[DAR-5588][External] Apply automated triaxial re-orientation of non-NifTI annotations for import

[JBWilkie]

Problem

When importing non-NifTI annotations, we do not consider if the target dataset items were medical items with flipped X, Y, or Z axes. Non-NifTI annotations generated for items that were re-oriented (All non LPI files) out of the platform will appear flipped in at least one axis when imported

Solution

This PR:

• 1: Determines the set of axes that were flipped on upload for items targeted by every annotation import. We do this by comparing affine with original_affine in the same way that we already compute transforms for NifTI exports
• 2: Applies the necessary X, Y, and Z axial flips to inbound annotations:
  • X & Y are flipped by subtracting their current coordinates from slot width & height
  • Z is flipped (only for sequential items) by reversing slot frame indices

Raster annotations require alternative handling to flip in X & Y. This requires decoding the dense_rle, flipping the decoded data, then re-encoding it into another dense_rle

We only apply these flips to post-MED_2D_VIEWER files. This PR also takes the opportunity to refactor _get_remote_medical_file_transform_requirements into the simplier _get_remote_file_medical_metadata which returns a single more understandable dictionary of data

This cannot be released until DAR-5589 is ready to merge on the backend

Changelog

Applied re-orientation of non-NifTI annotations upon import to medical dataset slots

[linear]

DAR-5588 Apply automated triaxial re-orientation of non-`NifTI` annotations for import

[dorfmanrobert]

This is true for DICOM, but for Nifti if we are using Nibabel to decide the primary plane, I'm not sure that this is enforced to be true (this was the diff in that other PR on orig implementation vs. nibabel on deciding what hte primary plane was for Niftis)

Rather than

    if primary_plane == "CORONAL":
        axial_flips[0], axial_flips[1], axial_flips[2] = (
            axial_flips[0],
            axial_flips[2],
            axial_flips[1],
        )
    elif primary_plane == "SAGITTAL":
        axial_flips[0], axial_flips[1], axial_flips[2] = (
            axial_flips[2],
            axial_flips[0],
            axial_flips[1],
        )

It might be safer to use the first column values of the nib.orientations.ornt_transform(original_ornt, ornt)

[dorfmanrobert]

Would deduplicate this and remove all the if else statements. One way to do it is to do two fns since we already decide on what axis is used based on an if/else when it is used in _apply_axial_flips_to_annotations:
_flip_annotation_in_x
_flip_annotation_in_y
and those both call a shared helper _flip_annotation_coordinates that just changes to either use key "x" or "y" based on which of _flip_annotation_in_x/_flip_annotation_in_y is calling it

[dorfmanrobert]

Maybe you can put the raster logic as another case you are matching against in flip_annotation_coords/flip_annotation_coords helper rather than having a separate method? and in that case you call the logic from another helper to declutter as well.

[dorfmanrobert]

If take approach suggested above, can remove

[dorfmanrobert]

This should maybe be a method of Annotation as well?

[JBWilkie]

I'll make it a method of VideoAnnotation

[dorfmanrobert]

Should we call these axial_flips still? It might also make it clearer if it was some kind of mapping

{
"Coronal": -1,
"Sagittal": 1,
"Axial": -1
}

so that its use in _apply_axial_flips_to_annotations is clearer than axial_flips[index], which is hard to read

[dorfmanrobert]

The logic below

1. Should probably be deduplicated out into helper functions

2. If the changes suggested for _flip_annotation_in_x_or_y and _flip_raster_layer_in_x_or_y are taken, then can remove a lot of the if elses here

3. If the changes suggested for the axial_flips renaming/changed structure are taken, should reflect here

4. Can we also loop over frame_idx, frame_annotation once? aren't the flips order independent?