1. BET should be reasonably robust against bias field effects as the
relevant thresholdings (etc) are computed locally. However if the bias is
really bad you could iterate BET-FAST-BET.
2. Well, there various tweaks possible to BET, though it is already
attempting to get this best segmentation - see the FSL email list archives
for other comments on tuning BET.
On Tue, 1 Jun 2004, Himachandra Chebrolu wrote:
> Hi Everybody,
> SInce this was started ,I wanted to pop in a few questions regarding the bias correction.I've two questions:
> 1. Does the bias field correction have any impact on the BET
> algorithm.i.e.if I bias correct the MPRAGE volumes before using BET,does
> that give better performance on the BET.
> 2. Is it practically possible to get a complete brain volume with
> absolutely no non-brain matter automatically?I've been running BET
> volumes but it is picking up considerable amount of non-brain matter
> particularly in the orbital fat region and the tongue etc.
> Clare Mackay <[log in to unmask]> wrote:
> Hi Krish,
> I'll be brave and answer this one - since I've asked the same questions
> myself in the past...
> 1) As I understand it, the bias field estimations and segmentation
> procedure run iteratively together, so you can't have one without the
> 2) Yes, but you need to output a dilated version of the bias field
> (-oba option) and then run avwmaths_32R (i.e. need to have floating
> point data). You can revert back to more sensibly sized bit depths after
> you've multiplied by the bias field
> fast -l 100 -i 8 -oba 100
> avwmaths_32R [input] _abias
> Clare E. Mackay, Ph.D.
> Research Fellow,
> Prince of Wales International Research Centre (POWIC),
> University Department of Psychiatry,
> Warneford Hospital, Oxford. OX3 7JX. UK
> tel: 01865 455910 fax: 01865 455922
> OCMR: 01865 221866 / 72
> [log in to unmask] www.psychiatry.ox.ac.uk/powic
> >>> [log in to unmask] 01-06-2004 13:00:59 >>>
> We have just started generating mp-rage volume scans from our new 3T
> Trio, using the 8-channel head coil.
> The images look great, with excellent grey-white matter contrast, but
> is a noticeable bias field.
> Fast seems to make a great job of bias-correcting these images, but I
> two questions:
> 1) I wonder whether it is possible (on the command line) to simply
> bias-correct the images, without doing the segmentation? Or does the
> segmentation not really contribute to the processing time?
> 2) Is it possible to apply the generated bias field image to another
> (i.e. The non-skullstripped image)? Is it simply a matter of
> multiplying in
> Thanks and all the best,
> Dr K.D. Singh
> Senior Lecturer and Convenor of the Neuroimaging Research Group,
> Director, MRI Research Centre,
> Neurosciences Research Institute
> Aston University, Birmingham B4 7ET, U.K.
> Tel/Fax: +44 (0)121 [359 3611 ext 5176/5190]/[333 4220]
> [log in to unmask], http://www.aston.ac.uk/lhs/staff/singhkd/
> Do you Yahoo!?
> Friends. Fun. Try the all-new Yahoo! Messenger
Stephen M. Smith DPhil
Associate Director, FMRIB and Analysis Research Coordinator
Oxford University Centre for Functional MRI of the Brain
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve