Pathology-aware deep network visualization and its application in glaucoma image synthesis

• This is the official repository of the paper "Pathology-aware deep network visualization and its application in glaucoma image synthesis" from MICCAI 2019[Paper Link][PDF Link]

1. Environment

• Python >= 3.5
• Tensorflow >= 1.4 is recommended
• opencv-python
• sklearn
• matplotlib
• matlab

2. Dataset

1. The training and test fundus images are from the [LAG-database].

2. The vessel images corresponding to the fundus images can be generated using the method in the paper [Vessel_segment]. A recent re-implementation of the method can be seen in [retina-segmentation-unet]

3. Obtain the ground-truth for visualization from [dropbox].

4. Put the data into the directory-tree of

./img_data/OURS/image(vessel)(label)

3. Prepare

Refer to the data_processing.py to generate the .tfrecord files.

4. Training and validation

    python main.py 

5. Results Display

Actually, there are two tasks as described in our paper, i.e., the network visualization task and image synthesis task. As a result, here we show some subjective results of thes two tasks.

1. The network visualization results

2. The image synthesis results

6. Citation

If you find our work useful in your research or publication, please cite our work:

@inproceedings{wang2019pathology,
  title={Pathology-aware deep network visualization and its application in glaucoma image synthesis},
  author={Wang, Xiaofei and Xu, Mai and Li, Liu and Wang, Zulin and Guan, Zhenyu},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={423--431},
  year={2019},
  organization={Springer}
}

7. Contact

If you have any questions, please contact [xfwang@buaa.edu.cn]