you could also still turn on the "alpha" and "boxed" visualization for just the positive and negative results simultaneously.

We now also have the gen_cluster_table program to help make the kind of overlap reports that are an improvement on peak voxel reporting---the case made here in the above paper.