Hi ptaylor,

Thanks once again for the detailed responses. It's been super helpful.

Re 1) Sorry but I am just now coming in to take over the project at the cluster forming stage, so I'm not sure the spatial resolution the output data has. Is there code I can run to check?

Re 4) My (perhaps oversimplified) takeaway from this point was that if you go 'mask then blur', it's reasonable to use intersection, and if you go 'blur then mask', it's reasonable to use 70% overlap. Our processing pipeline was the latter scenario, so I've created masks using 70% overlap now which certainly includes fewer voxels than 'union' did (it is 1,700,999 voxels now, to be exact) and therefore the analysis now runs a bit quicker. However, your exclamation at the number of voxels in the mask makes me wonder if something went wrong in an earlier stage of pre-processing. Are there any earlier stages I should be checking for errors perhaps?

Re 5) Okay great, thank you for all of these resources! In Figure 3 of the paper you linked, it talks about a conversion process when running two one-sided tests instead of a single bi-sided test. Am I correct in assuming that if one were to run a single bi-sided test (e.g., using the 'TTnew.CSimA.NN2_bisided' cluster simulation table output file) then no conversion is needed? Meaning, one can use the table as-is?

Re 6) Perfect, this was exactly what I needed to form the images! That paper you linked was also very interesting (Figure 1 and Box 1 were especially illuminating). I'll certainly consider featuring images with transparent thresholding for the sake of reproducibility. I liked the approach of Figure 1 where you can show a 'normal', easy-to-read activation pattern, and another set of images that show all the data. Thanks for the suggestion!