Joint model-based deep learning for parallel imaging.
Reference article: J-MoDL: Joint Model-Based Deep Learning for Optimized Sampling and Reconstruction by H.K Aggarwal and M. Jacob in IEEE Journal of Selected Topics in Signal Processing, (2020).
arXiv link: https://arxiv.org/abs/1911.02945
IEEEXplore: https://ieeexplore.ieee.org/document/9122388
We have tested the codes in Tensorflow 1.15 but it can run in other versions as well.
This git repository includes a test image with coil sensitivities in the file tstdata_jmodl.npz
. The testing script tst.py
will use this image by default and does not require full data download for the testing purpose.
We have also released a subset of the parallel imaging dataset used in this paper. This data is required in the training code trn.py
. You can download the dataset from the below link:
Dataset Download : https://drive.google.com/file/d/1GLqs2A5YpRN8RdDJgdhrspL3zjlG0Qha/view?usp=sharing
You can run the test code without making any change as:
$python tst.py
. It should give following output.
In this paper, we have proposed how to learn the sampling mask together with the reconstruction network in the parallel imaging settings. As shown in this diagram we train for the real valued sampling locations rather than the entire sampling mask.
This is the structure of the overall network. Please refer to the paper for details on notations.
A_theta
is the acquisition operator having trainable sampling parameter theta
.
D_phi
is the CNN network with trainable parameters phi
.
Q_theta
is the data-consistency block that allows to use imaging physics.
The code is provided to support reproducible research. If the code is giving syntax error in your particular python configuration or some files are missing then you may open an issue or directly email me at jnu.hemant@gmail.com