Mutual Information metric in Deformable registrations about simpleitk-notebooks

[myronzbrown]

Hi. I am looking for the same thing and don't see it in sitk now. Did you find an answer for this or an alternative? From experimenting with the sitk registration methods, demons appears to significantly outperform the other metrics for producing a well-regularized deformation field. Using ImageRegistrationMethod.SetMetricAsDemons or the four specific filters (e.g. FastSymmetricForcesDemonsRegistrationFilter), the result I get is well regularized if I specify a large Gaussian smoothing value. For the other metrics, I see fast deformations even if I set the variance very high. I am hoping to do a direct comparison of demons with squared error vs. mutual information. Any tips you could share?

Thanks,

[zivy]

Dear @myronzbrown,
You posted your question in a closed issue. In the future, please (re) open an issue or ask on ITK discourse.

With respect to the question, not clear to me what exactly it is that you are asking. What registration setups (optimizer-similiarity metric- inter/intra modal) do you want to compare?

[pubgeo]

Thanks @zivy. Sorry. I will do that in the future. I was hoping to roughly reproduce in python the inter-modal alignment results I get in MATLAB using demons with the mutual information metric. I can get roughly that for inter-modal case in sitk, and it works pretty well for intra-modal image pairs if I apply a Laplacian of Gaussian filter on the images first, but i was hoping to get a direct comparison with mutual information. Not urgent, but it seems like something that would come up from time to time so I wanted to ask if I'm missing something that's already available.

[zivy]

Hello @pubgeo,
No worries. You are not missing something that is available. SimpleITK doesn't have a MI based Demons registration.