The best thing to do is use an MR sequence that has distinct image intensities
for the different tissue classes. In the case of your data, you are finding
non grey matter voxels in the vicinity of grey matter, but which have similar
intensities. This makes eliminating such voxels rather more difficult.
In SPM2, there was a fudge included within the segmentation in order to do a
brain extraction and some cleaning up. This didn't work perfectly, but it
eliminated some misclassified non-brain tissue. I may (reluctantly)
incorporate that same fudge into the final version of SPM5.
For the moment, I don't really know what to suggest. Some segmentation
packages make use of Hidden Markow Random Fields in order to model spatial
dependencies between neighbouring voxels. You may be able to use a
segmentation package that includes such MRFs in order to reduce the number of
such misclassified voxels. If nothing will do it, then you'll have to resort
to a manual approach.
Best regards,
-John
> I'm working with anatomical T1-weighted images, scanned with a Siemens
> Symphony 1.5 Tesla MRI scanner (quantum coil system).
> I'm using SPM2 and SPM5b running in MATLAB 5.6.1 and Windows 2000.
> There are some problems doing a VBM analysis:
> Using the SPM2 optimized version there are still parts of dura mater
> parietal in the GM image after the automated extraction and segmentation
> procedure. Using SPM5b there are even more parts of dura mater not
> extracted. That's confusing because in the mailing list John Ashburner
> adviced among other things to use SPM5 when extraction problems occure in
> SPM2 (20. June 2005).
>
> Now my questions are
> 1. Are there possibilities to improve the results of extraction and
> segmentation within SPM2 or SPM5b? (I've allready tried to multiplcate the
> grey and white matter partitions withe the brain mask. There was no
> improve.) Which parameters can I change to improve the algorithm and how
> does it work? 2. Should I use an extern automated extraction algorithm
> alternatively? Which one performs significantly better than SPM? (Boesen et
> al. (2004) showed that there is no dramatic improve when using BSE or BET.)
> It has to run in Windows 2000 and it has to be freeware. And it has to be
> handeled easily. (e.g. it takes too much time and effort to install and
> setup McStrip.)
> 3. Or will I get best results by masking out dura mater manual using
> MRIcro? John Ashburner mentioned that there is a need to controll the
> header afterwards (20. June 2005). What should I'll be looking for?
> Thank you very much for any assistance you will provide me with.
>
> Best regards,
> Anja Heidgen
|