First, I’d like to point out that QC of the raw data should be performed before fMRIPrep (or any other processing for this matter). That means that the proper workflow is running MRIQC first, doing the QC on the original data. Pre-specify the participants to be excluded for low quality reasons and then run fMRIPrep only on those who survived.
That said, I’ll try to reply to your question. MRIQC generates one visual report per participant, modality, task, run etc. If time permits, ideally you should open up all these reports one-by-one, screen them on your browser and annotate them with a quality rating and potentially noting MR artifacts you might have found. About 1yr ago I started a walk-through guide about the anatomical reports of MRIQC that you might find useful: https://1387-54756129-gh.circle-artifacts.com/1/home/ubuntu/scratch/docs/interpreting.html
Please make sure you are using the latest MRIQC version (i.e. 0.15.0) because I fixed some critical bugs and improved the reports. They come with a handy widget for annotating the quality. Make sure you click the “Download” button (that will get you a json file per image, you can collate all those jsons into one excel table for your records). You may want to click also on “Post to API”, which basically sends your annotations to a database where they are crowdsourced (see https://doi.org/10.1038/s41597-019-0035-4).
If time is limiting you, then there are some lazy approaches using MRIQC “group reports”. That will help you catch outlier images that fall far from the common distribution of your sample. Those are the first images to look at (please have in mind that they could be outliers for good reasons too!).
Some (unfortunately muted) videos showing how that works are presented here: https://mriqc.readthedocs.io/en/stable/reports.html#demo-anatomical-reports