# SAML & A Multi-site Dataset for Prostate MRI Segmentation

by [Quande Liu](https://github.com/liuquande), [Qi Dou](http://www.cse.cuhk.edu.hk/~qdou/), [Pheng-Ann Heng](http://www.cse.cuhk.edu.hk/~pheng/). 

### Introduction

* The Tensorflow implementation for our MICCAI 2020 paper '[Shape-aware Meta-learning for Generalizing Prostate MRI Segmentation to Unseen Domains](https://arxiv.org/pdf/2007.02035.pdf)'. 

<p align="center">

  <img src="figure/saml.png"  width="650"/>

</p>

* A well-organized multi-site dataset (from six data sources) for prostate MRI segmentation, that can support research in various problem settings with need of multi-site data, such as Domain Generalization, Multi-site Learning and Life-long Learning, etc. For more details and downloading link of the dataset, please [Find Here](https://liuquande.github.io/SAML/).

<p align="center">

  <img src="figure/protocol.png"  width="650"/>

</p>

### Setup & Usage for the Code

1. Check dependencies:

   ```shell

   python==2.7.17

   numpy==1.16.6

   scipy==1.2.1

   tensorflow-gpu==1.12.0

   tensorboard==1.12.2

   SimpleITK==1.2.0

   ```

2. To train the model, you need to specify the training configurations (can simply use the default setting) in main.py, then run:

   ```shell

   python main.py --phase=train

   ```

2. To evaluate the model, run:

   ```shell

   python main.py --phase=test --restore_model='/path/to/test_model.cpkt'

   ```

   You will see the output results in the folder `./output/`.

### Citation

If this repository is useful for your research, please cite:

```

@article{liu2020shape,

  title={Shape-aware Meta-learning for Generalizing Prostate MRI Segmentation to Unseen Domains},

  author={Liu, Quande and Dou, Qi and Heng, Pheng-Ann},

  journal={International Conference on Medical Image Computing and Computer Assisted Intervention},

  year={2020}

}

```

### Questions

For further question about the code or dataset, please contact 'qdliu@cse.cuhk.edu.hk'