fMRIPrep: nonsteady states added back into AROMA non-aggressively denoised outputs

,
1

Thanks very much! This makes sense.

I have come across another issue that baffles me. I have 5 runs for one participant and at least for 2 of them, the *space-MNI152NLin2009cAsym_desc-smoothAROMAnonaggr_bold.nii file has its first three volumes non-smoothed, it looked like the nonsteady volumes are added back after ica-aroma, but not smoothed. Again, this happens for two runs out of the 5 and for several subjects. Why do you think that is?

1 Like
2

@jdkent could you have a look into this?

1 Like
3

Thanks for tagging me Oscar, and good question/observation ak_poorni, I’ll have more time to look into this this evening (1:31pm where I am now) or tomorrow.

-James

4

Thanks @jdkent Please let me know if you need any other information when you are looking at it. I will be online

5

@jdkent any updates?

6

Some more observations: Also, the number of volumes flagged for removal prior to ICA-AROMA varied quite a bit from subject/run (anywhere from 0-6). I noticed in the TSV file that it creates regressors for those volumes - that should you include then in the design matrix for task or seed analysis - will result in removing them. However, this gets confusing for task, where we don’t necessarily add any other regressors. I’m also not sure how this number is determiner, but this seems like a number that the user should specify depending on their pulse sequence and whether or not they have dummy scans etc.

7

The reason they are not smoothed is because the non steady state volumes are removed before smoothing is run. The reason they are not used in the smoothing process is because it may influence the thresholds passed to susan, but since the threshold is based on a median (not a mean), the impact on the parameters being passed to susan should be minor. Ideally, the non-steady state volumes should not influence an analysis, so using the non_steady_state_XX regressors is highly recommended. The non steady volumes are calculated on a per scan basis with the function is_outlier for reference

For task analysis, you do not add any additional regressors (besides the task model?). I believe it is still desirable to include some additional regressors for known confounds, is this not what you are accustomed to doing?

EDIT: questions for action items

  1. If we were to smooth the non-steady state volumes, they would still not be “denoised” and should be excluded from any model, (using the non steady state regressor columns). How would including smoothing change your processing pipeline?

  2. For a number that a user should specify, this is an option that you would like to see in fmriprep? (I feel like there has been discussion on this topic before, but I cannot find reference to it currently).

Best,
James

8

For task analyses, I have been including motion regressors, but since doing ICA-AROMA, I would not do this. I should include CSF/WM, I haven’t been doing this in the past.

  1. If we were to smooth the non-steady state volumes, they would still not be “denoised” and should be excluded from any model, (using the non steady state regressor columns). How would including smoothing change your processing pipeline?
    This was my another question, if I can change FWHM that is applied?

  2. For a number that a user should specify, this is an option that you would like to see in fmriprep? (I feel like there has been discussion on this topic before, but I cannot find reference to it currently).
    Yes, wouldn’t this be ideal? Rather than different number of volumes being excluded for different runs of the same participants? this is probably not a huge deal, but for eg., some runs will have 4 volumes of fixation in the beginning, whereas others might have just 2 (hypothetically). Also, especially if i have already accounted/removed for these dummy scans?

Ideally, it would be super helpful if this is highlighted somewhere explicitly in the fmrireaddocs that these volumes are added back to smoothed denoised data and we need to use nonsteady state regressors to not include them.

Just to confirm my processing - If I run ICA-AROMA, I use the smoothed denoised data, but use non-steady regressors as confound EV

9

Yes, if (and only if) you use the aggressively denoised BOLD runs.

If you use the standard (non-smoothed) BOLD runs, then you would include the nonaggressive regressors.

For both scenarios, using the non-steady regressors is highly recommended.

1 Like