Dear Mauro,

thank you for your interest in AROMA. I hope it works well for your data.

1) No, AROMA is intended to not need motion censoring. If you want to do motion censoring, there is no need for AROMA. See the ICA-AROMA evaluation paper.

2) yes, all the adverse effects associated with motion censoring, i.e., loss in degrees of freedom, problems with temporal auto-correlation. AROMA will not influence motion censoring as the frames to censor are determined on the original motion traces on which AROMA has no influence at all.

3) I’m not sure what you refer to with mirco- and macro-motion effects? Various evaluation papers have shown that AROMA does a great job in removing any correlation between head motion and long-distance connectivity. Likewise, given that AROMA operates at the voxel-level, head-motion related effects will be diminished in individual voxels, not just at the brain network level.

4) AROMA labels components it extracted, it does not label resting state data. That said, it is indeed possible that quite a lot of components are labeled as head-motion related (75% is not uncommon). If it’s less in your dataset, great! That means you either have data with little motion collected or you have some other very had artefact going on that is not related to head-motion, I hope for you the first is true :)

5) If you want to be absolutely sure that you’re regressing out any variance related to the head motion. The drawback is that you assume that there is no relationship whatsoever between head-motion and the rest of your data. This is usually not the case/what you want. In other words, if you apply aggressive denoising chances are higher that you will remove a certain proportion of good variance. That is why our default is non-aggressive. This is discussed in the ICA-AROMA methods paper.

Hope this helps,

Cheers,

Maarten

On Thu, Aug 24, 2017 at 4:34 PM, Mauricio Delgado <[log in to unmask]> wrote:

Dear FSL experts,

We are running resting-state functional connectivity analyses, in which we employ ICA AROMA to remove motion artefacts, and have some questions related to this.

1) We are wondering whether AROMA in combination with some basic preprocessing is sufficient to deal with motion-related effects, or whether one should follow up AROMA and temporal filtering with fsl motion outliers tool to identify remaining corrupted timepoints, and ignore those during subject-level statistics (AKA motion censoring).

2) Would there be any adverse effects anticipated if AROMA was followed up by motion censoring with fsl motion outliers (DVARS or FD metric)?

3) Does AROMA deal equally well with both micro- and macro-motion effects?

4) On average, what percentage of resting state data is commonly labeled as noise by AROMA? In one post on the forum I saw 75%, however, in our dataset this is more close to 35-50%. Does this make sense?

5) In which situations would one opt to implement the “aggressive denoising” option in AROMA? And what are the potential drawbacks?

Thanks!

Best,
Mauro

Maarten Mennes, Ph.D.
Senior Researcher
Donders Institute for Brain, Cognition and Behaviour
Radboud University Nijmegen
Nijmegen
The Netherlands

Google Scholar Author Link