2d-3d image registration?

[mdhe1248]

I tried 2d-3d registration and have quite a few naive questions:

First, while the 2d fixed image is "front view", the registered image is "side view". Is it an expected result? Shouldn't it be the same orientation as the fixed image, or am I missing something here? It seems that the orientation of moving volumetric image is different from that of fixed. If I rotate the moving image to match the orientation to the fixed, the results seems reasonable (to me). (fixed 2d image)

(2d-3d registration)

(2d-3d registration after orientation change) By the way, all images are upside-down; but this is not important here. This is because, my input images are upside-down.

Second, if I look at the mhd header file of the fixed image (and also 3d image), it has offset. Where are they coming from? If I change them in the fixed image header, the resulting registered image is off-centered.

Also, the values in offset are similar to those in origin in the parameter file. Are they related? It is hard to find why and how they are set. Please advise me. Along with this, interpolator values in the parameter file seem pretty important. If I don't set them (probably using default values), I get weird results. I don't understand how they are set.

Maybe related to my second question: Can I use an nrrd file as a fixed image? I loaded the example mhd file and save it as nrrd; and using it as a fixed image resulted in an off-centered, weird image:

My end goal is to register a 3d nrrd image to a 2d nrrd image. It would be nice if the 2d-3d registration demonstrated in this repository is applicable to many other data set.

[mdhe1248]

Progress update: Well, the issue with NRRD format was the header. Saving as .nrrd did not keep all the header information in the .mhd header file. I ended up with writing .nhdr (NRRD header file) with all information in the mhd file.

Still, I don't understand the interpolation parameters. Any comments?

[N-Dekker]

@ntatsisk Please check, as we discussed.