Edge Variational Graph Convolutional Networks for Disease Prediction

About

This is a Pytorch implementation of EV-GCN described in Edge-variational Graph Convolutional Networks for Uncertainty-aware Disease Prediction (MICCAI 2020) by Yongxiang Huang and Albert C.S. Chung.

Prerequisites

• Python 3.7.4+
• Pytorch 1.4.0
• torch-geometric
• scikit-learn
• NumPy 1.16.2

Ensure Pytorch 1.4.0 is installed before installing torch-geometric.

This code has been tested using Pytorch on a GTX1080TI GPU.

Training

python train_eval_evgcn.py --train=1

To get a detailed description for available arguments, please run

python train_eval_evgcn.py --help

To download the used dataset, please run the following script in the data folder:

python fetch_data.py 

If you want to train a new model on your own dataset, please change the data loader functions defined in dataloader.py accordingly, then run python train_eval_evgcn.py --train=1

Inference and Evaluation

python train_eval_evgcn.py --train=0

Reference

If you find this code useful in your work, please cite:

@inproceedings{huang2020edge,
  title={Edge-Variational Graph Convolutional Networks for Uncertainty-Aware Disease Prediction},
  author={Huang, Yongxiang and Chung, Albert CS},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={562--572},
  year={2020},
  organization={Springer}
}