I was wondering what are the plans for further development of this library

[mscipio]

Hi,
not really an 'Issue', but more of a praise and some curiosity-driven questions ...

I just started very recently to play around a bit with one of the dependencies you used for this project, torchkbnufft, and as soon as I was able to successfully use it to reconstruct some radial k-space data I had available I got to think about how cool it would be to have a proper MRI recon library built around it, like sigpy is based on cupy, for instance.

Luckily enough, a very quick search led me here, and even though I have had the chance yet to try out your work, this project looks super exciting!

I saw that you already have some basic regularization methods already implemented, but I was wondering if you had plan to further expand the library. I was thinking as a starter of a way to estimate coil sensitivity maps directly from here (maybe using ESPIRIT, as sigpy and BART do), and also some compress sensing algorithms.

The idea of having an imaging library based on PyTorch and ready to be used for DL research is just so exciting.
I would be happy to get in touch and help if you are interested and willing to keep working on this project!

Cheers

[guanhuaw]

Thanks, Michele, for your warm encouragement!
Currently, it has some 2D examples as jupyter notebooks, including B0-informed correction, blind compressed sensing, etc.
We are planning to release the first stable version this year; improvements including:

1. Algorithms like fiast, admm and pogm; applications like low-rank, cs algorithms ...
2. A training workflow for CNN-based methods.
3. Docs and (maybe) some tutorials on optimization.
   For sensitivity maps, I am worrying that an unsophisticated torch implementation will be much slower than BART.
   While for some cases where backpropagation may help the sensitivity maps calculation (calibrationless multi-coil recon), we would be very happy to reproduce them.
   Looking forward to your ideas!

[guanhuaw]

Thanks to the great work of @ZongyuLi-umich and @SoniaMinseoKim , now we add the support of SPECT and CT operators & demos!