add nibabel conversion (lacey import to prevent forced dependency)

[robert-graf]

Adding back the convert to and from nibabel, without forcing nibabel as a dependency if never used.

There was a lot of renaming going on from 0.4 to 0.5. I would like to have the same naming scheme as in the older version for backward compatibility, but I kept it the 0.5 way as I do not know why things are renamed.

nifti_to_ants works without the image decorating. So, I also did not use the add-to-image decorator.

See:
#693 #64 #52

[cookpa]

The new name is a bit confusing but I think the intent was to convey that they will convert any NIFTI image, not just nibabel. I guess the key functionality is the ability to translate an image object with an RAS transform to an ANTsImage with an LPI transform.

[cookpa]

Then again, while the methodology is generic, the implementation does seem to rely on having nibabel Nifti1Image objects. So I would agree the function naming of nibabel_to_ants makes most sense.

Other packages requiring RAS coordinates could call get_ras_affine_from_ants and get the voxel data via numpy.

[robert-graf]

I missed a ants_mni.to_nibabel(), which used the 0.4 version.

And then, I noticed the given nifti_to_ants were not invariant.

The get_ras_affine_from_ants form Torchio assumes that the axis is mirrored and mirrors the original as well.
The q-form does not do this; rotating the image with ants to the same orientation shows this is not invariant.

Maybe this Q-form stuff was a red herring, and that's why you had trouble writing this code.

[coveralls]

coverage: 83.789% (-0.9%) from 84.726%
when pulling 7dcc27f on robert-graf:nibable_conversion
into 5f550b8 on ANTsX:master.