Hälsningar, Paul-
The smoothness estimation of relevance here is the spatial scale of noise autocorrelation. For FMRI data, this typically means looking at how the autocorrelation of FMRI time series residuals drops off with distance (inherently, this estimation ignores directionality, so it is performed as a spherically symmetric operation, for better or worse).
It is hard to see how this could be done with anything other than modeling residuals from FMRI time series.
The perplexing thing for resting state FMRI is: the residuals are *also* the signal of interest! So this creates quite a conundrum IMHO, and I don't know why more people don't chat about this... At the moment, I think people just still use the residuals for both spatial noise ACF estimation (as a precursor to cluster size estimation) and then use that same data in their analyses.
--pt
