deep_diffusion is a reinforcement learning framework for generating asymmetric diffusion encoding gradient waveforms for high-resolution magnetic resonance diffusion imaging. It uses varying deep learning agents to learn the diffusion magnetic resonance imaging parameter space and generate optimized spin echo sequences. This allows multi-objective optimization for finding waveforms allowing high encoding efficiency, low echo time (i.e. high SNR), improving motion sensitivity and concomitant field effects. It is written in Python 3 using the deep learning library Keras.
Install from Github source:
git clone https://github.com/alen-mujkanovic/deep_diffusion.git
cd deep_diffusion
python setup.py install
If you want to run the examples, you'll also have to install:
- keras-rl by Matthias Plappert:
pip install keras-rl - gym by OpenAI:
pip install gym - h5py: simply run
pip install h5py
If you use deep_diffusion in your research, you can cite it as follows:
@misc{mujkanovic2018deepdiffusion,
author = {Alen Mujkanović},
title = {deep_diffusion},
year = {2018},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/alen-mujkanovic/deep_diffusion}},
}- keras-rl, Plappert, 2016 [GitHub]
- OpenAI Gym, Brockman et al., 2016 [GitHub]
- Playing Atari with Deep Reinforcement Learning, Mnih et al., 2013
- Human-level control through deep reinforcement learning, Mnih et al., 2015