Skip to content

Latest commit

 

History

History
46 lines (34 loc) · 2.23 KB

README.md

File metadata and controls

46 lines (34 loc) · 2.23 KB

SEFRN-MRI

This is the official code hub for the SEFRN: Lightweight and Accurate Self-Ensemble Feedback Recurrent Network for MRI Reconstruction

Requirements

  • Python==3.6.6
  • numpy==1.14.3
  • opencv-python==3.4.1.15
  • scipy==1.1.0
  • pytorch == 1.7.0+cu110
  • matplotlib==2.2.2
  • scikit-image == 0.15.0
  • h5py

How to Train

  1. Prepare data.
  2. run run_job.sh.

Prepare Data

1 Cardiac dataset: The original data is established from the work Alexander et al. Details can be found in the paper. You can download the original data from Here.

For us, we use the converted png images provided by Here, and the convert code is Here. The data should be placed in ./data/cardiac_ktz/.

2 Brain dataset: this dataset is establisded by Souza et al. you can download it from Here

3 FastMRI dataset: This is a large-scale MR dataset jointly established by Facebook AI Research and NYU Langone Health. In our work, we use the single-coil Knee MR data for model evaluation. Data can be download from Here

3 In our training process, we pre-generate a quantity of random sampling masks in the mask/, named like mask_rAMOUT_SAMPLINGRATE.mat. These masks will be applied in for the Cardiac dataset; For the other datasets, we use the masking function provided by FastMRI.

Some Visual Result

For the Cardiac dataset, we plot the motion of our predicted image under sampling rate 15%, in order to check whether the predicted motion is normal. Please see the figure below.

Example 1 (From left to right: Zero-Filled, Ground Truth and Predicted Image)

Example 2 (From left to right: Zero-Filled, Ground Truth and Predicted Image)