[WIP] ENH: Load/save linear transform array (LTA) files

[effigies]

This is a first pass at supporting loading and saving of LTA files. This is a pretty flexible structure that encodes the geometry (equivalent to the VOX-RAS affine) of the source and target spaces, and an affine transform between those spaces.

I'm using LabeledWrapStruct essentially as a way to enforce numpy typing on a dict. The VolumeGeometry type seems like it may recur a bit in FreeSurfer, as it shows up in geometry files, as well.

The end goal is to support conversion between different affine formats. Because I'm working with the three formats regularly, I'll aim for LTA (.lta), FSL (.mat) and ITK (.txt) formats first.

A not-incidental secondary goal will be to document these different formats, as most of their existing documentation is the parsing/emitting code in the various packages that support these formats.

TODO:

• Support conversion between LTA types (VOX_TO_VOX and RAS_TO_RAS, at minimum)
• Support concatenation of LTA transforms

[coveralls]

Coverage decreased (-0.4%) to 95.882% when pulling b8844ec on effigies:enh/linear_transform_array into b7eee37 on nipy:master.

[codecov-io]

Codecov Report

Merging #565 into master will decrease coverage by 0.70%.
The diff coverage is 26.11%.

@@            Coverage Diff             @@
##           master     #565      +/-   ##
==========================================
- Coverage   91.85%   91.15%   -0.71%     
==========================================
  Files          97       98       +1     
  Lines       12357    12490     +133     
  Branches     2177     2190      +13     
==========================================
+ Hits        11351    11385      +34     
- Misses        675      774      +99     
  Partials      331      331              

Impacted Files	Coverage Δ
nibabel/freesurfer/lta.py	25.56% <25.56%> (ø)
nibabel/wrapstruct.py	97.01% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2f86a4a...d5f4a08. Read the comment docs.

[matthew-brett]

Chris - this may not be relevant - but there's some stuff on the FSL mat format here : https://github.com/nipy/dipy/blob/master/dipy/external/fsl.py#L69

[coveralls]

Coverage decreased (-0.4%) to 95.875% when pulling 7a3f71a on effigies:enh/linear_transform_array into 2817d1b on nipy:master.

[matthew-brett]

Chris - where are we with this one?

[effigies]

The status here is that it's stalled. Now I have a better understanding of the different affine conventions (i.e. FSL using a voxel-based origin, ITK ignoring sforms) the urgency of working through all of this has fallen away a little bit.

I'd be up for trying to make a roadmap for handling affine transforms, but I think that's going to involve settling on a common internal representation. I actually like the LTA approach, which includes both the source and target spaces, which means that this information is sufficient to produce an FSL or ITK matrix.

But #567 is my highest priority when I have time, followed by #550.

[oesteban]

@effigies do you want to move this one to poldracklab/nitransforms until that is finished and we pull everything together in nibabel's proper?

[pep8speaks]

Hello @effigies, Thank you for updating!

• In the file nibabel/freesurfer/lta.py:

Line 82:9: E722 do not use bare 'except'

To test for issues locally, pip install flake8 and then run flake8 nibabel.

[effigies]

This should be superseded by nitransforms.