AC/PC spotting vs FSL normalisation

[VvnYang]

Hi BrainVISA team, I am planning to do manual classification of each subject's sulcal pattern using the cortical meshes reconstructed by the Morphologist pipeline. Before running Morphologist, I noticed that one can either manually define the AC/PC landmarks or use a T1 image that has been normalized with FSL. I would like to clarify whether these two approaches are strictly equivalent in terms of the resulting cortical meshes. In particular, I am concerned that using an FSL-normalized T1 image might alter the native sulcal geometry and potentially bias subsequent manual sulcal classification, compared to using the AC/PC-based rigid alignment. Could you please clarify the differences between these two options and whether one preserves native cortical geometry more faithfully than the other? Thank you very much in advance!

[denisri]

Hi,
Actually the choice is not exactly the one you describe, although what you say should work also. Normally Morphologist proposes several alternatives to get affine normalization information for a raw, unnormalized (native) T1 image. You may use either the manual AC/PC landmarks, or SPM12 normalization "OldNormalise", or FSL "flirt", or "AimsMIRegister" (a Brainvisa/Aims command). SPM or FSL options do actually make use of these external software (the process will run SPM or FSL) to process a native T1 image. SPM is now shipped with the BrainVisa distribution so should be always available, while FSL has to be installed externally and configured in Brainvisa so that it can be used from BrainVisa.
What you describe is different, if I understand: you propose to run Morphologist on an already normalized image. This should work also for segmentation purpose, but with 3 noticeable issues:

• you will still need to provide a transformation to the MNI or Talairach space to Morphologist in a .trm file, which is not automatic
• the image is resampled, so its quality may be slightly decreased (contrast loss, blurring, due to interpolation).
• all segmentations and measurements will then happen in the normalized space, not in the native space. So all measurements, brain volumes, cortical thicknesses, and sulci measurements will not be the correct native ones.

Morphologist normally processes native space images, and uses normalization information on-the-fly when needed when it needs to use a common orientation.
Morphologist results are mainly in native space, so will not be very sensitive to normalization inaccuracies and differences. But sulci morphometric measurements are provided in both (native and normalized) spaces, and the normalized ones may vary depending on the accuracy and robustness of the normalization method.
The manual AC/PC method was historically the first one, but is not very accurate, since it relies only on internal points and the brain bounding box. Now it is still a fallback when another automatic normalization method fails. Because all known normalization methods sometimes fail, depending on image quality and potential preprocessings which may have already been run on some public datasets (FOV changes, defacing, skull-stripping, aligment in a different orientation, missing orientation in a format conversion, etc).
Denis

[VvnYang]

Hi Denis,
Thank you very much for the clarification, very helpful indeed! If I understood correctly, it would be better if I do normalisation as part of the Morphologist pipeline instead of inputing a normalised image. Regarding the parameters, could I also confirm with you that on the right, I should input an original (native) T1 in the t1mri box, and tick the perform_normalisation box? On the left I guess I should select the parameters as shown in the screenshot (using SPM)? The tutorial online is a bit outdated...

[denisri]

Yes that's it: if you input a pre-normalized image, then the native size and proportions of the subject are lost and we cannot recover them.
So yes we generally use the SPM normalization, which works "sufficiently well", but if you have FSL installed and have setup the FSL paths in Brainvisa preferences, then a FSL normalization variant will appear in Morphologist normalization pipeline. Not sure it is more robust but you can try.
You are totally right, the tutorial is completely outdated, we need to fix it...
In any case you need to use the importation process before Morphologist.

[VvnYang]

Got it, thanks a lot!

[VvnYang]

I just tried the above pipeline and got this error message:

Process spm12 - Old normalise: Estimate & Write - generic started on 2026/03/10 22:25 
in spm12 - Old normalise: Estimate & Write - generic 1 RuntimeError: SPM process has failed, exit code: 127 
RuntimeError RuntimeError: SPM process has failed, exit code: 127

I didn't do anything specific about SPM (since it's included in the distribution), and I'll take a look at my environment to ensure evrything's in place. But meanwhile may I check that in principle, should I download / configure anything before running Morphologist with SPM? Thanks a lot!

[denisri]

Oh. That's not normal, it should work out of the box. Can you show the parameters of the pipeline, and maybe the log outputs (ctrl-L in the main window of BrainVisa, then open the process tree down to the process which had the error) please ?

[VvnYang]

That's good to know, I just solved it by manually installing a (somehow) missing package, and now it's working fine! Thanks again for your help!

[denisri]

There was a missing package ? Then it means that the dependencies of our packages are missing something... Which package was it ?

[VvnYang]

So after I got the exit code: 127 above, I went to check if SPM is working fine. I entered the paths in preference - SPM and tried to launch the SPM standalone on its own.
Then I got error while loading shared libraries: libcrypt.so.1: cannot open shared object file: No such file or directory. So I installed libcrypt. After this I can launch SPM okay and Morphologist worked.
But I'm not sure if it's just me or this is indeed a missing package.

[denisri]

Thanks for your feedback.
OK it appears that the Matlab runtime is making use of this library.
libcrypt.so.1 is available in conda/pixi as the libxcrypt1 package, so you can try also by uninstalling it from your system, and adding in inside the brainvisa pixi env:

pixi add libxcrypt1

then try again.
We will add this dependency to the matlab-runtime package.

[denisri]

We have rebuit a matlab-runtime package, so if you type pixi update matlab-runtime-9.7 it should trigger the installation of libxcrypt1.