This sounds similar to some problems that I had with BET on some legacy data
which was taken along an unusual slice plane as a result it had a lot of
neck and jaw in the FOV. BET initializes using a triangulated sphere located
at the centroid of the 3D image. As a result, when there is this jaw-neck
"chaff" far away from the brain, the sphere can initialize with some of that
chaff within it's radius. It them deforms outward to include even more
garbage. If you prune (set to zero) the voxels well below the brain, this
pushes the centroid closer to where you want it (in the brain) and you
should notice significant improvement in the BET performance.
CJ
----- Original Message -----
From: "Amanda Bischoff-Grethe" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Wednesday, July 31, 2002 3:42 PM
Subject: [FSL] BET with archival data
> Hello,
>
> I'm seeking advice regarding some difficulties I've had with
> retrospective data. As part of our interest is using the images with
> FreeSurfer, I've read with interest the documentation on FSL and
> FreeSurfer. We are also interested in using BET as part of an
> anonymization process. I am working with 3 datasets all collected on a
> GE 1.5 T using Signa 4.05. All three were SPGRs, with two collected
> sagittally (one elderly and one young subject) and one collected
> coronally (elderly subject). Because I've run into some difficulties in
> getting FreeSurfer to properly skull-strip and segment white matter, one
> suggestion has been to use BET to remove the skull before processing in
> FreeSurfer. BET has been somewhat successful, but it has a tendency
> with these images to leave behind the much of the jaw and neck,
> particularly if I relax the threshold to 0.3 - 0.35 to ensure an intact
> brain (otherwise it tends to remove too much from superior parietal
> areas). The coronal subject looks far worse, however, possibly due to
> image quality. I tried using the -g option to adjust the gradient, but
> it seems to work anterior to posterior, rather than top down as
> suggested in the help file. I've also used avwswapdim to rotate the
> images to that suggested in the FSL with FreeSurfer documentation, but
> that does not appear to improve the problem; it still seems to adjust
> the gradient anterior/posterior. Any ideas on what might be causing the
> problem and how to solve this?
>
> Incidentally, after using BET FreeSurfer improves its surface detection
> with the two sagittal images but not the coronal one.
>
> Thank you.
>
> Amanda Bischoff-Grethe, PhD
> Post-Doctoral Fellow
> Laboratory of Cognitive Imaging
> VA San Diego Healthcare System
> 3350 La Jolla Village Drive, M/C 151A1
> San Diego, CA 92161
>
> Phone: 858/552-8585 x1120 (Lab)
> Fax: 858/642-6393
|