Description
Comes from #658, where @satra proposed:
- non-visual qa
[...]- image similarity metrics (coregistration histogram separability), exposing bbregister mincost as a basic cut-off, distribution variance of ratio of group intersection epi mask to individual epi mask
- mostly metrics around coregistration, normalization, and artifactual components.
with the following questions from @effigies:
These two seem related. While mincost is straightforwardly parsed from bbregister, do you have references for the metrics you'd like to use? And to the extent that you have experience with them, do you have thresholds you'd set for flagging, or would you expect to form a distribution across subjects/runs and look for outliers?
We've also been discussing collecting mask volumes (for skull stripping, CSF/WM ROIs, etc.).
Basically, this is something we're actively discussing, and the more concrete the proposal, the more likely we'll implement it or something close to it.
Let's define that concrete proposal here. A somewhat related PR (merged) is #694. Another related issue is the warning you get on the reports when x-forms don't make sense.