Tim,
Thanks for the explanation. I agree that the partial volume maps appear
to have more information. Now for my next problem.
I am running mfast on 200x200x200 mm isotropic brain data with 5 input
channels and requesting 6 compartment output with partial volume
estimates. The data are stripped of non-brain and the input is trimmed to
eliminate as much non-image edge as possible to reduce the size of the
dataset. I have also scaled the input channel data to 8 bits to further
reduce the load. The routine takes up to 3+ hours on a 1.6 GHz linux
processor and uses over 2 G of memory.
initial segmentation by KMeans....is quick
4 main iterations ...takes about 30-40 minutes
estimating the partial volumes takes hours
Is this normal performance?
Dolf
On Mon, 29 Sep 2003,
Tim Behrens wrote:
> Hi there -
>
> There are a few different stages to fast -
> After the initial k-means segmentation, the first thing it tries is to fit
> a Gaussian mixture model (with spatial priors) - The underlying assumption
> here is that each voxel has a true underlying "hard" segmentation - i.e.
> there is only ever _one_ matter type in a voxel, and that there is a
> probability given the data of that matter type being Grey matter, white
> matter or csf. These probabilities are the values you get if you use the
> -op option.
>
> If you specify the -ov option, fast goes a stage further. It then fits a
> different model called a partial-volume model, where each voxel can
> contain more than one matter type. the output images are then the best
> estimate of the _fraction_ of each matter type in the voxel.
>
> These are conceptual differences. From a practical point of view, my
> experience is that the partial volume images look better than the
> posterior probability maps, but I don't have a huge amount of experience.
> I guess steve can comment here...
>
> Hope this helps
>
> Tim
>
> -------------------------------------------------------------------------------
> Tim Behrens
> Centre for Functional MRI of the Brain
> The John Radcliffe Hospital
> Headley Way Oxford OX3 9DU
> Oxford University
> Work 01865 222782
> Mobile 07980 884537
> -------------------------------------------------------------------------------
>
> On Mon, 29 Sep 2003, Adolf Pfefferbaum wrote:
>
> > How do the mfast -op and -ov options differ, it seems at first thought
> > that they might be very similar?
> >
> > -op output probability maps (one for each class)
> > -ov: output partial volume images (one for each class)
> >
> >
> > Thanks,
> >
> > Dolf
> >
> > ____________________
> > Adolf Pfefferbaum,MD
> > Neuroscience Program
> > SRI International
> > 650-859-2927 phone
> > 650-859-2743 FAX
> >
> >
> > CONFIDENTIALITY NOTICE: Information contained in this message and any
> > attachments is intended only for the adressee(s). If you believe that you
> > have received this message in error, please notify the sender immediately
> > by return electronic mail, and please delete it without further review,
> > disclosure, or copying.
> >
>
--
____________________
Adolf Pfefferbaum,MD
Neuroscience Program
SRI International
650-859-2927 phone
650-859-2743 FAX
CONFIDENTIALITY NOTICE: Information contained in this message and any
attachments is intended only for the adressee(s). If you believe that you
have received this message in error, please notify the sender immediately
by return electronic mail, and please delete it without further review,
disclosure, or copying.
|