Manual import of external segmentations

[Olivier Coulon]

Hi everybody.
I have some unusal infant data that I would like to put in the morphologist piepeline to get surfaces and possibly sulci. I have segmentations, mask, split brain that have been made by collaborators. So basically I import an image in my database, then use a personnal script to import allsegmentations by hand. But in order to get proper surfaces I still need to run the Bias Correction. Not a problem on principle except that I encouter a weird behavior. When I look at the 'whiteridge' image, it seems that ridges have been extracted only on the left side ('see image attached):



Any clue about why this strange behaviour ?

Olivier

[riviere]

Hi Olivier,
We generally see that when AC/PC is not set or not correct: if I remember, ridges extraction is restricted to a standard bounding box to avoid getting ridges outside the brain.
Have you used the normalization step, or set AC/PC points ?
Denis

[Olivier Coulon]

Thanks for the answer. I have used a manual setting of AC-PC... And I'm confident I set them right. Maybe a problem with the voxel size ? At least it's a direction to work with, so I'll look into that.

Olivier

[riviere]

The voxel size should be taken into account in the .APC file, I don't think it could cause the problem.
So it's not what I thought. However I'm sure I have already seen this problem several times, but I don't remember the context right now... I'll think about it, and if I don't find, we will ask Clara, she's younger so should not have lost all her memory yet
I'm still inclined to think that it's related to coordinates systems...
Denis

[Olivier Coulon]

I tried to run the exact same thing with a denoised version of the image, same APC file. The behaviour is strange, I get the ridges on one side only again, but it is on the other side, see the snapshot below...




Olivier

[Olivier Coulon]

Indeed the difference between the original and denoised file headers is one transformation. The original file has a 'Scanner-based anatomical coordinates' referential and associated transformation, and a 'Coordinates aligned to another file or to anatomical truth' . The denoised version has twice the second referential/transformation. For some reason, SPM has removed the first referential and duplicated the second one... So anyway, the problem might be related to referentials...

[riviere]

weird. Funny, but weird...
I'm surprised it is this 2nd transform which changes anything: in Morphologist, at the moment you have the .APC file setup, everything is handled in the native image space and external transforms are not used, so it should not make a difference.
Unless...
Something that happens sometimes is when you are re-processing an image which already has a referential and transformations in brainvisa: in some situations when you re-click AC/PC, they are taken (or transformed) in the wrong coordinates system (because transformations are automatically loaded in Anatomist by Brainvisa), and it may result in wrong .APC coordinates. It happens typically if you overwrite the input raw T1 MRI with another one with a different size/field of view for instance, or a flipped version. You can check using the viewer for the .APC file. I don't know if this could happen here.
I'm not sure it is the right clue...
Denis

[Olivier Coulon]

No, unfortunately this is not the issue here. Field of view, volume dimensions, voxel size are the same, and when I check the ACPC positions with the viewer they are alright...