[ENH] Derive masks to support native-space data

[tsalo]

References #2 and stems from Brainhack-Donostia/ica-aroma-org#49.

Changes proposed in this pull request:

• Add function to derive masks from EPI data and CSF tissue probability map or mask.
• Have the ICA function return the relevant ICA outputs instead of inferring the filenames in the main workflow function.
• Use the new derived masks throughout the workflow.

[eurunuela]

Thank you @tsalo ! I tried running the integration tests and it fails.

I think it's related to nibabel. Could you check it out please?

[tsalo]

I didn't get any errors in the workflow itself, but it looks like the masks are different enough that the ICA produces different components, which is causing the test to fail.

[oesteban]

I tried running the integration tests and it fails.

You guys mean locally right? Because the integration tests on Circle don't seem to have ever worked, correct?

[tsalo]

Right. The integration tests are going to fail on CircleCI until we get rid of FSL completely- unless we set up a Dockerfile with FSL installed, which we haven't done because we ultimately want to drop the dependency anyway.

[oesteban]

I think this is a good idea. Please check #29 which will make these kinds of refactors easier to check for regressions

[eurunuela]

That's a weird error you got here. Is it somehow related to nilearn's load_niimg() function?