I’m looking into ways of denoising data that has been/will be preprocessed using fMRIPrep. I’m aware this topic has been discussed at length in other posts, which has been very helpful, however, I wanted to take this opportunity to consolidate my understanding and gain more specific insights.
Here are the questions I currently have (note, these questions only relate to task-based data where no connectivity analyses will be performed):
From my understanding, if I were to enable ICA-AROMA it would apply motion correction, spatial smoothing and non-aggressive denoising on my functional data?
If the above is the case, does that mean I wouldn’t need to regress out the 6 motion parameters that are given in the confounds.tsv file?
Is it recommended to enable ICA-AROMA in fMRIPrep, are the results comparable to not using it?
Also, as a general question - this was suggested in one of the threads as a way of denoising: https://github.com/arielletambini/denoiser - is it still a work in progress or is it relatively stable now?
Yes. ICA-AROMA is a ICA method to remove motion artifacts. Images are spatially smoothed (6mm FWHM gaussian kernel). The “non-aggressive” artifacts are removed. If you need an unsmoothed image, you can manually denoise a non-aroma output using the AROMA components. Depending on your analysis, these images may not need to be denoised any further and may be ready to analyze.
This is a bit more complicated, and is dependent on your application, hypothesis, and raw data. You should take a look at these papers (Parkes et al., 2018, Mascali et al., 2021) and those cited within.
Final question - you say that a non-aroma output can be denoised with AROMA components, does that mean that when ICA-AROMA is enabled it produces one output with it applied and one without (the one without being the one you manually denoise using AROMA components?)