Hello,

as far as I understood, the confound "effect of rest" corrects for instabilities at the beginning of the scanning session.

During preprocessing, I removed the first 5 scans of each session. Does it then make sense to still keep "effect of rest" as a confound for denoising or is it advisable to remove it from the confounds list? Is there a common practice?

Best,
Sascha

Hi Sascha,

If explicitly removing the first few scans then you probably do not need the 'effect of rest' confounding effect any longer. This is easy to check, if you still have residual effects at the beginning of your timeseries those should be visible in the Denoising tab plots. For example, look at the global signal timeseries and carpet plots to see if there are any visible effects at the beginning of your scans, and/or look at the distributions of voxel-to-voxel correlations with and without the 'effect of rest' regressor in the confounding-effects list to see if you see any noticeable improvement when adding this regressor. If in doubt I would recommend leaving it in, since it only really takes a couple of degrees of freedom from your design. In general the recommendation in the Denoising tab (which applies to all potential confounding effects) is to try to find the minimal combination of confounds that results in robust voxel-to-voxel histograms; i.e. after denoising theses histograms should look: 1) reasonably centered around 0; and 2) almost identical across subjects and sessions. That often happens just right out-of-the-box with the default settings, but if it does not then you should spend a little time here making sure that your data is as clean as possible before proceeding to the first-level analyses. 

Hope this helps
Alfonso

 Originally posted by Sascha Froelich: