Identifying Autism Spectrum Disorder Based on Individual-Aware Down-Sampling and Multi-Modal Learning

This is the GitHub repository for our paper: Identifying Autism Spectrum Disorder Based on Individual-Aware Down-Sampling and Multi-Modal Learning

Requirements

torch == 1.8.1+cu102

torch-cluster == 1.5.9

torch-geometric == 1.7.0

torch-scatter == 2.0.7

torch-sparse == 0.6.10

sklearn

nilearn

Install torch-geometric:

pip install torch-scatter -f https://data.pyg.org/whl/torch-1.8.1+cu102.html
pip install torch-sparse -f https://data.pyg.org/whl/torch-1.8.1+cu102.html
pip install torch-cluster -f https://data.pyg.org/whl/torch-1.8.1+cu102.html
pip install torch-geometric

Download ABIDE I dataset

python download_ABIDE.py

This script will automatically download ABIDE I dataset preprocessed by CPAC and construct the graph representation of the brain imaging. After downloading, the raw data (i.e. fMRI) will be deleted. The graph representation of brain will be saved at './data/ABIDE/'.

Run the diagnosis framework

Run the framework:

python main.py

main.py includes:

1. The unsupervised graph pooling. The results will by default be saved to /data/ABIDE_downsample/.
2. Training phase of the MLP under nested 10-fold cross-validation
3. Use the best model from the trained MLP to extract the further learned features from pooling results under the same cross-validation of 2.
4. train/test classifiers under the same cross-validation of 2.

Evaluation

python visualize.py --roc

will plot the ROC curve and confusion matrix for the framework using the classification results saved to /results/

Visualization

python visualize.py --embedding --group=gender

will compare the 2D distribution difference between the features learned by MLP and the node embeddings later learned by GCN and renders the nodes according to different groups (gender, age, site). The parameter --group could be 'gender', 'site', or 'age'.

Citation

@article{pan2021identifying,
  title={Identifying Autism Spectrum Disorder Based on Individual-Aware Down-Sampling and Multi-Modal Learning},
  author={Pan, Li and Liu, Jundong and Shi, Mingqin and Wong, Chi Wah and Chan, Kei Hang Katie},
  journal={arXiv preprint arXiv:2109.09129},
  year={2021}
}