Hi - the point here is just that if you can estimate the partial
volume fractions (e.g. 35% GM and 65% WM in a given voxel) then using
those to estimate total volume of any given tissue type should be a
lot more accurate than hard-classifying the segmentation and using
that to estimate volumes. We have shown (e.g. see the old NeuroImage
SIENA/SIENAX paper) that this is indeed the case - e.g. that test-
retest repeatability of estimated tissue volumes is better if you use
partial volume estimation.
The partial volume estimation is primarily done on the basis of the
intensity mixture model - once you've modelled each class's mean and
variance intensity, any given voxel intensity will sit somewhere
'between' two class peaks, and where it sits is used to estimate the
partial volume fractions.
Cheers, Steve.
On 26 May 2009, at 15:41, Ranganathan, Sudarshan wrote:
> Hi
>
> Thank you for the reply. I would like to find out how the partial
> volume images are generated and how they are analyzed in the FAST
> segmentation algorithm. Would you like to give me some pointers
> about this? I am also trying to understand why the volume
> measurement results are different when FSLSTATS are used on the
> binary images.
>
> Regards,
> Sudarshan.
>
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
|