I’m trying to figure out which ICA components to pick as regressors for fMRI analysis. fMRIPrep spit out a ton of them and different for each subject/runs, and I’m pretty sure I don’t need to use all. But then, I am also not sure which ones do I keep?
(My dream would be to keep same number of components for all subjects)
I’m also planning to throw in 6 head motion parameters, and lateral and 4th ventricles timeseries as extra regressors.
Would love to hear your thoughts or any tips you’ve got on this topic!
If you ran AROMA then it should have classified the ICA components as noise or signal for you. If you want to manually classify components on top of that, then I would recommend following the instructions in Griffanti et al. (2017).
Hi tsalo,
Thanks for the reply. Do you think all the components of AROMA classified as noise should be selected as regressors in the first level analysis?
That’s generally how people use the AROMA regressors, I believe. I don’t think you’ll want to include motion parameters as well though, since AROMA classifies components based on their correlation with those motion parameters, so the signals are probably redundant.
Thanks!
I am also considering to correlate Aroma noise components with acompcor components and then select only which are highly correlated. I realised there are many strategies but the time is limited
Thanks again for your time!
