Denoising and dummy scans

1

Hi all,

I’m using the following nuisance regressors to denoise my data: 6 parameter motion and motion outliers. I will be using fsl_glm to do this. From my understanding, the motion outliers column will regress out volumes that are subject to large amounts of movement. Does this mean I will have to edit my timing files to take into account the volumes that were regressed out?

I also need to regress out dummy scans, which I will do using non_steady_state_XX columns. At what stage should this be introduced? Should this be included in my denoising stage above, if so, does this mean I will have to edit my timing files further to take this into account? If I include it in my deonoising stage, would I still need to include the dummy scan regressor in my first and second-level analyses so that parameter estimates are not generated for these scans? Or do I not include them in my denoising step and just include them as regressors in my analyses?

Finally, If I have some participants where non-steady states were correctly identified for some runs but not in others, is it worth rerunning this but including the flag --dummy-scans N on these participants or is the effect on confound regressors only slight?

Apologies for all the questions, but many thanks in advance for your time!

2

Hello,

All of these regressors can be included in your first level design matrix without editing your events timing.

As per your last point, if you know that you want the first N volumes to be excluded for non steady state, you can just manually add those regressors to your design matrix using the same kind of one-hot encoded vectors fMRIprep outputs.

Best,
Steven

3

Hi Steven,

Thanks for your response. As per your last comment, does this mean I could edit the non_steady_state_XX column in the confounds.tsv so only the first 3 volumes are regressed out for all runs and participants?

Many thanks,

4

The column name isn’t important, just as long as you have the one hot encoded vectors for the first three indexes.