[DAR-4932][External] correct pixdim application in mask to polygon

[dorfmanrobert]

Problem

When pixdims[0] ≠ pixdims[1], 3D mask annotations are incorrectly scaled when imported as polygons from a NifTI file. This is due to applying the pixdims to the incorrect axes.

Solution

The key thing to note is that in the 2D mask slice arrays:

• rows are indexed top to bottom
• columns are indexed left to right

But in visualisation space, the columns represent the x axis and the rows represent the y axis, which is why we switch x and y. Now, since the pixdims entries correspond to the axes of visualisation, not array index axes, we need to switch the application of the pixdims.

Changelog

Correct import of 3D NifTI masks to polygons when pixdims vary across axes.

[linear]

DAR-4932 Correct NifTI polygon import scaling

[dorfmanrobert]

The pixdims are (0.7986109, 0.798611, 6.0000024), so this test covers the code changes as pixdims[0]!=pixdims[1]. Due to the fix, the expected outputs needed to be recreated.

[dorfmanrobert]

This is the real change in logic.

[dorfmanrobert]

These changes to process_nifti are unrelated and make the function more explicitly able to handle arbitrarily oriented outputs, even though this is not needed for importing to a fixed orientation of LPI.

Before, the pixdims were taken from volume reoriented to RAS, meaning the output orientation needed to be in RAS order as well for the pixdims to be ordered correctly. Now the output's axes can be any order.