[MED-23][external] Support for export from Raster Layer to NifTI

[tomvars]

Problem

Users of Darwin can currently annotate as masks but not export these results to NifTI

Solution

We create a new code path within the same NifTI export logic. This means that for darwin JSON that contains polygons we will still parse these as before and output a NifTI volume. If the user has K mask classes they will receive K output files called SeriesInstanceUID_classname_m.nii.gz (NOTE: for nifti uploads SeriesInstanceUID is the filename)

Changelog

Introduced initial support for NifTI export from Masks

[linear]

MED-23 Export raster_layer annotations to NIFTI

[Nathanjp91]

is there a reason you use '0' here but XYPLANE above? Are these totally different or is the magic number here meant to be XYPLANE? Same for below

[shernshiou]

Implemented in the new commit

[Nathanjp91]

Are these intrinsinc to nifti and potentially used by other functions? If so can we make this a type

Something like

class Plane(Enum):
    XY = 0
    XZ = 1
    YZ = 2

if view_idx == Plane.XY.value

[shernshiou]

Implemented in the new commit

[Nathanjp91]

It would be great to generalize this function, if only so that it's named and has a docstring. I can't actually tell by inspection what it's meant to be doing too easily

def transform_plane(plane, volume, ...): #or some other more relevant name
    """ docstrings explaining high level so that numpy logic is easier to reason about """
    if plane == XYZ:
        transform_indexes = [np.s_[:], frame_idx, np.s[:]]
    ...
    volume[annotation.annotation_class.name].pixel_array[transform_indexes] = np.logical_or(...)

[Nathanjp91]

and I know this is probably some nifti format specific thing but jumping into the code for these specific data formats from the outside is a lot easier if these sorts of transforms are easier to reason about, with names and descriptions

[JBWilkie]

LGTM - Tested this commit in a deploybot to ensure we can export multiple mask & polygon classes as NifTI volumes