This repository is an official PyTorch implementation of the paper "Instance Importance-Aware Graph Convolutional Network for 3D Medical Diagnosis" [paper] from Medical Image Analysis 2022.
- Considering the high cost of collecting exhaustive annotations for 3D data, a sustainable alternative is to develop diagnosis algorithms with merely patient-level labels. We propose the Instance Importance-aware Graph Convolutional Network (I2GCN) under the multi-instance learning (MIL).
- Using a preliminary MIL classifier, we first calculate the instance importance of each slice towards diagnosis. In the refined diagnosis branch, we devise the Instance Importance-aware Graph Convolutional Layer (I2GCLayer) to exploit complementary features in both importance-based and feature-based topologies. Moreover, the importance-based Sub-Graph Augmentation (SGA) is devised to alleviate the deficient supervision of 3D dataset.
The processed CC-CCII dataset can be downloaded from Google Drive. Put the downloaded .npy files in a newly-built folder ./data/
. Please note that among the three-fold cross-validation with random split, the performance of split1
and split2
is slightly higher than the split0
.
- Python 3.6
- PyTorch >= 1.3.0
- numpy 1.19.4
- scikit-learn 0.24.2
- scipy 1.3.1
Clone this repository into any place you want.
git clone https://github.com/CityU-AIM-Group/I2GCN.git
cd I2GCN
mkdir experiment; mkdir data
- Train the I2GCN with default settings:
python ./main.py --theme default --test_split 1 --online_flag 1
We provide the dataloader with two ways of loading npy files, including online
and offline
.
If you find our work useful in your research or publication, please cite our work:
@article{CHEN2022102421,
title = {Instance Importance-Aware Graph Convolutional Network for 3D Medical Diagnosis},
author = {Zhen Chen and Jie Liu and Meilu Zhu and Peter Y.M. Woo and Yixuan Yuan},
journal = {Medical Image Analysis},
pages = {102421},
year = {2022},
issn = {1361-8415},
doi = {https://doi.org/10.1016/j.media.2022.102421}
}
- CC-CCII dataset from China National Center for Bioinformation, the largest public COVID-19 dataset of 3D lung CT scans until publication.