I am trying to register (rigid) a MRI image of a phantom to its CT image. However, I am not getting the results I need. I am using MattesMutual Metric, and regularstepgradient, Euler3DTransform, etc. Has anyone encountered similar problem and solved it?
It looks like repetitive grid pattern quickly puts the registration metric into a local minimum. This could benefit from multi-resolution approach, especially with a lot of initial downsampling.
Is the initialization good? It looks from your code that you’ve scaled the CT to be positive, which is a good start, but it’s worth doing a visual check of where the optimization is starting from.