Dear Thierry,

You may find some additional information/confirmation in these posts (http://www.nitrc.org/forum/message.php?m... , http://www.nitrc.org/forum/message.php?m... ) but, briefly, the "effect of rest" included by default during the Denoising steps, is in fact simply only removing from the BOLD timeseries a small trend at the very beginning of each scanning session. It is similar but not the same as the "session effects" modeled in first level SPM analyses. In fact those exact "session effects" regressors (a constant value across all scans in a session) are always included as part of the Denoising regression step (this has the effect of setting the average BOLD signal value from each session to zero after denoising), while the "effect of rest" (a step function convolved with the hemodynamic response function) can be included or not as part of Denoising simply by entering/removing the associated regressor from the 'Confounds' list (and when included it has the additional effect of removing any associated small ramp/trend at the beginning of each session from the BOLD timeseries after denoising).

If you have a purely resting state dataset and you are already confident that there are not ramping effects at the beginning of your sessions it is perfectly fine to remove the "effect of rest" from the 'confounds' list in the Denoising tab (and you should see in the associated plots in the right of that gui that this produces minimal to no effects in the estimated voxel-to-voxel correlation distributions) 

Hope this helps
Alfonso
Originally posted by Thierry Chaminade: