Hi,
This is going to be a difficult registration problem.
You can use mutual information, but that won't necessarily work very well either.
If you stick with linear registration then there will be a limit to how well it can match the MNI.
If the ventricles and areas around them are of more interest then you might want to create a segmentation of them in the MRI and the CT and then register these segmentations together.
Good luck,
Mark
> On 20 Apr 2017, at 02:03, Sam Choi <[log in to unmask]> wrote:
>
> Dear FSLers,
>
> I am working with some CT scans of stroke patients and want to register the brain and lesion mask from CT space to MNI space. We do not have MR images of these patients, so we are unable to perform a two-stage registration.
>
> Currently, we are using the commands (below) to register the brain and lesion mask to MNI space. However, the outcome of this registration is not very good since the ventricles do not match up between the images (i.e., CT scan in MNI space and the MNI template).
>
> 1) flirt -ref $FSLDIR/data/standard/MNI152_T1_2mm_brain.nii.gz -in ct_brain.nii.gz -inweight ct_lesion_mask.nii.gz -omat ct2mni.mat
>
> 2) flirt -in ct_scan.nii.gz -ref $FSLDIR/data/standard/MNI152_T1_2mm_brain.nii.gz -applyxfm -init ct2mni.mat -out ct_scan_std.nii.gz
>
> 3) flirt -in ct_lesion_mask.nii.gz -ref $FSLDIR/data/standard/MNI152_T1_2mm_brain.nii.gz -applyxfm -init ct2mni.mat -interp nearestneighbour -out ct_scan_std.nii.gz
>
> We were wondering if a different cost function option in FLIRT should be used. I appreciate any advice you may have to improve the registration.
>
> Thank you,
>
> Sam
|