If you’re interested, I have some Python code that extracts the WM, CSF, and global signals from AROMA-corrected data (unsmoothed), and then uses 3dTproject to do simultaneous highpass cosine / WM / CSF regression, followed by ROI extraction. See my the script fmriprep2denoised_3dtproject.py in my GitHub repo for this project ( https://github.com/sjburwell/fmriprep_denoising ). I admit, the code is a little kloodgy, so if you have a questions please don’t hesitate to ask or post an issue to the GitHub!