Surface reconstruction defects

[MSh]

Hi

I'm attaching the iflated mesh I've obtained which is not ideal as it's got shapes like pin or holes at certain points. I'd like to know how I can correct for these defect? Is there any way to prevent them occuring in the first place, for example during surface reconstruction steps ?

Thanks
MSh

[Jean-Francois Mangin]

There is a surface cleaner somewhere here, but it is not distributed yet. I am afraid the only solution is to
remove by hand the spurious chains of voxels at the origin of the problem. You can also try to
tune the segmentation parameters (white matter/grey matter stats, or pressure inside white matter)

[MSh]

Hi

My impression of what you mean the spurious voxels is for example voxels that are already recognized as white matter but are not connected to voxels correctly recognized as white matter voxels. If so, would labelling the whole image and removing the unconnected voxels give the same result as removing them by hand? Sorry if this question is a bit off the subject and not directly related to Brain Visa but I'm trying to find an alternative way to manual correction.

Thanks

[Jean-Francois Mangin]

No,
the segmentation process is imposing a spherical topology to the white matter surface.
This is at the origin of your problem, sometimes a thread of voxels is connected
in the final result. You can not remove it with simple connectivity analysis.
But tuning the segmentation parameters could work.

[MSh]

With all the possible choices for pressure vlaue (from 0 to 150) the problem still exists and I don't get considerablly improved result. Neither am I sure how to choose spurious voxels.

[Jean-Francois Mangin]

Is it happening with only one dataset or a whole database?

[MSh]

I repeated the whole procedure on another dataset (varying the pressure value from 0 to 150) and again got no improvement.

[Jean-Francois Mangin]

What process do you use to create the surface? Do not use the one called 5-something, unfortunatelly
I never had time to finish it... YOu should use the one active in the complete pipeline.

In case nothing works, you may have problem with bias correction. Let more freedom to the correction
through lowering the field regularization parameter.