"Training Sets" for Manual QC of MRI Data

Does anyone know of any “Training Sets” with manual QC of MRI data (T1/T2/BOLD)? That is, MRI images that have corresponding quality ratings attached to them.

I was thinking of using something like the database of MRIQC visual reports on openneuro, AWS S3 Explorer, but I can’t find the corresponding expert ratings. I think they might be on the MRIQC Web API: https://mriqc.nimh.nih.gov/ but it seems to be a broken link? Anyways, the idea was to look at those reports, of say, a T1w image, give a rating of “exclude, poor, acceptable, or excellent”, then look at the expert rating to see if I was off.


For the DMCC we use the HCP/CCF structural rating system, as described in the StructuralTrainingSlides.pptx (“CCF Structural QC Training”) made by Erin Reid. We may be able to release the poor-quality structurals for the DMCC55B participants that required repeated acquisitions; I am investigating.

1 Like

We were able to release the T1 and T2 images for the 13 DMCC55B people whose initial structural scans we rated of poorer quality, as well as our ratings for both sets of images (the initial rejected images and better repeated ones). The rejected structural images (with session name “doNotUse”) and ratings are in a new DMCC55B OSF sub-component, [OSF | postPublication_rejectedStructurals], rather than at OpenNeuro with the released dataset, to avoid confusion about which images should be used for processing. Thanks go to Rachel Brough for preparing the images and ratings for release.