Image Registration with masking

[codeenthusiast]

I am trying to register two CT's and would like to specifically focus on the nasal cavity. As per the image below, I need very accurate registration for the nasal septum and inferior turbinates. However, when I overlay the registration output on the fixed image, there are definitely mismatches. A few questions:

• Masking would be a good way to go about this but how should I code this into my current platform?
• Should I segment the inferior turbinate?
• If yes, do I simply pass the segmentation as the "mask" argument for ants.registration?
• Should I segment the fixed image in that case?
• My current code first does a 'similarity' transform and then a 'SYN'

Thanks!

[ntustison]

I actually worked on this problem for quite awhile with a collaborator. Unfortunately, collaborator interest waned and the project never went anywhere. However, I played around enough with the data to know that you're probably not going to get a good alignment with registration alone. With my prior collaboration, our thought was to try to get a good parcellation of the regions and then do a segmentation-based registration. At the time, joint label fusion was the go-to approach for something like this (I think I still have the 10-20 templates that were hand-labeled) but with deep learning, there is a much greater potential for the degree of accuracy you're probably looking for.