README.md

NVUM

This repository is the official implementation of MICCAI 2022 (Early Accept) NVUM: Non-volatile Unbiased Memory for Robust Medical Classification.

Requirements

• To install requirements:

  pip install -r requirements.txt
  

• 1 * NVIDIA RTX 2080ti

Datasets Preparation

Refer to dataset_preparation

Training

Train on NIH

python train.py --trim_data --run_name NVUM_NIH --train_root_dir <NIH_dir> --openi_root_dir <OPI_dir> --pc_root_dir <PDC_dir> --save_dir <save_dir> --nvcm --lm

Train on CXP

python train.py --trim_data --run_name NVUM_CXP --train_root_dir <CXP_dir> --openi_root_dir <OPI_dir> --pc_root_dir <PDC_dir> --save_dir <save_dir> --batch_size 64 --lr 0.0001 total_epochs 40 --resize 224 --num_classes 8 --nvcm --lm --train_data CXP 

Credits

• Part of the code are from TorchXrayVision.

Citation

If you find this repo useful for your research, please consider citing our paper:

@article{liu2021NVUM,
  title={NVUM: Non-volatile Unbiased Memory for Robust Medical Classification},
  author={Liu, Fengbei and Yuanhong Chen and Tian, Yu and Yuyuan Liu and Chong Wang and Belagiannis, Vasileios and Carneiro, Gustavo},
  journal={arXiv preprint arXiv:2103.04053v2},
  year={2021}
}