Improve component plots

[tsalo]

Closes #1042.

Changes proposed in this pull request:

• Use plot_stat_map instead of identifying and plotting individual slices. This should address the potential orientation issues from Check orientation of brain/acquisition in figures #1042. It will also select the "best" slices for each component independently, so different components will have different slices rendered.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.34%. Comparing base (1f1845b) to head (71992f5).
Report is 41 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1045   +/-   ##
=======================================
  Coverage   89.34%   89.34%           
=======================================
  Files          26       26           
  Lines        3444     3444           
  Branches      622      616    -6     
=======================================
  Hits         3077     3077           
  Misses        213      213           
  Partials      154      154           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[tsalo]

Argh, tight_layout doesn't help the line plots, but using plot_stat_map should fix any orientation issues.

Basically, I just need to fix up the spacing between the different axes in the figure.

[tsalo]

Here's what an example component figure looks like:

[handwerkerd]

This looks much better. One thing I noticed is that the presented slices vary by component. For example, for the three-echo integration test, here are the images for components 15 & 51. The axial slice numbers are different. This is not great because we really want to compare the component maps between components, but I don't see an obvious way to fix this within plot_stat_map. Thoughts?

[tsalo]

plot_stat_map will identify the "best" slices to present (probably the ones with the highest values, given a certain amount of space between slices). I think that would be better for getting a sense of the component's spatial pattern than being able to directly compare across components.

[handwerkerd]

LGTM

So that we remember this change, note that the presented slices may be different for each component in the opening comment for this PR.

[tsalo]

Done!

[dowdlelt]

I'm sad to see the originals go - but happy with similarly oriented brains.

I'm of two minds about the "best slices" approach - because it is nice to flip between components and see how the structure changes in the same area....but I'm convinced enough that it is better to have a better grasp of the individual components.

I think, based on what I've seen, that there is still enough information there to identify good/bad components, and perhaps it can do even better. Not worth holding up the overall fix, and potentially it could even improve that process.