Use only subset of regions within an atlas

[PeerHerholz]

Is your feature request related to a problem? Please describe.
Given the work on a few of my current projects within which I want to use NiBetaSeries, I was wondering if the usage of only a subset of regions within an atlas should be supported in any way? Let's say I want to compute a correlation between certain regions but not on the whole-brain level: how should that be supported/implemented? Should this be a part of NiBetaSeries or conducted with other tools after NiBetaSeries was run?

Describe the solution you'd like
One option, which would potentially reduce computation time, would be to only compute correlations between a user-defined subset of regions within an atlas. This would however include the necessity of knowing which regions are included in a certain atlas.

Describe alternatives you've considered
Another option would be to run NiBetaSeries as is and implement a small function that only returns the correlation of a user-defined subset of regions or easily "extracts" them from the whole-brain correlation matrix.

What do folks think about this?

[jdkent]

would this be possible by specifying only the regions you wanted in the atlas look up table and pass that into nibetaseries? I suppose I'm not positive about the behavior of nilearn's NiftiLabelMasker if you pass it an image file with more regions than you have defined in your labels file, as it is used in nibetaseries.

[PeerHerholz]

Without having tested it, I would say it's not possible without changing NiftiLabelMasker itself as it operates on the atlas *.nii!? Hence, the atlas image would need to include only the regions one wants to compute correlations between. I think the second option mentioned above, computing on the whole-brain level and then extracting only regions one is interested in, is more straightforward and should be easy to implement using e.g. pandas. With that it could be an additional argument, for example --subset regions 1 2 3 4 5.