Unified Attentional Generative Adversarial Network for Brain Tumor Segmentation From Multimodal Unpaired Images

Installation

• Run on python3.6, Pytorch0.4 and CUDA 8.0.
• Install tensorboardX.
• Clone this repo.

Data Preparation

1. Download the Task01_BrainTumour dataset from Medical Segmentation Decathlon.
2. Pre-process, save as png files and split train-test list.

   
   cd process

python to_png.py --brain_dir /path/to/Task01_BrainTumour \ --save_dir /path/to/png_dataset \ --crop 200 --resize 128

python split.py --brain_dir /path/to/png_dataset \ --save_dir .

## Train
All model will stop at `max_epoch` and make predictions at the last epoch.
```bash
cd ..
./uagan.sh

Acknowledgement

Part of the code is revised from

• yunjey/stargan
• JorisRoels/domain-adaptive-segmentation

Citation

@inproceedings{yuan2019unified,
  title={Unified attentional generative adversarial network for brain tumor segmentation from multimodal unpaired images},
  author={Yuan, Wenguang and Wei, Jia and Wang, Jiabing and Ma, Qianli and Tasdizen, Tolga},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={229--237},
  year={2019},
  organization={Springer}
}