[BUG] "Check orientation" button in napari not consistent with registration results for hemisphere

[pderdeyn]

Describe the bug
Orientation doesn't seem to be consistent between the "Check orientation" button in napari brainreg plugin versus when running the registration. The "left-right" lateral direction seems to be flipped between the two for my useage. Note that I am using a single hemisphere

To Reproduce
I have one hemisphere of an imaged brain to align with an atlas. Using the check orientation button in napari's brain reg plugin, I find that my data orientation should be lpi for a brain geometry of Left hemisphere (screenshot 1). However, when I run this hemisphere through brain reg, I find that the atlas is actually the incorrect lateral orientation and left and right need to be flipped (screenshot 2). I'm able to fix my brain my switching to rpi (screenshot 4), even though the "check orientation" results don't look correct anymore (screenshot 3).

Expected behaviour
Registration should occur with the correct orientation if the "Check orientation" results look good

Screenshots

"Check orientation" results for lpi

Registration results for lpi

"Check orientation" results for rpi

Registration results for rpi

Computer used (please complete the following information):

• OS: Windows
• Version: Windows 11, Version 23H2
• Hardware specs: 32 GB RAM.

[adamltyson]

Thanks for raising this @pderdeyn, we'll take a look.

@alessandrofelder I wonder if this is related to what we saw with napari in 2D/3D on Friday? Is it possible that there's something going on with 2D and 3D visualisation in napari, so that in one of the views, the data is (or at least appears to be) flipped?

[alessandrofelder]

I wonder if this is related to what we saw with napari in 2D/3D on Friday?

Maybe... we should find a clear way to check this for ourselves (that doesn't require neuroanatomy knowledge) and then discuss it with the napari community if needed?

[adamltyson]

A quick test on the latest napari suggests that it isn't a napari thing, this looks fine to me:
2D

3D

[adamltyson]

I had a quick look at this last week, and it looks like the issue is somewhere in the intersection of setting the hemisphere and creating the projections.

[adamltyson]

I had a quick look at this last week, and it looks like the issue is somewhere in the intersection of setting the hemisphere and creating the projections.

Actually I think I'm wrong, I think the issue is just in creating the projections. I (very scientifically!) created an image with an L marked into it in two axes, so that you can always tell which axis is which.

This brain is psl and registering it with this orientation appears to work correctly. The downsampled_standard.tiff image is correctly oriented:

This suggests that brainreg and brainglobe-space are working properly. Of course they could be cancelling each other out, but I'm assuming that these two mature packages are correct.

However, if I click on Check orientation then the brain appears flipped (hard to see, but there's a slight mark in the left hemisphere, which is now rendered on the right).

I assume this is the reason for the issue reported here. I haven't worked out why this is happening yet.

[adamltyson]

I think I'm missing something here. In the orientation checking function, the input data (napari layer) is reoriented (using brainglobe-space) to the atlas orientation. This looks correct (L is in the correct place):

However, when looking at the projection images in the above comment, it seems that the first projection (bottom left) is correct (assuming the view is front to back), but the second (bottom middle) is not (assuming the view is top down). I don't think we can say anything about the third (bottom right).

[alessandrofelder]

@IgorTatarnikov and I spent some time diagnosing this.
Our conclusions are:

• there is a bug in the check_orientation code, which hides the left hemisphere when we actually want to show it (and same for right hemisphere)
• there is an additional false assumption by us that the middle view of the "Check orientation" result is top-down. It is actually bottum-up, because that's is the view that results from the anterior-most, right-most pixel of a 2d projection along the S-I axis displayed on the top-left.

[pderdeyn]

Fantastic. Thanks for the fix you all <3

[adamltyson]

Hey @pderdeyn, a new version with this fix should be out in an hour or so. pip install brainreg -U will update.