Dear Bruce,
Thanks for your detailed answer!
> it's tough to tell for sure on just a single slice, particularly if
it is the intensity normalized one that
removes a lot of the contrast if it is (as you suspect) doing the
wrong thing. What does the orig look like? We try
pretty hard to avoid including deep, highly myelinated gray matter
inside the white matter surface, but of course if
the contrast is low enough we can fail. If you scroll backwards and
forward a few slices does or look in a different
orientation does it look different/better? Sometimes the surface is
curving sharply and this kind of thing will show
up on a single slice, but actually isn't a significant inaccuracy.
Yes, when scrolling throug the slides it really appears consistent across
several slides (actually the attached screenshots
showed Freeview in coronal, sagittal and axial perspective of the same
region). Here is the orig attached together with the wm,
pial and white.
So how should I treat it?
> As for the dura, that certainly looks like a problem for thickness
estimates. Have you tried using the graph cuts
skull stripping? That is more aggressive than mri_watershed. You can
also play with the watershed parameters to try
to get more dura removed. Finally, if you have a T2-space FLAIR scan
we can use that to properly reposition the
surfaces (or less ideally, a T2-space scan without a FLAIR
inversion).
No, unfortunately there are no other images at disposition.
How would one go with the graph cuts ?
I found "mri_gcut [-110| -mult <filename> |-T <value>] in_filename
out_filename",
but how is it run within recon-all in order to repair (FS version 5.3)?
Thank you so much!
Markus
2014-04-07 14:47 GMT+02:00 Bruce Fischl <fis...@nmr.mgh.harvard.edu>:
Hi Markus
it's tough to tell for sure on just a single slice, particularly if
it is the intensity normalized one that removes
a lot of the contrast if it is (as you suspect) doing the wrong
thing. What does the orig look like? We try pretty
hard to avoid including deep, highly myelinated gray matter inside
the white matter surface, but of course if the
contrast is low enough we can fail. If you scroll backwards and
forward a few slices does or look in a different
orientation does it look different/better? Sometimes the surface is
curving sharply and this kind of thing will show
up on a single slice, but actually isn't a significant inaccuracy.
As for the dura, that certainly looks like a problem for thickness
estimates. Have you tried using the graph cuts
skull stripping? That is more aggressive than mri_watershed. You can
also play with the watershed parameters to try
to get more dura removed. Finally, if you have a T2-space FLAIR scan
we can use that to properly reposition the
surfaces (or less ideally, a T2-space scan without a FLAIR
inversion).
cheers
Bruce
On Mon, 7 Apr 2014, Markus Gschwind wrote:
Dear all,
I would like to show some very frequent and typical cases of
segmentation
problems in my bunch of data of normal subjects (after
recon-all), of which
I couldn't find a description in the
wiki(http://surfer.nmr.mgh.harvard.edu/fswiki/FreeviewGuide/FreeviewWorkingWith
Data/FreeviewEditingaRecon).
I am therefore not sure if these types of error are serious and
will affect
later results like thickness, curvature, GWR, etc.
Here are the examples (c.f. attached screenshots from Freeview)
1) In the primary sensory-motor cortex it frequently happens
that the
"?h.white" line lies several mm within the wm border (c.f.
yellow arrow), I
did not observe this behavior in other places.
>>Would I need to edit that?
2) Sometimes the "?h.pial" line includes also dura or parts of
the skull
(green arrow).
>>Is it necessary to edit that as well?
3) Another question concerns the yellow parts of the wm which
sometimes are
much larger than the ventricles and also include both caudate
nuclei (not
shown).
>>Is this a problem?
Thanks in advance!!
Markus
