search

Project-MONAI / GenerativeModels

MONAI Generative Models makes it easy to train, evaluate, and deploy generative models and related applications
Apache License 2.0
626 stars 87 forks source link

Implement Score-MRI #182

Open HJ-harry opened 1 year ago

HJ-harry commented 1 year ago

Score-MRI (paper, code) is one of the simplest MRI reconstruction algorithms that can be implemented with a general pre-trained checkpoint.

Once a diffusion model checkpoint is ready in the model zoo, I can start making a tutorial about this.

Questions:

  1. In order to include realistic components that are relevant to MRI reconstruction (e.g. MRI sampling pattern, coil sensitivity map estimation in parallel imaging), it would be much easier if we could include e.g. sigpy as a dependency. Is this okay, or something to be avoided?
  2. Another related question: forward/inverse Fourier transforms and some of the related utilities can either be implemented from scratch, or taken from e.g. fastmri. Which way should I try?
Warvito commented 1 year ago

Hi @HJ-harry , thank you for helping with the components for the MRI Reconstruction application! I think your Score-MRI method would be a great addition to the package!

About your question:

  1. It is not an issue to add new dependencies like sigpy. As you mentioned, this dependency looks quite essential, with several features that could be used not only for your method but for others related to MRI-reconstruction. In the CONTRIBUTING.md file in the MONAI core, they mention that dependencies can be added, and it would be included as an optional dependency when installing MONAI.

  2. Would it be regarding the functions you used here? I guess if it is a small part of the fastMRI, it would not be an issue to use a similar approach that you did in your repository. Maybe, it might be good to include at the beginning of the file something like this:

    # =========================================================================
# Adapted from https://github.com/facebookresearch/fastMRI
# which has the following license:
# https://github.com/facebookresearch/fastMRI/blob/main/LICENSE.md
# =========================================================================
HJ-harry commented 1 year ago
  1. Great
  2. Sounds good, will do. Thanks! :)