Resample multimodality image mismatch

Algorithms
, ,
1

Hi!
I’m trying to create a new 3D image from two medical images (stir=fixed image and axial 3d=moving image). This new image is wanted to be in the same space that the STIR image.

I’ve tried with sitk.resampleimage.

With some images that method works.
STIR image:
image
New obtained image from axial 3D:
image

But in others, the result image differs a lot from what is desired:
STIR
image
Obtained image:
image

Could someone tell me why this is due?

Thank you very much in advance :slight_smile:

2

Hello @AmericaBG,

Is the transform between the two images obtained via registration? I suspect it is, given the usage of “fixed/moving image”. If this is the case, two things can possibly account for the issue:

  1. The registration failed to converge to the correct solution. This is likely due to registration initialization being too far from the correct transformation (see Jupyter notebook on registration initialization).
  2. Registration succeeded but there is an issue with your visualization. It looks like you are using matplotlib to display images and it assumes isotropic data (pixel sizes are the same in the two directions). This is not the case with medical images, see fundamental concepts and the Jupyter notebook on results visualization.
1 Like