Registering subjects with large ventricles

[9999sotelom]

Hello

I have a stroke subject with large ventricles (Figure 1, below). I used LINDA to obtain a lesion mask (small lacunar lesion in the left basal ganglia area).

My template is a control subject (Figure 2).

I am registering the stroke subject to this template using the command below: ANTS 3 -m PR[${templateImBrain}, ${strImBrain}, 1, 2] -i 50x20x10 -o str2template_ants.nii.gz -t SyN[0.3] -r Gauss[3,0] -x ${lesionMask}

The result is shown in Figure 3. the Axial view looks OKAY, other than the inferior horn of the ventricles. The Coronal view shows the largest issue (left ventricle is not warped properly).

Question: Am I right to consider this a failed registration? Are there any recommendations to improve my registration? Some ideas I have is modifying the iterations, the cost function. Anything else?

Thank you!

- Miguel

Figure 1

Figure 2

Figure 3

[ntustison]

Hi,

The ANTS program is somewhat deprecated so try switching to antsRegistration. To make using the latter easier, start with the scripts antsRegistrationSyNQuick.sh and antsRegistrationSyN.sh. Once you've looked at the results from both of those runs, come back and we can discuss potential refinements.

Nick

[dorianps]

You can also try registerLesionToTemplate from the Lesymap package. That will perform brain extraction, denoising, etc., before running the registration.

Looks like your original command has a single SyN deformation, without an affine. That would certainly be bad for the registration. The scripts advised by Nick have all the necessary transformations. But I wouldn't use the quick if you want the best registration.

On Fri, Dec 22, 2017, 12:00 Nick Tustison notifications@github.com wrote:

Hi,

The ANTS program is somewhat deprecated so try switching to antsRegistration. To make using the latter easier, start with the scripts antsRegistrationSyNQuick.sh and antsRegistrationSyN.sh. Once you've looked at the results from both of those runs, come back and we can discuss potential refinements.

Nick

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/ANTsX/ANTs/issues/532#issuecomment-353637409, or mute the thread https://github.com/notifications/unsubscribe-auth/AIqafVPEHORLYjBTwXsrQzDrtlvagYBmks5tC-A_gaJpZM4RLOcA .

[9999sotelom]

Thank you all for the replies, and I hope the holidays were filled with more food digesting than image processing.

The default antsRegistrationSyNQuick.sh script produced an OKAY result (Figure 2), without the erroneous ventricle warping from ANTS(coronal view, right hemisphere ventricle, Figure 3), but unfortunately less ventricle registration than I would like (registered Figure 2 ventricles looks more similar to native space Figure 1 than template space Figure 0). *I will not use the Quick script

The default antsRegistraionSyn.sh script (with histogram matching) produced BETTER results (comparing Figure 4 with Figure 0), as predicted by @dorianps . My concerns are 1) in the coronal view there are 'radiating' streaks near the ventricle horns--the result of warping the large ventricle to a smaller volume--and 2) the axial view shows a longer ventricle inferior horn (Figure 4: posterior part of the ventricles is longer). The results are workable, but I am wondering if there is a way to address these two concerns.

Thank you very much.

Figure 0: template Figure 1: stroke subject native space Figure 2: synQuick.sh Default Figure 3: ANTS + lesion mask Figure 4: syn.sh + histogram matching