An experimental framework for evaluating the performance for various techniques for segmenting the BraTS 2018 data.

These instructions are for how to build and run this framework on an Azure Ubuntu VM which is the environment in which this work was originally done.

The BraTS 2018 data is required to run this module. The data can be found here. It is required to make an account and request the data.

Download and install Anaconda.

Install Nvidia toolkit:

sudo apt-get install nvidia-cuda-toolkit 

Once you've initialized a new Conda environment install the following packages: Most importantly, install PyTorch:

# torch
conda install pytorch torchvision cudatoolkit=9.0 -c pytorch

Note that this will install torch for Cuda 9.0. You should check that this is the correct version by running

nvcc --version 

Then

# nibabel
conda install -c conda-forge nibabel

# tqdm
conda install -c conda-forge tqdm