Skip to content
Try several methods for MRI reconstruction on the fastmri dataset.
Jupyter Notebook Python Other
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.


Type Name Latest commit message Commit time
Failed to load latest commit information.
fastmri_recon H5 deprecation warning (#41) Feb 10, 2020

fastMRI reproducible benchmark


The idea of this repository is to have a way to rapidly benchmark new solutions against existing reconstruction algorithms on the fastMRI dataset single-coil track. The reconstruction algorithms implemented or adapted to the fastMRI dataset include to this day:

All the neural networks (except the U-net) are implemented in both keras and pytorch. I mainly used keras to develop, but I realized at some point that pytorch might just be faster for fourier transform operations (see However, the main documentation is still for the keras models.

How to train the neural networks

The scripts to train the neural networks are located in fastmri_recon/training/. You just need to install the package and its dependencies:

pip install . &&\
pip install -r fastmri_recon/requirements.txt

TensorFlow is not listed as a dependency to let you chose if you want gpu supported TensorFlow.

How to write a new neural network for reconstruction

A good example of a simple neural network on which you can improve is the zerofill_net which is simply performing zero-filled reconstruction using keras. The building blocks can then be found in fastmri_recon/helpers/


This work will be presented at the International Symposium on Biomedical Imaging (ISBI) in April 2020. If you use this package or parts of it, please cite the following work: Benchmarking Deep Nets MRI Reconstruction Models on the FastMRI Publicly Available Dataset

You can’t perform that action at this time.