- 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]
- Python >= 3.5
- Tensorflow >= 1.4 is recommended
- opencv-python
- sklearn
- matplotlib
- matlab
-
The training and test fundus images are from the [LAG-database].
-
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]
-
Obtain the ground-truth for visualization from [dropbox].
-
Put the data into the directory-tree of
./img_data/OURS/image(vessel)(label)
Refer to the data_processing.py
to generate the .tfrecord
files.
python main.py
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.
- The network visualization results
- The image synthesis results
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}
}
If you have any questions, please contact [xfwang@buaa.edu.cn]