For resting-state analyses, the consensus seems to be:
Pre-process the data. Extract the seed time course, get the motion
correction terms,
and enter these into the stats model, and run feat on the filtered-func-data
afer orthogonalizing the EVs.
How about the following? Is this legitimate?
From the original dataset, Enter a blank EV and add motion to model, and
from the resulting residuals (res4d), extract the timecourse of the seed,
and use res4d in feat. Then there is no need to enter motion correction as
EVs, since it has been removed already.
I ask because I get differing results for each approach, mostly in the level
of significnace.
-BRad
|