fix hemispheres being cut slightly off-centred

[carlocastoldi]

This fixes a small problem with ~symmetric atlases with an even-sized frontal axis to be cut slightly off-centred.

Description

What is this PR

• Bug fix
• Addition of a new feature
• Other

Why is this PR needed?
This produces

What does this PR do?

References

Please reference any existing issues/PRs that relate to this PR.

How has this PR been tested?

import numpy as np
from brainglobe_atlasapi import BrainGlobeAtlas

atlas = BrainGlobeAtlas(atlas_name="allen_mouse_10um")
round(atlas.metadata["shape"][2] / 2)
all(atlas.hemispheres[0,0,571:] == 1)
all(atlas.hemispheres[0,0,:571] == 2)

last two commands should be True only with 570, not with 571 as the splitting axis depth.

Additionally, this is also shown in ABBA, as atlas annotations imported with the auxilium of BrainGlobe are offcentred:

From the image, you can notice that BrainGlobe's (yellow) and ABBA's ASR annotations fully overlap, however BrainGlobe splits the hemispheres slightly on the right.

P.S. the left annotations are from ABBA's RSA atlas. For some reason the transformation shifted it one unit (µm). Dunno the reason, but that's not the point.

this is the result with this PR:

Is this a breaking change?

Not that I am aware of.

Does this PR require an update to the documentation?

No

Checklist:

• The code has been tested locally
• Tests have been added to cover all new functionality (unit & integration)
• The documentation has been updated to reflect any changes
• The code has been formatted with pre-commit

[alessandrofelder]

Thanks @carlocastoldi - looks good at first glance. Will have a slightly deeper look tomorrow.

[alessandrofelder]

I have confirmed the bug in napari on the allen_mouse_10um atlas. Here is a closeup of the midsagittal plane in a coronal section:

One can also be convinced of its presence by examining the values below in the Allen Mouse 10um atlas:

In [48]: atlas.shape[2]
Out[48]: 1140

In [49]: (atlas.hemispheres==2).sum()-(atlas.hemispheres==1).sum()
Out[49]: np.int64(2112000)

In [50]: atlas.shape[0]*atlas.shape[1]*2
Out[50]: 2112000

Which shows that the number of pixels in one hemisphere differs by 2x number of pixels in a sagittal plane from the number of pixels in the other hemisphere. This is despite the number of LR pixels being even (so there should be no difference).

[alessandrofelder]

Thanks a lot for fixing @carlocastoldi 🙌