TC-MGAN

Pytorch implementation for multi-modality GAN with tumor-consistency loss for brain MR image synthesis

Magnetic Resonance (MR) images of different modalities can provide complementary information for clinical diagnosis, but whole modalities are often costly to access. Most existing methods only focus on synthesizing missing images between two modalities, which limits their robustness and efficiency when multiple modalities are missing. To address this problem, we propose a multi-modality generative adversarial network (MGAN) to synthesize three high-quality MR modalities (FLAIR, T1 and T1ce) from one MR modality T2 simultaneously. The experimental results show that the quality of the synthesized images by our proposed methods is better than the one synthesized by the baseline model, pix2pix. Besides, for MR brain image synthesis, it is important to preserve the critical tumor information in the generated modalities, so we further introduce a multi-modality tumor consistency loss to MGAN, called TC-MGAN. We use the synthesized modalities by TC-MGAN to boost the tumor segmentation accuracy, and the results demonstrate its effectiveness.

Dependencies

Python 3.6
PyTorch 1.1.0

Dataset

Download 285 cases training dataset of BRATS18.

Then move all the 285 cases to ./brats18_dataset/train/

run the folowing command to pre-process, split and save the data as .npy files.

python generate_data.py

The saved files are as below.

├── brats18_dataset
│   ├── npy_gan
│   │   └── gan_flair.npy
│   │   └── gan_t1ce.npy
│   │   └── gan_f2.npy
│   │   └── gan_t1.npy
│   ├── npy_test
│   │   └── test_flair.npy
│   │   └── test_t1ce.npy
│   │   └── test_f2.npy
│   │   └── test_t1.npy
│   ├── npy_train
│   │   └── train_flair.npy
│   │   └── train_t1ce.npy
│   │   └── train_f2.npy
│   │   └── train_t1.npy

Train network

Firstly, train a conditional unet which will be used for GAN training later.
```
python train_pre_unet.py
```
Then, train the GAN model
```
python train_gan.py
```

Test network

python test_gan.py

the generated images will be saved as below.

├── brats18_dataset
│   ├── npy_pred
│   │   └── pred_flair.npy
│   │   └── pred_t1ce.npy
│   │   └── pred_t1.npy

Demo

You can download pretrained model here [BaiduNet](code:qnw3), and run the command below to try the demo.

python demo.py

License

This work is licensed under MIT License. See LICENSE for details.

If you find our code/models useful, please consider citing our paper:

@inproceedings{Xin2020Multi,
  author = {Xin, Bingyu and Hu, Yifan and Zheng, Yefeng and Liao, Hongen},
  title = {Multi-Modality Generative Adversarial Networks With Tumor Consistency Loss for Brain MR Image Synthesis},
  booktitle = {The IEEE International Symposium on Biomedical Imaging (ISBI)},
  year = {2020}
}

Acknowledgments

Codes borrowed a lot from pix2pix and StarGAN.

hellopipu / TC-MGAN