Date: Wed, 23 Apr 2014 06:41:54 +0000
From:
[log in to unmask]Subject: Re: [FSL] first_flirt three step registration
To:
[log in to unmask]
Dear Alain,
In general the use of masks makes the registration less robust and more likely to go catastrophically wrong, as it has in this case. When the mask is very tight this is more likely, as it removes the useful border area from the calculation, and if it
is only within one region without any border at all then the registration has no information from the edges at the boundaries, which is crucial for the registrations to work. So we normally dilate our masks so that the border information is included.
This being said, I'm not sure if you will gain much from using even a dilated mask, unless you are clearly seeing systematic misregistrations in your data. It isn't something where a little bit more accuracy in the registration is likely to make much
difference, as the boundary mesh is deformed from the initial starting position and small differences in the initial starting position normally have no discernible impact.
All the best,
Mark
Dear FSL list,
I'm trying to play around a little bit with FIRST in order to get the best in term on striatum (putamen + caudate, no accumbens) segmentation. One of the things that I thought to try is to include a third registration step in first_flirt using a mask of the
striatum.
So I created a mask of the striatum tresholding the Harvard Oxford subcortical atlas (1mm) and I entered this image in first_flirt with the following call
first_flirt ${dirroot}${subject}_T1 ${dirroot}${subject}_firstflirt_p4
-strucweight ${Striatum}
What I obtain can be seen attached: the registrations look totally wrong, in different ways (mainly over-zoomed or strangely rotated).
Have any one an idea of what I'm doing wrong ? Do you think the cost-benefit ratio of applying the third registration is interesting ?
Thanks in advance and best
Alain
Bad registration 1
Bad registration 2
Striatum mask