Artificial Neural Network for Myelin Water Imaging: ANN-MWI

Source code to train and test the network (.py), and data preprocessing (.m) from the manuscript "Artificial neural network for myelin water imaging", submitted to Magnetic Resonance in Medicine.

Overview

Two different artificial neural networks (ANN I and ANN II) for generating myelin water imaging were proposed in the manuscript.

In the trained_networks folder, final parameters from three different networks were uploaded.

(Three networks: ANNI_mwf for myelin water fraction, ANNI_gmt2 for geometric mean T2, and ANNII for T2 distribution)

Please read the usage below for details.

Requirements

• Python 2.7

• TensorFlow 1.9.0

• NVIDIA GPU (CUDA 8.0)

• MATLAB R2017b

Data acquisition

• 3T Trio MRI scanner (Siemens Healthcare, Erlangen, Germany)

• 3D multi-combined gradient and spin echo sequence

Usage

Training

• make_trainingset.m: to make the training set with normalizing data.

• train_ANNMWI.py: to train the network with the normalized training set.

Test

• make_testset.m: to make the test set with normalizing data.

• test_ANNMWI.py: to test the trained network with the normalized test set.

  For the test, you can use our results (.ckpt files) in each corresponding network folder in the trained_networks folder.

References

Please cite the following publication if you use our work (https://doi.org/10.1002/mrm.28038).

   @ARTICLE{
   author = {{Lee}, Jieun and {Lee}, Doohee and {Choi}, Joon Yul and
   {Shin}, Dongmyung and {Shin}, Hyeong-Geol and {Lee}, Jongho},
   title = "{Artificial neural network for myelin water imaging}",
   journal = {Magnetic Resonance in Medicine},
   year = "2020",
   month = "May",
   eid = {Volume 83, Issue 5},
   pages = {1875-1883},
   }