pOSAL: Patch-based Output Space Adversarial Learning for Joint Optic Disc and Cup Segmentation.

---

We provide the Keras implements based on Tensorflow Backend for REFUGE challenge segmentation task. <img src="https://github.com/EmmaW8/pOSAL/blob/master/imgs/overview.png" width="800px"/>

Getting Started

Install requirments

conda create -n posal python=3.5
conda activate posal
pip install keras==2.2.0
pip insatll tensorflow-gpu==1.4.0
conda install tqdm
conda install -c anaconda scikit-image
conda install opencv

Prerequisites

• GPU, CUDA=9.0

Running Evaluation

• Clone this repo:

git clone https://github.com/EmmaW8/pOSAL.git
cd pOSAL

To reproduce the results for the rank in REFUGE challenge in MICCAI 2018, please do

python predict.py 0 # 0 is the avaliable GPU id, change is neccesary

Running Training for Dri-GS dataset

Remember to check/change the data path and weight path

python train_DGS.py 0
python test_DGS.py 0

The CDR values used for glaucoma diagnsis are generated with MATLAB.

cd matlab-code

Please change the input and output path in the generate_CDR_values.m file.

Acknowledge Some codes are revised according to selimsef/dsb2018_topcoders, HzFu/MNet_DeepCDR and evaluateion code . Thank them very much.

Citation

@article{wang2019patch,
  journal={IEEE Transactions on Medical Imaging},
  title={Patch-Based Output Space Adversarial Learning for Joint Optic Disc and Cup Segmentation},
  author={Wang, Shujun and Yu, Lequan and Yang, Xin and Fu, Chi-Wing and Heng, Pheng-Ann},
  year={2019},
  volume={38},
  number={11},
  pages={2485-2495},
  publisher={IEEE},
  doi={10.1109/TMI.2019.2899910},
  }