Dear Alfonso,

Thank you so much for the quick response! I actually have some updates about this question:

The dataset used in the original question was an fMRIprep preprocessed dataset, which I imported into CONN. And there were actually quite a amount of people (nearly 1/4 of the whole group) have been removed out most of the scans in the denoising stage (~150 out of 200). 

However, I imported the exactly same bids dataset (but all nifiti, not fMRIprep preprocessed yet) and ran the whole preprocessing in CONN - the QC turned out fine. Only a couple of invalid scans were removed for most people and only a few movers out of them have more than 40 invalid scans.

I wonder if you have any guesses on why the denoising output were so different? For the fMRIprep I ran the default pipeline without AROMA for func scans, and did not set up any other parameters. I wonder if there might be something wrong with the data importing part.

Thank you so much!
Meishan