The problem with this issue is subjectivity - your criteria for exclusion may be only relevant to your intended use and current state of the data processing tools. This can change in the future, the data can be used for other types of analyses and tools will improve helping to deal with noisy data.
So what I would recommend is the following:
- Add _scans.tsv file for each sub/ses and include a new column in it called qc_exclude (or something similar). See section 8.10
- Add a data dictionary (scans.json) explaining the nature of the new column and the exclusion criteria. See section 4.2
- Explaining the exclusion cryteria once again in the README.
I hope this helps!