/noddi_undersampling

Use deep learning to create NODDI maps

Primary LanguagePython

Noddi processing project

This is the code for generating NODDI maps (ODI, FISO, and FICVF) rapidly and simultaneously through deep learning methods.

Running this code

This library will require the following the dependencies:

  • python3.5+

  • tensorflow1.8+ (pip install --upgrade tensorflow-gpu)

  • keras2.1+ (pip install keras)

  • webcolors (pip install webcolors)

  • ants (see their github for install)

  • matplotlib

  • scikit-image

  • numpy

  • h5py

  • scipy

  • nibabel

  • matlab.engine

If you don't have these installed, the easiest is through the anaconda virtual environments.

conda create -n [your_env_name] python=3.6

This will install the conda environment. Now you need to install each of the pacakages either through pip or conda install. You will need CUDA9.0 and CUDNN7. You can get that through

conda install cudnn

The exception to this is the matlab.engine package, which will be needed separately. This can be done by

cd "matlabroot\extern\engines\python"
python setup.py build --build-base="builddir" install --prefix="installdir"

To use the virtual environment, just run

source activate [your_env_name]

You will need to add my python_utils library to the PYTHON_PATH by

source path_setup.sh

Generating the figures for the paper

To generate the figures for the paper

cd validation
python [whatever_figure_you_want]_fig.py

This should spit out the figure window pane through x11 and also save it in the results folder (relative to the base). You will need to make the results folder first, however.