Dear list
Note: This is not a howto - in fact I'm looking for one!
I'm running SIENA analyses on a group of patients and controls scanned
at 2 different time points. A first problem that I had with the analyses
was that the required BET settings for the time points were not the
same. This was fixed by modifying the script so that it allows different
BET parameters for both images.
Now the SIENA analyses finish for all image pairs. On closer inspection
though, we found that sometimes, although BET chops large parts off the
brain for one of the time points (eg temporal lobe or cerebellum) the
registration parameters and the FAST segmentation seems to be accurate.
In other cases the BET, registration and masking seem to have gone well,
but a complete gyrus seems to have moved (resulting in a large red/blue
blob in the atrophy map)
So what we have is a number of SIENAs with outright failed BETs, whose
results can obviously not be used, a number of SIENAs with good-looking
BETs and atrophy maps that look sensible. The worry is about SIENAs
where at least one of the BETs has gone wrong, but the atrophy map does
not show strange or unexpected things.
Example SIENA results pages (as screenshots) of these 3 cases can be
seen here:
http://www.wbic.cam.ac.uk/~amw71/sienafailcrop.png
http://www.wbic.cam.ac.uk/~amw71/sienasuccesscrop.png
http://www.wbic.cam.ac.uk/~amw71/sienasuccesscrop.png
I am curious to know
1. Is it possible to get a good SIENA after a non-good BET?
Specifically, if the BET excludes a whole section of the brain at one
time point, but the pairwise registration and the FAST segmentation
afterwards look OK (don't exclude that section), can we trust the
atrophy map?
2. Would there be a an evaluation criterion for SIENA results in general?
If the pairwise registration is bad, or there are visible differences
between the segmented versions of the two time points, we found that
atrophy estimates (in %) tend to be higher -we found around twice as
high for SIENAs that still finish, but with problems.
However we also have results (often with a dodgy BET but seemingly good
registration and segmentation) where the atrophy estimates are very low,
about two to three times lower. Are these two (dramatically higher/lower
atrophy estimates) indicators of a failed SIENA analysis which is
visually OK?
There are very few clues in the literature or in the form of HOWTOs, so
any help is very welcome!
With best wishes
Alle Meije Wink
|