Hi,
The chances are that this is due to a limited FOV in the inferior direction. Try taking the most inferior slice (use fslroi to get this) and attach multiple copies of it to the inferior part of the image (using fslmerge). Then try running FIRST on this and see if it fixes the problem.
All the best,
Mark
> On 11 Dec 2014, at 16:33, Gabor Perlaki <[log in to unmask]> wrote:
>
> Dear all,
>
> I've difficult images of older brains. In order to improve the segmentations by FIRST, I've tried to use FNIRT to non-linearly register the brain to the 1mm standard space and then run FIRST on the registered output. (After that the results can be warped back to the original image using invwarp.)
>
> Since, I registered my T1-weighted MPRAGE to the MNI152_T1_1mm by fnirt, I used an identity matrix in the run_first_all command:
>
> run_first_all -d -a ident.mat -i input_already_registered_to_MNI_1mm_by_fnirt -o output
>
> However, this command fails when segmenting the brainstem with the following message: "*** Error in `/usr/share/fsl/5.0/bin/first': free(): invalid next size (fast): 0x000000000385ab90 ***"
>
> The command "run_first_all -d -a ident.mat -s BrStem -i input_already_registered_to_MNI_1mm_by_fnirt -o output" also fails. The segmentation of the other subcortical structures (e.g. Thalamus, Putamen, .....) seems to be OK.
>
> I've also tried it using MNI152_T1_1mm.nii.gz as an input ("run_first_all -d -a ident.mat -s BrStem -i MNI152_T1_1mm.nii.gz -o output"), which failed with the same error message.
>
>
> I've used FIRST on non-standard space inputs (in 50+ cases) without any error message during the segmentation of BrStem or other structures.
>
> Any idea about this issue?
>
> Best regards,
> Gabor
>
>
>
>
>
>
> -s BrStem
>
>
>
>
>
> I've read in an earlier comment that in some cases (e.g. in elderly subjects) it is often worth trying to use
>
>
> But if you
> have difficult images of older brains it is worth trying.
>
> We haven't come up with a general solution that always
> works, but it is often worth trying to use FNIRT to
> non-linearly register the brain to the 1mm standard space
> and then run first on the registered output. The
> results can be warped back to the original image
> using invwarp. Whether this helps or not depends on
> how well the registration works usually
|