ACN

This repo holds code for ACN: Adversarial Co-training Network for Brain Tumor Segmentation with Missing Modalities. (MICCAI 2021 Accepted)

Usage

Dataset

You need to download the BraTS2018 or other multi-modality datasets into <root_dir>/ACN/data The dataset directory should have this basic structure (BraTS as an example):

<root_dir>/ACN/data/<DATA_NAME>/*/case_name/*_flair.nii.gz      
<root_dir>/ACN/data/<DATA_NAME>/*/case_name/*_t1.nii.gz   
<root_dir>/ACN/data/<DATA_NAME>/*/case_name/*_t1ce.nii.gz   
<root_dir>/ACN/data/<DATA_NAME>/*/case_name/*_flair.nii.gz
<root_dir>/ACN/data/<DATA_NAME>/*/case_name/*_seg.nii.gz     # groundtruth 

Pre-requsites

Python 3.6
Pytorch >= 0.4.1
CUDA 9.0 or higher

Please use the command pip install -r requirements.txt for the dependencies.

Train/Val

Run the code for both train and validation on a multi-modality dataset. Note: This is an example for training a model when only T1ce modality is available.

python train_val_ACN.py

Citation

If you find this paper or code useful for your research, please cite our paper:

@misc{wang2021acn,
      title={ACN: Adversarial Co-training Network for Brain Tumor Segmentation with Missing Modalities}, 
      author={Yixin Wang and Yang Zhang and Yang Liu and Zihao Lin and Jiang Tian and Cheng Zhong and Zhongchao Shi and Jianping Fan and Zhiqiang He},
      year={2021},
      eprint={2106.14591},
      archivePrefix={arXiv},
      primaryClass={eess.IV}
}

Acknowledgement

This repo is borrowed from the Reproduction of BraTS18 top1's solution and ADVENT

TO DO

This is an initial version, we will re-organize it after the final publication.