Awesome
MRI Reconstruction Using Deep Energy-Based Model
The Code is created based on the method described in the following paper: MRI Reconstruction Using Deep Energy-Based Model.
NMR in Biomedicine
https://analyticalsciencejournals.onlinelibrary.wiley.com/doi/10.1002/nbm.4848
Author: Yu Guan, Zongjiang Tu, Shanshan Wang, Yuhao Wang, Qiegen Liu*, Dong Liang*.
Date : Sep. 7, 2021
Version : 1.0
The code and the algorithm are for non-comercial use only.
Copyright 2021, Department of Electronic Information Engineering, Nanchang University.
Overview of the MRI reconstruction.
<div align="center"><img src="https://github.com/yqx7150/EBMRec/blob/main/Figs/Fig1.png" width = "815" height = "470"> </div>Detailed comparison of characteristics and structures in the flow chart of GAN and EBM.
<div align="center"><img src="https://github.com/yqx7150/EBMRec/blob/main/Figs/Fig2.png" width = "781" height = "450"> </div>Complex-valued reconstruction results on brain images at R=3 various 1D Cartesian under-sampling percentages in 15 coils parallel imaging.
<div align="center"><img src="https://github.com/yqx7150/EBMRec/blob/main/Figs/Fig5.png" width = "844" height = "556"> </div>Complex-valued reconstruction results on brain image at R=6 pseudo random sampling in 12 coils parallel imaging.
<div align="center"><img src="https://github.com/yqx7150/EBMRec/blob/main/Figs/Fig6.png" width = "893" height = "558"> </div>Pretrained Models
We provide pretrained checkpoints. You can download pretrained models from Baidu Drive. key number is "gygy "and unzip into the folder cachedir.
Train
If you want to train the code，please
python3 EBM_train.py --exp=fastMRI256 --dataset=fastMRI --num_steps=50 --batch_size=16 --step_lr=100 --lr=3e-4 --zero_kl --replay_batch --ResNet128_model --cclass --swish_act
All code supports horovod execution, so model training can be increased substantially by using multiple different workers by running each command.
mpiexec -n <worker_num> <command>
For example: "mpiexec --oversubscribe -n 1" or "mpiexec --oversubscribe -n 4"
Test
If you want to test the code，please
python3 EBM_test.py --exp=siat256 --resume_iter=164250 --step_lr=300 --swish_act
python3 EBM_test_ddp.py --exp=siat256 --resume_iter=164250 --step_lr=50 --swish_act
python3 EBM_test_modl.py --exp=siat256 --resume_iter=164250 --step_lr=10 --swish_act
Acknowledgement
The implementation is based on this repository: https://github.com/openai/ebm_code_release.