Most segmentation is done on T1 weighted images, rather than T2 or
FLAIR, but note that the accuracy of the resulting segmentation will
depend on your images. It is possible that the segmentation may work
extremely well on your T2 or FLAIR images. One possibility would be to
try to combine (e.g. average, or weighted average) the results of
segmenting the individual images. For best results, try a little trial
and error. In SPM2, there was the option to do multi-spectral
classification, but this option has gone from the SPM5 user interface.
Tissue probability maps used for segmentation should not depend on the
image contrast. These tissue probability maps should just represent the
probability of finding a particular tissue class at a particular
location. I don't know how this relates to the Dice metric question you
are asking though. For the evaluations in the paper, the Dice metric
was measured by comparing the c1* with the grey matter used to simulate
the images, and the c2* was compared with the white matter image used by
the simulator.
The bias correction is about whether you want the inhomogeneity to be
corrected or not. If your images have no inhomogeneity, then it is
better not to correct for it. If there is a lot of inhomogeneity, then
it may be helpful to let the bias correction have a lot of flexibility
in how it estimates the bias (i.e. low regularization). Whether you
save a bias corrected image or not will have no effect on how well the
segmentation actually works.
Best regards,
-John
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
On Behalf Of Riandini -
Sent: Tuesday, January 23, 2007 2:00 PM
To: [log in to unmask]
Subject: [SPM] more questions on SPM5
Hi SPM-ers,
Again, more questions about SPM from me.
These are my questions (to ease up my confusness) and perhaps I've
already
asked these questions:
* according to SPM-VBM, it said that before doing a VBM analysis ,
the images should be checked its alignment by choosing T1.nii. How about
T2 and FLAIR images? Since in SEGMENT if we choose the Tissue Probablity
Maps option it always refer to GM-WM-CSF.nii of T1 images.
* in the paper " Unified Segmentation ". It mentioned about
eavaluations with BrainWeb phantoms for images of T1, T2 and PD. Also
mentioned about Dice metric which measured or compares images with
"true"
grey and white matter. Do you familiar with those? If yes, regarding the
T2 and PD images, which files should I specified in the Tissue
probability
Maps.
* Bias correction. Could you explain me more in "a simple way" on
its concept (also Bias regulation and Bias FWHM)? In the paper, it said
that it should be handled properly. So far what I did in my study I
always
choose SAVE bias corrected. Would it be influenced my volume
calculation?
As far as I understand it has relation with the inhomogenity of the
images. Am I right?
Thank you.
kind regards,
Dini
|