Hi, Ruilin-
Re. the correlation matrices-
The images showing the overlap of the atlas regions with the FMRI data are useful to see. The overlap is not great. The salmon/orange region in particular appears to stick out above where the brain data actually is. That means many of the voxels within it are really just containing noise, and averaging them into the ROI mean signal washes it out/introduces noise or artifact. When you intersect that atlas map with the brain mask, much of the atlas map will go away, restricting the voxels more. I think something in some alignment step is not looking good. Is that atlas really in the same space as the FMRI data?
For fat_mat_tableize.py, hmm, I have not looked at that program for a long time. I think you would be better off starting with the *.netcc files directly, currently. Those are just text files you can open up.
--pt