Diffusion convolution recurrent neural network for brain network analysis

This is the implementation of the graph neural network model used in our paper:

S. Wein, W. M. Malloni, A. M. Tomé, S. M. Frank, G. -I. Henze, S. Wüst, M. W. Greenlee & E. W. Lang, A graph neural network framework for causal inference in brain networks, Scientific Reports 11, 8061 (2021).

The implementation is based on the DCRNN proposed by:

Y. Li, R. Yu, C. Shahabi & Y. Liu, Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecasting, ICLR 2018.

This model was also used in our recent manuscript:

S. Wein , A. Schüller, A. M. Tome, W. M. Malloni, M. W. Greenlee, and E. W. Lang, Forecasting brain activity based on models of spatiotemporal brain dynamics: A comparison of graph neural network architectures, Network Neuroscience 6 (3): 665–701 (2022).

Requirements

• scipy>=0.19.0
• numpy>=1.12.1
• pyaml
• tensorflow=1.xx

Also a conda environment.yml file is provided. The environment can be installed with:

conda env create -f environment.yml

Run demo version

A short demo version is included in this repository, which can serve as a template to process your own MRI data. Artificial fMRI data is provided in the directory MRI_data/fMRI_sessions/ and the artificial timecourses have the shape (nodes,time). The adjacency matrix in form of the structural connectivity (SC) between brain regions can be stored in MRI_data/SC_matrix/. An artificial SC matrix with shape (nodes,nodes) is also provided in this demo version.

The training samples can be generated from the subject session data by running:

python generate_samples.py --input_dir=./MRI_data/fMRI_sessions/ --output_dir=./MRI_data/training_samples

The model can then be trained by running:

python dcrnn_for_brain_connectivity_train.py --config_filename=./configs/dcrnn_demo_config.yaml --save_predictions=True

A Jupyter Notebook version is provided, which can be directly run in Google Colab with:

https://colab.research.google.com/github/simonvino/DCRNN_brain_connectivity/blob/main/dcrnn_for_brain_connectivity_colab_demo.ipynb

Data availability

Preprocessed fMRI and DTI data from Human Connectome Project data is publicly available under: https://db.humanconnectome.org.

A nice tutorial on white matter tracktography for creating a SC matrix is available under: https://osf.io/fkyht/.

Citations

If you apply this graph neural network model for MRI analysis, please cite the following paper:

@article{Wein2021,
  title = {A graph neural network framework for causal inference in brain networks},
  author = {Wein, Simon and Malloni, Wilhelm and Tomé, Ana and Frank, S. and Henze, Gina-Isabelle and Wüst, S. and Greenlee, Mark and Lang, Elmar},
  year = {2021},
  month = {04},
  volume = {11},
  journal = {Scientific Reports},
  doi = {10.1038/s41598-021-87411-8}
}

And the model architecture was originally proposed by Li et al.:

@inproceedings{li2018dcrnn_traffic,
  title={Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecasting},
  author={Li, Yaguang and Yu, Rose and Shahabi, Cyrus and Liu, Yan},
  booktitle={International Conference on Learning Representations (ICLR '18)},
  year={2018}
}

Other implementations

A PyTorch implementation of the DCRNN model can be found at: https://github.com/chnsh/DCRNN_PyTorch

A Graph WaveNet implementation for brain network analysis can be found at: https://github.com/simonvino/GraphWaveNet_brain_connectivity