Transform to MNI space - Visualization

[stavrizinonos]

Hi,
I have segmented some brain MRI images, where i saved the initial MRI with the corresponding brain mask in a Subject object. Then, i tried to visualize them using the interactive plot volume, which seems to work fine. However, when i do transformations, it also works fine, until i transform it to MNI space. Then, i can no longer visualize the images as shown in the 4th screenshot. Why is that happening? The same happens and when i use subject.plot() , i can see the visualization at first but after transformation no. I am wondering how i can set starting points on the plots so i can view it when it is transformed to the new space. Also, if i want to save the stripped image in my subject object as well, as a new parameter, shall i save it as a ScalarImage or LabelMap? I have tried both and i was also not able to visualize the stripped image not even before transformation, but when i try to plot it when it is not in the subject object it appears fine. Why is that happening again?
Visualisation_brain.pdf
Thank you in advance

[fepegar]

Hi, @stavrizinonos. I think your registration was not successful and the image is empty. Look at the RuntimeWarning you get. The code is here. Please take a look at those values you want to visualize.

[stavrizinonos]

Thanks for your reply. And i have tested that, I believe the image is not empty, because when i try different plotting i get something but of course not what i should get, but at least from this i can verify that the image is not empty. Also, the registration was applied to the whole dataset and i noticed that there is nothing wrong when the images come from the IXI dataset, and this problem only appears when the image comes from the FCP dataset. The only difference i saw and it might be related between those 2, is their initial orientation ,as in the IXI images initially had the PSR+ orientation, while the FCP ones come with LAS+ orientation. Is there a possibility this is related with the registration?
Vis2.pdf
This is an example of a non skull-stripped('anonymized') & a skull-stripped images from the FCP dataset:
scan_mprage_skullstripped.nii.gz
scan_mprage_anonymized.nii.gz

[stavrizinonos]

@fepegar Hi, was there any progress regarding this issue? Sorry for messaging again, I am really struggling to understand what is happening and why. Is seems that orientation is not a problem as in when I try different LAS+ orientation images the registration works for some. However, for many images it works perfectly and for many not at all, as the pdf files i attached above show. Does this only happen to me? When trying any of the example images i attached the transformation to MNI work for others?
Thank you!

[fepegar]

Hi. Sorry, I'm quite busy at the moment. Can you please share some code so I can reproduce the problem? Also, it could be great if you could post images and text in your comment, instead of attaching a PDF.

[stavrizinonos]

Yeah of course, sorry!

The images that i used as inputs in this example are :
corr-Beijing_sub11072.nii.gz
corr-AnnArbor_sub00306.nii.gz
skull-stripped-IXI002-Guys-0828-T1.nii.gz

The code after the images are skull stripped and with corrected bias are then fed into images A and B and C and need to be transformed to MNI space as follows:

#3 images
imageA = tio.ScalarImage('skull-stripped-IXI002-Guys-0828-T1.nii.gz')

imageB = tio.ScalarImage('corr-AnnArbor_sub00306.nii.gz')

imageC = tio.ScalarImage('corr-Beijing_sub11072.nii.gz')

#to_RAS function
to_ras = tio.ToCanonical()
imageA_ras = to_ras(imageA)
imageB_ras = to_ras(imageB)
imageC_ras = to_ras(imageC)
#to_mni function
mni = tio.datasets.Colin27()
to_mni = tio.Resample(mni.t1.path)
imageA_mni = to_mni(imageA_ras)
imageB_mni = to_mni(imageB_ras)
imageC_mni = to_mni(imageC_ras)
#plotting
imageA_mni.plot()
imageB_mni.plot()
imageC_mni.plot()
#before
print(imageA)
print(imageB)
print(imageC)
#after
print(imageA_mni)
print(imageB_mni)
print(imageC_mni)

Result is the following:

The 3 plots:

The details of the 3 images before and after transformation:

[romainVala]

Hello
I think that there is a confusion between coregistration to MNI and resampling to MNI.
The tio.Resample transform, is just a resampling ie a re-griding of the data to a given reference grid.

it is independant of a coregistration, which aim to realign 2 volume (event if they are differente grid (size, number of voxel). A affine coregistration can be achieve with traditional software, flirt from fsl, nifti_reg, ants, ect ..., Usually it compute a Affine.txt matrix ( and if needed, a resampled version of the input volume moved to the reference volume (mni template in your case)

looking at your 3 volumes (in mrtrix viewer) one can see that corr-AnnArbor_sub00306.nii.gz is at a different spatial position (almost a FOV away) from the 2 others, I guess this is due to its affine, that is not well set.
So the solution here is to use a coregistration tool, that will compute the affine transformation to apply. If then you add this affine as argument of tio.resample (pre_affine_name) you should then reslice with the correct allignment (== coregistration)

[fepegar]

I agree with @romainVala. Please read carefully the docs for Resample, the Fundamentals Concepts docs from SimpleITK and Coordinate Systems from NiBabel. There is a tool to register to MNI in resseg, you might be interested. We will probably include a transform to register to the MNI space soon (#428).

[fepegar]

Well, it's installed in the second line: "pip install resseg antspyx".

[stavrizinonos]

Thanks both of you very much. Yeah now i get it , i thought 'Resample' will do the registration. I ended up using flirt instead.
Thanks for your help!

[lkshrsch]

how exactly does one use the resseg-mni function? I don't understand where the input data is specified in that line. Thanks!

[romainVala]

hi,
I never used it, but it looks like the input data is the first argument of resseg-mni funtion
python -c "import torchio as tio; print(tio.datasets.FPG().t1.path)" will indeed gives you the path of the image that will be coregister (and resample ) to mni

[fepegar]

@lkshrsch please take a look at https://linuxhint.com/mastering-backticks-linux-bash-scripts/ and https://github.com/fepegar/resseg/blob/ecf256c9ba8593c9fb2f33b184245125eb6cd91b/resseg/cli/resseg_mni.py#L8-L18