Hi Mark,
Thanks for your response. I tried the multi-channel approach with 3 classes and with 4 classes. I didn't find it to be reliable either. I had wanted to check that I wasn't overlooking something obvious.
Thanks again,
Keith
________________________________________
From: FSL - FMRIB's Software Library [[log in to unmask]] on behalf of Mark Jenkinson [[log in to unmask]]
Sent: Wednesday, January 05, 2011 3:58 PM
To: [log in to unmask]
Subject: Re: [FSL] misclassification of white matter using FAST
Hi Keith,
If you have white matter intensity inhomogeneities that are disease- or
age-related then FAST will not segment them properly, as you've found.
The only solution we have is to manually identify the "lesions" by creating
a manual mask (or any other way you know of) and then either excluding
them from the FAST segmentation (setting them to zero is enough for this)
or "filing" them with "normal" white matter intensities (ideally taken from the
surrounding voxels) prior to segmentation.
You could also try a multi-channel segmentation with 4 classes, but my
experience is that it tends not to work very reliably.
All the best,
Mark
On 5 Jan 2011, at 21:53, Keith Hulsey wrote:
> I have used FAST to segment T1 weighted MPRAGE images into three classes. (Skull stripping and other image preparation has been done as used in SIENAX.) White matter regions which are very bright in FLAIR images tend to be darker in the T1 than other white matter. As a result FAST tends to misclassify this white matter as gray matter. Do you have any suggestions for automatically identifying FLAIR hyperintensities in the white matter that are misclassified as gray matter by FAST? I have MPRAGE, FLAIR and T2 images of my subjects.
>
> Thanks for your help,
> Keith
>
________________________________
UT Southwestern Medical Center
The future of medicine, today.
|