Code for reproducing the results from our paper on cnn-based medical image segmentation

For any questions or issues, contact Baris via: [email protected].

Whenever you use this code, please refer to our publication

@article{kayalibay2017cnn,
  title={CNN-based Segmentation of Medical Imaging Data},
  author={Kayalibay, Baris and Jensen, Grady and van der Smagt, Patrick},
  journal={arXiv preprint arXiv:1701.03056},
  year={2017}
}

Requirements:

• Python 2.7
• Theano (0.9.0) (along with CUDA and cuDNN)
• cudamat
• this fork of breze
• climin
• h5py
• SimpleITK

Some of these requirements can be installed with pip install package (theano, h5py, SimpleITK), others (breze, cudamat, climin) should be cloned from the github links provided and installed via pip install -e .

Usage:

To segment new images and get test results, you will need to train the network first. The following steps need to be taken to create a data set, train and segment new images:

Acquire the BRATS 2015 data set:

Go to the official brats website and download the BRATS 2015 data. Store the training data in this directory under a directory called BRATS2015_Training.

Create a data set:

Run the following line on the terminal:

python brain_data_scipts/read_images.py

This will create a .hdf5 file called brats_fold0.hdf5 under data/datasets. This .hdf5 file contains three randomly created partitions train, valid and test for training, validation and testing. You can now use it to train a neural network.

Train the network:

Run:

python train.py fcn_rffc4 brats_fold0 brats_fold0 600 -ch False

This will train the network used in our paper on the data set brats_fold0 for 600 iterations over the data set and store the results at the path models/brats_fold0.

Test or reuse the trained network:

Once you've trained a network, its parameters and hyperparameters are stored in a subdirectory of the directory models (read the docstring of the module train.py on how to select this). You can then reuse those parameters using the API provided in the module segment.py. In segment.py you will find a function segment that can be used in the following way to segment new images:

segment('BRATS2015_Training/HGG/brats_2013_pat0001_1', 'results2', 'fcn_rffc4', 5)

In this example snippet, we are using the network with the id fcn_rffc4 along with the parameters stored at models/results2 to segment a medical image contained at the path BRATS2015_Training/HGG/brats_2013_pat0001_1. The general usage is:

segment([image_path], [params_path], [model_id])