search

multimodallearning / Lung250M-4B

Lung250M-4B: A Combined 3D Dataset for CT- and Point Cloud-Based Intra-Patient Lung Registration
Apache License 2.0
30 stars 3 forks source link

Usage with anisotropic CT data #1

Open cagatayalici opened 1 year ago

cagatayalici commented 1 year ago

Thank you very much for this amazing repo!

I am currently using intensity-based registration for pre-treatment and post-treatment CT pairs. I was investigating if I can improve the results by using correspondence point guidance.

I wanted to use Corrfield to obtain matching key points for my CT pairs. As far as I understood before running Corrfield data should be preprocessed such that its origin is at zero and voxel spacings are 1 mm.

I was wondering how I can use the provided code with anisotropic CT data with random image origin. I can move image origin to zero but I want to use my data with full resolution.

Thank you!

FenjaF commented 1 year ago

Hi @cagatayalici,

computations for corrField are performed based on the same amount of voxels in each direction. While you could run corrField on anisotropic data, this might result in uneven penalization of deformations in different directions. A resampling to exactly 1mm isotropic resolution however is not necessary, a higher resolution works fine as well. In that case you might want to consider changing the parameters slightly.

Best, Fenja