Coregistration can be made more robust by starting with the images in
closer alignment with each other. The Display and Check Reg buttons
can be used to change the initial orientation/positioning of the
images, which reduces the chance of getting stuck in a local optimum.
The way the algorithm works is to try one set of parameters, see what
the objective function is and use this to try to get a better set of
parameters where the objective function is better. Unfortunately,
doing it this way can result in not finding the best parameters.
An alternative would be to do some sort of grid search. For example,
trying 20 values for each of the parameters. Unfortunately, there are
six of them to find, which would make this infeasible. If there were
two parameters, there would be 20*20 different combinations to try.
With six there would be 20*20*20*20*20*20 different combinations, and
it would involve reslicing one of the images 64,000,000 times.
Therefore, the alternative approach is not really possible.
Stripping the skull and nonuniformity correcting the anatomical scan
can also help. Segmentation can be used to separate brain from
non-brain, as well as doing the nonuniformity correction. Combining
the output (grey, white, CSF as well as a bias corrected version of
the image) can be done using ImCalc.
Best regards,
-John
On 4 October 2010 22:04, Blefari Maria Laura <[log in to unmask]> wrote:
> Dear All,
> I am very happy to decide to join the forum. It seems very active and helpful.
> My problem is very basic. In my fMRI experiment I am using 20 slices. Can this fact generate problems during the coregistration with the anatomical?
> I am a little confused about why I got that result (see the attachment).
> I have been told that I probably need to have a 'peeled' version of my anatomical data. Because it seems that in my functional images, the skull is not visible, making it hard for SPM to coregister the data with the anatomical data, where the skull is visible. Any advise?
> Thanks in advance,
> ML
>
|