Control points might be helpful in both
cases, but you are right it is more often the case when the wm
is too dark. Another thing you can try is using expert options
for mri_segment to constrain the intensity ranges of wm/gm. Or
you could try using the samseg stream (someone else will have
to help you with this, but it might well do better for your
acquisitions)
This
looks like a helpful solution for the places where
the wm is not identified- we will definitely try it!
We
have one other type of issue which is more prevalent,
that I was not sure if control points will be able to
help with (we have not worked with control points before
so sorry if this is a misinformed question)
The
main issue we're facing is that the final segmentation
we get is too aggressive, that is, in many places too
*many* voxels are categorized as wm and thus we have
lower ability to characterize properties of the gray
matter in the occipital cortex. What we usually do is
clean the wm segmentation to make it less bulky, but now
these edits are not being incorporated in the final
segmentation. We are very confused by this because this
type of edits were easily incorporated in v6, but not
with v7.
From
my understanding- the control points would be helpful in
the opposite case- where not enough voxels are
recognized as wm (such as in the rightmost part of the
attached image)- but not much in the rest of the image
where we can see that the magenta borders (x.orig.nofix)
are 'tighter' around the wm when compared to the borders
after running through recon -wm (teal,
x.white.preaparc).
Is
there a way to leverage the control points for this
cause, or is there something else we can try to make the
recon -wm less stubborn and accept our manual edits?
For
reference, the full command we run is as follows:
recon-all
-autorecon2-wm -autorecon3 -subjid '
freesurferfoldername ' -cc-crs ' num2str(corpus_point)
Attaching
another image - please let me know if the
visibility here is better or if you'd like to see a
different visualization.
Edan @
PrincetonBrainDevLab