Dear Julia,
elderly subjects can sometimes be a bit tricky. On the one hand the large ventricles means that one needs to do non-lin reg. On the other hand I have observed that elderly often have a high signal from the fat in the marrow of the skull bone. Since that signal is not present in the MNI template one sometimes see cases where that marrow signal at the crown of the head gets registered to the top part of the brain in the MNI. You can then get the effect you see.
One can sometimes solve the problem by experimenting a little with how the masking is applied.
Jesper
On 30 Mar 2016, at 15:26, Julia Schumacher <[log in to unmask]> wrote:
> Dear FSL experts,
>
> I am trying to do registration on data from elderly dementia patients with quite large amounts of atrophy and enlarged ventricles.
> I am performing a two-step registration procedure using the MELODIC GUI, so first using FLIRT for registration of the functional to structural images (Normal Search, 12 DOF) and then using FNIRT for the nonlinear registration of structural (and functional) images to the MNI152 template (Normal Search, 12 DOF, 10 mm warp resolution).
> The first step works fine.
> However, the nonlinear registration of the structural to standard image results in bad registrations for some of the subjects where the upper part of the brain seems to get pulled in (https://www.dropbox.com/s/qd6krdkfvkcn2fa/highres2standard_nonlin.png?dl=0).
> We thought that the problem might be caused by FNIRT trying to align the enlarged ventricles with the much smaller ventricles of the template and thereby pulling the rest of the brain in.
> So, I also tried using linear registration for both steps, which results in nicely registered brain boundaries, but then obviously the ventricles and subcortical structures are registered poorly (see https://www.dropbox.com/s/r1dsputl0ksllw4/highres2standard_lin.png?dl=0).
>
> Has anyone experienced similar problems with brains that show atrophy and large ventricles?
> Is there any solution to this? Specifically, is there any way to adjust FNIRT parameters to maybe prevent this problem or to find a tradeoff between good registration of the ventricles and good registration of the cortex?
>
> I would be very grateful for any help!!
> I am also happy to provide more data if that helps.
> Kind regards,
> Julia
|