Canine templates coregistration

[frederik-lam]

Hello,

I am trying to coregister two canine atlases:

1. Jhonson - Cortical atlas, labeled on population-averaged template: https://ecommons.cornell.edu/items/e9241402-f9a8-4881-b104-73f6db113541
2. Czeibert - Detailed canine brain label map, labeled on a single subject: https://figshare.com/s/628cbf7d4210271ffe70

I would like to coregister the Czeibert atlas template onto the Johnson atlas template and then apply the transformations for the Czeibert atlas labels.

So far, I manually aligned the Czeibert template onto the Jhonson template and set origing to the anterior commisure using SPM Set Origin. I am attaching an overlap of the two images in MRIcroL. I also verified the overlap in other programs (3D slicer and SPM chekreg), and the images are aligned pretty well, although not perfectly.

Then, I tried registering the two images using antsRegistrationSyNQuick.sh with the following inputs:

-d 3 \
-f .../Jhonson_canine_population_template.nii \
-m .../Czeibert_Brain_template_moved.nii \
-o .../CzeibertToJohnson

However, I am receiving the following errors:

Stage 0
*** Running Euler3DTransform registration ***
Exception caught: 
itk::ExceptionObject (0x600001c4c080)
Location: "unknown" 
File: /Users/lampert.frederik/Documents/antsInstallExample-master/build/staging/include/ITK-5.4/itkMattesMutualInformationImageToImageMetricv4.hxx
Line: 311
Description: ITK ERROR: MattesMutualInformationImageToImageMetricv4(0x141f049f0): All samples map outside moving image buffer. The images do not sufficiently overlap. They need to be initialized to have more overlap before this metric will work. For instance, you can align the image centers by translation.

same for stage 1 and then

Stage 2
*** Running SyN registration (varianceForUpdateField = 3, varianceForTotalField = 0) ***
Exception caught: 
itk::ExceptionObject (0x600001c4c080)
Location: "unknown" 
File: /Users/lampert.frederik/Documents/antsInstallExample-master/build/staging/include/ITK-5.4/itkImageBase.hxx
Line: 79
Description: ITK ERROR: Image(0x141f0f910): Zero-valued spacing is not supported and may result in undefined behavior.
Refusing to change spacing from [1, 1] to [0.0341809, 0]

Do you have any suggestions as to what might be causing this? I am also attaching a full log of the run
errorlog.txt

I also tried switching between the fixed and moved image but with the same results. Moreover, I thought that the fact that one image has a stripped skull whereas the other one doesn't might not disrupt the analysis, but when I previously tried to coregister the Jhonson atlas (with stripped skull) to my image data without skull stripping, it did a fine job.

[cookpa]

I'll try to help, could you please make a runnable example including the files you're using (ie, your preprocessed data)? I'll download and take a look

[frederik-lam]

Sure, I am attaching a zip file
data.zip
Thank you!

[cookpa]

One thing I see is that you are trying to register a brain-extracted to a non brain-extracted image. I would make a mask from the labels and register brain to brain.

[cookpa]

To make a mask:

1. Choose labels - some at the anterior appear to not be in the other template
2. Morphological closing to include gyral CSF
3. Fill in any holes by hand

[frederik-lam]

Hi, thanks for the tips.
I made a mask based on the labels from Czeibert atlas and exported it as a binary mask; see attached files
example_data.zip

.
I tried running the antsRegistrationSyN.sh with the following inputs:
-d 3 \ -f .../Jhonson_canine_population_template.nii \ -m .../Czeibert_Brain_template_moved.nii \ -o .../Test \ -x .../Czeibert_Brain_mask_moved.nii
But once again, I received the following error Description: ITK ERROR: MattesMutualInformationImageToImageMetricv4(0x126f04770): All samples map outside moving image buffer. The images do not sufficiently overlap. They need to be initialized to have more overlap before this metric will work. For instance, you can align the image centers by translation. and the analysis got stucked in the loop.

Moreover, as I mentioned in my previous post, I tried to coregister the Jhonson atlas (brain-extracted) to my image data (non brain-extracted), and it did a fine job, so I don't think the issue is there.

Since these errors occur straight away in the first stages of analysis, I tried running just rigid transform (with and without applying mask), and it fails even there (see attached files TestRigidWarped.nii.gz and TestRigid_woMaskWarped.nii.gz). The affine matrixes of these transformations show strange translations (-61; -674; 351) that appear out of nowhere.

I tried to call the full antsRegistration command (prompt attached in ANTs_call.txt) to see if the issue is not in setting the center of images into center of mass instead of origin, but I ran into some zsh errors. I am not very experienced in using Zshell or in creating bash scripts. I couldn't find step-by-step tutorial on how to create a bash script for antsRegistration and how to run it. Would you be able to help me?

[cookpa]

I think I see the issue now. I think the main problem is negative intensities in the images, which causes the center of mass initialization to fail. I fixed this by using -i Identity, which initializes with an identity transform. Another option is to threshold and then mask out the negative values before registration. Using a mask during registration with -x doesn't help here, because the center of mass is applied over the whole image domain.

[frederik-lam]

Using -i Identity worked! Thanks!

Full call for anybody with the same issue:
antsRegistrationSyN.sh \ -d 3 \ -f .../template.nii \ -m .../moved.nii \ -o .../moved_to_template \ -i Identity