This repository contains the code for the following paper: (https://arxiv.org/abs/1711.05702)

@misc{Han2017Brain,
    title        = {Brain Extraction from Normal and Pathological Images: A Joint PCA/Image-Reconstruction Approach},
    author       = {X. Han and R. Kwitt and S. Aylward and S. Bakas and B. Menze and A. Asturias and P. Vespa and J. Van Horn and M. Niethammer},
    year         = {2017},
    howpublished = {arXiv:1711.05702}}

We support use of docker. If you want to use the GPU for decomposition, please follow below to install nvidia-docker: https://github.com/NVIDIA/nvidia-docker

We also include a sample PCA basis images as well as atlases. You can download the zip file at (https://github.com/uncbiag/PStrip/releases/download/v1.0.0-alpha/data.zip). Unzip to the data folder. You can use your own datas, but you need to make sure the file name and the folder strucure are the same as mine.

The following script can be used, for example:

nvidia-docker run -i --rm -v=your_image_folder:/input:rw -t xhs400/pstrip:latest python /PStrip/main_code/pstrip.py -i /input/your_image_name

docker run -i --rm -v=your_image_folder:/input:rw -t xhs400/pstrip:latest python /PStrip/main_code/pstrip.py -i /input/your_image_name -p CPU

nvidia-docker run -i --rm -v=your_image_folder:/input:rw -t xhs400/pstrip:latest python /PStrip/main_code/pstrip.py -f /input

docker run -i --rm -v=your_image_folder:/input:rw -t xhs400/pstrip:latest python /PStrip/main_code/pstrip.py -f /input -p CPU

What the above script does are following:

1. Pull the docker repository. (https://hub.docker.com/r/xhs400/pstrip/)
2. Initiate a docker container. (Mount the image_folder to '/input').
3. Run the python pstrip.py code: the first two perform brain extraction on the input image and the results will be saved in the same folder, using GPU and CPU respectively; the last two perform brain extraction on all images inside the input folder and the results will be saved in the same folder, using GPU and CPU respectively

For all options for pstrip.py please go to main_code folder, or you can run

python pstrip.py -h

1. Pull the docker repository. (https://hub.docker.com/r/xhs400/pstrip/)

docker pull xhs400/pstrip

2. Run the docker container

nvidia-docker run -i --rm=false -v=your_image_folder:/input:rw pstrip:latest

Once log into the docker container, the source code is located in the '/PStrip'

1. Download the Dockerfile in this repository or at (https://hub.docker.com/r/xhs400/pstrip/~/dockerfile/)
2. Build the dockerfile.
3. Run the docker container

Even if you don't use docker, the Dockerfile is the best reference to build our code on your own.
To test the source code, you may go to PStrip/main_code and run the following code to view all the options

python pstrip.py --help

Or you may run

python pstrip.py -i ../data/atlas/atlas_w_skull.nii  

which simply tries to extract the brain from atlas image.
For users without GPU, you can add '-p CPU'.

This section introduces the main steps to build PStrip. The exact command can be found in Dockerfile, which is the best referece.
This code has been tested on Ubuntu 14.04 with Python 2.7 (Nvidia Titan X GPU, cuda 8.0)

2017/10/19: CPU implementation is added. (No need for cuda, if only cpu is used)
2017/10/06: the code works on Ubuntu 16.04 with Python 2.7 (cuda 9.0)

Download and Install virtualenv (Option for no-sudo privilege user) Download virtualenv

wget https://github.com/pypa/virtualenv/archive/15.1.0.tar.gz

Untar virtualenv

tar -xvzf 15.1.0.tar.gz

Install virtualenv

cd virtualenv-15.1.0
python virtualenv.py ~/  (assume ~ is the location for virtual environment)

Start virtualenv

source ~/bin/activate

Download and Install Numpy

Follow https://scipy.org/install.html

Download and Install SimpleITK

pip install SimpleITK

Download and Install CUDA (GPU is used for decompositions only)

CPU implementation of the decompostion is relatively slow, and we haven't uploaded it.

Download and Install niftyreg https://cmictig.cs.ucl.ac.uk/wiki/index.php/NiftyReg_install

For future purpose, in order to use the GPU, set the CUDA_FLAG to on. (Currently all registration steps are using CPU only)

Download and Install pycuda/scikit-cuda Follow https://wiki.tiker.net/PyCuda/Installation/Linux

obtain the scikit-cuda code at https://github.com/lebedov/scikit-cuda

2017/10/06: Do NOT use pip install for scikit-cuda, although it is instructed from scikit-cuda webside

DO NOT USE pip install scikit-cuda

I receive OSError: cusolver library not found, if installing from pip.