Slice pixel to CCF transformation depends on axes at initialization

[aaronbwong]

I am programmatically using abba-python to convert some coordinates in a slice image to Allen CCF coordinates. I noticed the output seems wrong, and realized it's a strange behavior of getSlicePixToCCFRealTransform() depending on how Abba is being initiated.
Looks like the x_axis of the .abba file was not taken into account. We have been aligning in ABBA with x_axis set to LR, instead of the default RL.

# try default opening
abba = Abba('Adult Mouse Brain - Allen Brain Atlas V3p1')
print(f"X_axis: {abba.x_axis}\nY_axis: {abba.y_axis}\nZ_axis: {abba.z_axis}")
abba_axis_1 = abba.x_axis

The output is

X_axis: RL
Y_axis: SI
Z_axis: AP

abba.state_load(abba_file) # full absolute path needed
print(f"X_axis: {abba.x_axis}\nY_axis: {abba.y_axis}\nZ_axis: {abba.z_axis}")
abba_axis_2 = abba.x_axis

The output is

X_axis: RL
Y_axis: SI
Z_axis: AP

# repeat with x_axis='LR'
abba = Abba('Adult Mouse Brain - Allen Brain Atlas V3p1',x_axis='LR')
print(f"X_axis: {abba.x_axis}\nY_axis: {abba.y_axis}\nZ_axis: {abba.z_axis}")
abba_axis_3 = abba.x_axis

The output is

X_axis: LR
Y_axis: SI
Z_axis: AP

abba.state_load(abba_file) # full absolute path needed
print(f"X_axis: {abba.x_axis}\nY_axis: {abba.y_axis}\nZ_axis: {abba.z_axis}")
abba_axis_4 = abba.x_axis

The output is

X_axis: LR
Y_axis: SI
Z_axis: AP

I tried transforming in both cases with coordInImage [100,100,0].

mp = abba.mp
transform_pix_to_atlas = mp.getSlices().get(0).getSlicePixToCCFRealTransform()
transform_pix_to_atlas.inverse().apply(coordInImage, coordInCCF)
print(f"coordInImage: {coordInImage}\ncoordInCCF: {coordInCCF}")

Here are the results. First run:

coordInImage: [100.0, 100.0, 0.0]
coordInCCF: [7.353574035872652, -0.5013995496068988, -0.1447920757930703]
Second run:

coordInImage: [100.0, 100.0, 0.0]
coordInCCF2: [9.882429330755478, -0.5013995496068988, 11.544792075793072]

[NicoKiaru]

Thanks for the detailed report - I think that the behaviour, while looking weird, is what is supposed to happen.

It's kind of a mess because python uses zyx convention, while Java is xyz, and the Allen Brain Atlas decided to use x for the AP axis (whereas I personnaly think z would make more sense).

Anyway, the last coordinate or your output swaps from -0.1447920757930703 to 11.544792075793072, which looks like the RL to LR switch.

Maybe before digging into this complex reference issue, can you try to re-run the same registration but using the atlas named allen_mouse_10um_java instead of 'Adult Mouse Brain - Allen Brain Atlas V3p1' and see if the results make more sense to you ?

[NicoKiaru]

See this long discussion: maybe there's some useful information:

#21

[aaronbwong]

I had finally have time to get back to this issue. I wanted to run some test as suggested by with allen_mouse_10um_java, but when I run

from abba_python.abba import Abba
abba = Abba('allen_mouse_10um_java')

I got the error from the brainglobe_atlasapi that it is not a valid atlas:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Users\972021\miniconda3\envs\abba-env\lib\site-packages\abba_python\abba.py", line 255, in __init__
    bg_atlas = BrainGlobeAtlas(atlas_name)
  File "C:\Users\972021\miniconda3\envs\abba-env\lib\site-packages\brainglobe_atlasapi\bg_atlas.py", line 91, in __init__
    raise ValueError(f"{atlas_name} is not a valid atlas name!")
ValueError: allen_mouse_10um_java is not a valid atlas name!

My version of the packages:

# Name                    Version                   Build  Channel
brainglobe-atlasapi       2.0.11                   pypi_0    pypi
brainglobe-space          1.0.2                    pypi_0    pypi
abba-python               0.10.1.dev0              pypi_0    pypi

[NicoKiaru]

Can you try updating abba python ? (0.10.6 now)

https://pypi.org/project/abba-python/