This does not look like a good idea to me. The BOLD response usually varies across trials, and it's rarely the case when you have a constant response among all the trials. If you model the task response with a presumed response function, that would be another layer of inaccurate modeling. When you remove the task effects from the original signal by subtracting the modeled responses, you can only remove the *average* effects, and what is left still contains a lot of task effects. So it would be very problematic to simply treat the residuals or unmodeled signal as resting or intrinsic state independent of the tasks.
Gang
Edited 1 time(s). Last edit at 02/27/2015 04:03PM by Gang.