Hi Darren, no - nothing has changed - I'm not sure if you were maybe
thinking of a discussion involving single channel (fast) as opposed to
multi-channel (mfast) segmentation? Xavier is right in thinking that it
appears often to be the case that for two-channel segmentation, 4 classes
gives better results than 3.
Thanks, Steve.
On Wed, 2 Jul 2003, Darren Weber wrote:
> Hi all,
>
> hmmm, very interesting... a while ago we discussed whether fast can actually
> run with 4 classes (or more) and I recall (possibly in error) that fast was
> designed and effectively limited to 3 classes (or at least optimal at 3
> classes rather than 4). Has something changed in the way that fast does the
> initial k-means segmentation? Is this process an adaptive one for different
> contrast images (eg, T1/T2)? Perhaps another processing stage has been
> introduced? Are these results specific to a T2 contrast?
>
> Best regards, Darren
>
>
> ----- Original Message -----
> From: "Xavier Chitnis" <[log in to unmask]>
> To: <[log in to unmask]>
> Sent: Wednesday, July 02, 2003 10:26 AM
> Subject: Re: [FSL] mFAST
>
>
> > Dear Steve,
> >
> > Thanks for taking a look at that image. The three class segmentation
> > does produce a good result on that subject.
> >
> > However, when I started using mFAST on this dataset I initially
> > segmented a subset of scans using the default settings. I found that the
> > three class segmentation didn't work well, with generally a poor
> > grey/CSF separation. I read on the FAST webpage that when segmenting T2
> > images, it tended to work better with a 4th class for dark non-brain
> > tissue. Hence I tried this and found that generally it produces
> > excellent results. Out of 64 subjects, 56 have segmented very well with
> > 4 tissue classes. It is a relatively small number where this approach
> > hasn't worked.
> >
> > I have put on our webpage two new tar files (01_4CLASS and 01_3CLASS).
> > They are the same subject segmented into 3 or 4 tissue classes (I
> > apologise, the files are quite large...). The 4 class segmentation is
> > very good, but the 3 class hasn't to my eye distinguished between grey
> > matter and CSF.
> >
> > http://www.iop.kcl.ac.uk/iop/Departments/BioComp/BIAU/pickup.shtml
> >
> > Given that the 3 class segmentation may work for those images that fail
> > the 4 class method (which works well for the vast majority), I am
> > wondering whether it would be valid to combine data from scans segmented
> > using two differing sets of parameters?
> >
> > As always, I am very grateful for any advice or suggestions
> >
> > Best wishes
> >
> > Xavier
> >
> >
> >
> > -----Original Message-----
> > From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On
> > Behalf Of Stephen Smith
> > Sent: 01 July 2003 13:52
> > To: [log in to unmask]
> > Subject: Re: [FSL] mFAST
> >
> > Hi - I'm a little confused - your questions certainly made sense, but
> > running mfast with the defaults on this dataset gave very nice results!
> > I
> > ran:
> >
> > mfast -s 2 -c 3 -od 27_m_pd 27_m_pd.hdr 27_m_t2.hdr
> >
> > does this not give good results for you?
> >
> > thanks :)
> >
> >
> >
> > On Mon, 30 Jun 2003, Xavier Chitnis wrote:
> >
> > > Dear All,
> > >
> > >
> > >
> > > I have been using mFAST to segment some dual-echo (proton density/T2)
> > > images.
> > >
> > >
> > >
> > > Following the suggestions on the FSL webpage, I have been segmenting
> > > them into 4 classes, using partial volume classification.
> > >
> > >
> > >
> > > This has worked on the majority of my images. However, the
> > segmentation
> > > is failing on around 10% of my data. The segmentation finds a white
> > > matter class, and (what I term) a dura class consisting of non-brain
> > > tissue left by BET. However, it fails to separate grey matter and CSF
> > > accurately.
> > >
> > >
> > >
> > > I have found in this dataset that 3 class segmentation failed. I have
> > > tried changing some of the options e.g. disabling bias field
> > correction,
> > > or going for 2D segmentation, however have been unable to improve
> > > matters.
> > >
> > >
> > >
> > > I would be very grateful for any suggestions. I have put an example
> > > subject's data (27_fse.tar.gz) at
> > > http://www.iop.kcl.ac.uk/iop/Departments/BioComp/BIAU/pickup.shtml if
> > > anyone would like to take a look at it.
> > >
> > >
> > >
> > > Many thanks
> > >
> > >
> > >
> > > Xavier Chitnis
> > >
> > >
> > >
> > >
> > >
> > > Neuroimaging Research Group
> > >
> > > Institute of Psychiatry, London
> > >
> > >
> > >
> > >
> >
> > Stephen M. Smith MA DPhil CEng MIEE
> > 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
> >
>
Stephen M. Smith MA DPhil CEng MIEE
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
|