Compatibility with 3D kspace

[tamas-borbath-orbem]

Hi,
I am considering to use the fastMRI model for an undersampled 3 dimensional k-space (two phase encoding directions) and not 2 dimensional slices of k-space (one slice encoding and one phase encoding), as it is currently in all the datasets.
Does anyone have experience with this?
Thank you very much!
All the best,
Tamas

[mmuckley]

Hello @tamas-borbath-orbem, our current code base in open source would not work with this. You would probably mostly need to rewrite how [subsampling] is done and then write a new DataTransform for your case.

I don't know of any projects that have done this but there might be some out there.

[tamas-borbath-orbem]

Thank you very much for your answer! Much appreciated!
In case I will solve it, will contribute with a pull request

[mmuckley]

@tamas-borbath-orbem just one thing to note: in terms of including your code in the repo: we would merge it only if it were in scope (i.e., works with the fastMRI data). Otherwise, you would have your implementation in your own repository. That being said if you wanted to advertise your 3D models here on the Discussions page I think that would be fine :).

[tamas-borbath-orbem]

Sure! :)