Deep Diffusion

What is it?

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.

Installation

Install from Github source:

git clone https://github.com/alen-mujkanovic/deep_diffusion.git
cd deep_diffusion
python setup.py install

Examples

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

Citing

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}},
}

References

1. keras-rl, Plappert, 2016 [GitHub]
2. OpenAI Gym, Brockman et al., 2016 [GitHub]
3. Playing Atari with Deep Reinforcement Learning, Mnih et al., 2013
4. Human-level control through deep reinforcement learning, Mnih et al., 2015