Download clinical.csv and FC/ from Open Access: The Effect of Neurorehabilitation on Multiple Sclerosis – Unlocking the Resting-State fMRI Data and add it to the root folder.

For this repository to run, Python 3.7 is required. I specifically used Python 3.7.9 and created a virtualenvironment in a env/ folder with

python -m venv env
source env/bin/activate

Install all required libraries with pip via

pip install -r requirements.txt

and clone GCN into this repository.

git clone https://github.com/tkipf/gcn .

Copy the load_data.patch and the train.patch into the cloned repository and apply the changes with

git apply load_data.patch
git apply train.patch

This modifies the load function in order to work with our data and adds a section to the train script which saves the hidden layer outputs as numpy arrays to files in gcn/gcn/embeddings.

Then install the module with

python setup.py install

The version number in requirements file of this repo should work with the versions in the gcn implementation. For training with our dataset and parameters for a single patients treatment, use:

python train.py --dataset p001_1 --learning_rate 0.01 --epochs 10 --early_stopping 10 --hidden1 10

Run the preprocessing.ipynb notebook and follow the commands. You need to execute ./generate_node_embeddings.sh at some point. Make sure that its executable with

chmod +x generate_node_embeddings.sh