Hi, Dan-

Thanks for checking out the afni_proc.py QC (APQC) stuff. I am in the process of building a help page about it (and actually, there is a newer version of it in the distribution now, with additional features+images).

I guess you have a task data set-- there will be a basic F-stat calculated for whatever you are modeling, testing the significance of your overall regression model (quoting from the 3dDeconvolve help). THis is not a stat of any of your specific, individual glts or anything-- at some point, we will probably add in that functionality.

This is supposed to give you a sense of your model+data being in unison. It might help trouble shoot very noisy data, or stimulus timing files mistakes (wrong units, wrong file, etc.). The threshold is set to be the 90th %ile of the F-stat distribution of values iwthin the brain-- so you are seeing (within the brain) the top 10% of values in opaque coloration; values below that threshold can still be show, but they have increasing transparency (-> so you don't lose all the information of the sub-threshold parts).

If you have a strong visual+audio set of tasks, for example, you might expect bright clusters in the visual and auditory cortical regions-- that would mean things are likely well for your data+model. If you see lots of speckles, and no real clusters in expected areas, there might be a problem with the stimulus timing file, or there might be lots of motion, or... something else (maybe the subject fell asleep and didn't do the task?). Other plots in the APQC should help discern the nature of the issue. However, note that FMRI dsets are veeery noisy in general, so that might also just be part of hte issue.

Hope that is useful.

--pt