Dear Renzo,

Could you please clarify a bit the procedure? what is the reference for the 45dB units, BOLD std? did you add noise after preprocessing but before denoising? I do not see any obvoius mechanism why higher noise would result in higher sensitivity to group differences in ICC, there could be some changes on the effect of the subject-level dimensionality reduction step, and some changes in the operating range of the functional connectivity values that make up the ICC composite measure, but overall I do not see a clear mechanism here, perhaps learning a bit more about the specifics of your tests (and their motivations) could help?

Best

Alfonso

Originally posted by Renzo Torrecuso:

Dear Alfonso and CONN experts,

I believe this is an important issue for all users.

After I denoised my data I introduced 45 dB of white Gaussian noise to my nifti files.

I then ran a ICC analysis comparing patients versus controls and got an increase in blue and red cluster sizes, therefore more voxels with decrease and increase in ICC.

Would you have any input on why this happens?

Thanks a lot

All best

Renzo