Hi Donna
I've gone back and have re-run segmentation using the peak values you've
suggested. It reduced my handles from 21 to 10. I have more questions
though.
In the manual, it is suggested to not fill ventricles when doing a partial
hemisphere, but this is what you suggested
but I suspect that rather large, unfilled ventricle will still need
spackling.
For that job, I recommend the apply only to visible slice in the
coronal view, and starting from slice 46 (where it starts to open up
around 49,46,28), keep patching all the way up to slice 108, where the
segmentation ends.
Should I just re-run the segmentation with fill ventricles? I have
attempted to manually patch the ventricle and managed to reduce the error
count by three. But how much of the ventricle should I fill? Would you be
able to take a look at my spackle job?
Also, the manual gives a procedure for locating errors using inflated
surfaces. I followed this but it does not seem to work. I scroll to within
the limits of one of the errors. I press p on the slice window, and I get a
red sphere on the surface. I place my cursor on the sphere and press p
again. The slice window updates, but the cross-hairs do not stay within the
limits, where the manual says it should. How reliable is this method?
Sorry for all the questions, especially on a Monday!
marc
----- Original Message -----
From: "Donna Hanlon" <[EMAIL PROTECTED]>
To: "Caret, SureFit, and SuMS software users"
<[EMAIL PROTECTED]>
Sent: Thursday, May 20, 2004 11:35 AM
Subject: Re: [caret-users] surefit correcting residual errors
Hi Marc,
Your first error object (16-22,71-78,3-9) was a tiny handle that was
easy to fix. I always scroll to the slice midway between the limits.
In the horizontal view (z component of the limits), I scrolled to slice
6, then moved my crosshairs to x=19, y=74 (roughly midway between the
limits). Scrolling between horizontal slices 5-7, I could see a gap
that probably needed filling. Swithing to coronal view, I set mask
center at 23,77,6, mask 5; this dilated the gap, but didn't reduce the
handle count. Switching back to horizontal view, I could see more
holes, which were easily filled by SMC 22,75,6 mask 5, dilate. This
reduced the handle count by 1.
The next object was 39-48,94-102, 11-18. On horizontal slice 15, near
x=43, y=98, I could see this object was near the notoriously problematic
insula/hippocampus region. In your case, you have a partial hemisphere,
so fill ventricles leaves you even more "subcortical mess" than normal
to patch.
After 24 patches, I gave up documenting the centers and masks. And I
still hadn't reduced the handle count. In fact, at some points, I had
increased it. But realize that in this area, you're often not so much
handle patching as manually reconstructing the surface where the
intensity of the white matter drops well below the peaks.
In this area, your best bet is to switch to coronal view; scroll to the
posterior extent of the trouble area (in your volume, slice 85 was okay,
but the trouble starts at slice 86 -- well before the limits cited in
the Locate Objects list); and check the apply only to visible slice
checkbox. Then, start patching, slice-by-slice, until you fill the
problem areas. See attached screen capture that shows the large gap in
the segmentation in this region (near the crosshairs).
The problem extended all the way up into slice 104, and as I mentioned
before, your handle count could go up during this kind of major spackle
job.
In the process of doing this, however, I noticed a couple of things:
What looked like cortex was being omitted from the segmentation, and
the segmentation itself looked very biased toward the white matter. As
a result, your surface looked "skinnier" than normal SureFit surfaces
(but very much like Freesurfer white matter surfaces). Your gray matter
peak was set to 82, but my histogram showed 73 better reflected the left
peak. I re-ran the segmentation with this peak, and had only 5 handles
(vs your 21).
Unless you have some reason for using a high gray matter peak, I'd go
back to your original, unpadded intensity volume. In Volume
Preparation, change your gray matter peak to 73; save your settings; and
then re-run SureFit (segmentation and automatic error correction). It
requires far less patching in the insula/hippocampus region, but I
suspect that rather large, unfilled ventricle will still need spackling.
For that job, I recommend the apply only to visible slice in the
coronal view, and starting from slice 46 (where it starts to open up
around 49,46,28), keep patching all the way up to slice 108, where the
segmentation ends.
Out of curiosity, why are you cropping dorsally? If you can crop even
further in the posterior direction, you'll avoid some patching near the
medial wall.
Donna
On 05/20/2004 08:08 AM, Marc Malloy wrote:
Hi Donna
I have uploaded the files.
Set Mask Center 58 43 47
seed [58, 43, 47, 6, 6, 6] -> limits 55 60 40 45 44 49
Mask Extent 55 60 40 45 44 49
Perform Patch 2 erode
seed [58, 43, 47, 6, 6, 6] -> limits 55 60 40 45 44 49
Patch type = erode 55 60 40 45 44 49
Limits 55 60; 40 45; 44 49
0 dilation iters, 1 erosion iters
I could not find this information, but I sent you the Object Limits and
some
information on one of the errrors.
Thanks!!
marc
----- Original Message -----
From: "Donna Hanlon" <[EMAIL PROTECTED]>
To: "Caret, SureFit, and SuMS software users"
<[EMAIL PROTECTED]>
Sent: Wednesday, May 19, 2004 5:14 PM
Subject: Re: [caret-users] surefit correcting residual errors
Hi Marc,
Well, at least the number of handles didn't go up. ;-)
My guess is that while you made changes to your segmented volume, you
didn't completely remove the handle (by either filling or zapping enough
voxels). You don't have to save your changes for update handle count to
consider them; it operates on the pre-save volume in memory -- not
whatever was last saved to disk. And I wasn't just kidding: It's
entirely possible to increase the number of handles while patching, so
if this happens consider using the Undo button.
There is no "Patch and Evaluate" button; on page 29 of the SureFit
users' guide, the Patch and Evaluate items refer to general procedural
steps -- not interface labels. Page 30 shows how to use the
dilate/erode features.
Fixing handles isn't easy. If you'd like, I can have a look at your
volume and segmentation and give you pointers specific to your dataset.
Upload the segmentation, intensity volume, and params file here:
http://pulvinar.wustl.edu/cgi-bin/upload.cgi
You can also send me excerpts of the output that scrolls to the terminal
window from which SureFit was launched. I'm specifically interested in
lines like this, that tell me your mask center, extent, and patch type:
Set Mask Center 58 43 47
seed [58, 43, 47, 6, 6, 6] -> limits 55 60 40 45 44 49
Mask Extent 55 60 40 45 44 49
Perform Patch 2 erode
seed [58, 43, 47, 6, 6, 6] -> limits 55 60 40 45 44 49
Patch type = erode 55 60 40 45 44 49
Limits 55 60; 40 45; 44 49
0 dilation iters, 1 erosion iters
Donna
On 05/19/2004 03:39 PM, Marc Malloy wrote:
Hello
I cannot seem to reduce the number of handles. I make changes to my
segmented volume, but each time I press Update Handle Count, it always
reports the same number of errors. In the manual it refers to a Patch
and Evaluate button, but I cannot seem to find either. Thanks!
marc
------------------------------------------------------------------------
_______________________________________________
caret-users mailing list
[EMAIL PROTECTED]
http://pulvinar.wustl.edu/mailman/listinfo/caret-users
_______________________________________________
caret-users mailing list
[EMAIL PROTECTED]
http://pulvinar.wustl.edu/mailman/listinfo/caret-users
_______________________________________________
caret-users mailing list
[EMAIL PROTECTED]
http://pulvinar.wustl.edu/mailman/listinfo/caret-users
----------------------------------------------------------------------------
----
----------------------------------------------------------------------------
----
_______________________________________________
caret-users mailing list
[EMAIL PROTECTED]
http://pulvinar.wustl.edu/mailman/listinfo/caret-users
_______________________________________________
caret-users mailing list
[EMAIL PROTECTED]
http://pulvinar.wustl.edu/mailman/listinfo/caret-users
