DeepGF: Glaucoma Forecast Using the Sequential Fundus Images
- This is the official repository of the paper "DeepGF: Glaucoma Forecast Using the Sequential Fundus Images" from MICCAI 2020[Paper Link][PDF Link]
1. Environment
- Python >= 3.5
- Tensorflow >= 1.4 is recommended
- opencv-python
- sklearn
- matplotlib
2. Dataset
- The training data and testing data is from the [SIGF-database]. Contact [liu.li20@imperial.ac.uk] or [xfwang@buaa.edu.cn] for password of the shared data in dropbox. Below is an example of our SIGF database.
- Put the training and test images and the labels in the directory:
'./data/train(test)/image(label)/all/'
- Obtain the polar and attention data from the [Dropbox]. Below is an example of the polar and attention map of a glaucoma fundus image.
- Put the attention and polar images in the directory:
'./data/'
3. Training
The details of the hyper-parameters are all listed in the train.py
. Use the below command to train our model on the SIGF database.
python ./train.py
4. Test
Download the pre-trained model in [Dropbox]. Then put the file in tghe directory of
pretrained_model
. Use the below command to test the model on the SIGF database.
python ./test.py
5. Compared Methods
The network re-implenmentation of [Chen et al.] is in the file of:
chen_net.py
and from the directory of ./Compared Methods
6. Ablation Study
If you are interested in our ablation study, please see ./Ablation study
7. Network Interpretability
-
If you are interested in the visualization method and results used for showing the interpretability of our method, please refer to the directory of
./saliency
-
Or you can just see the images in the directory of
./visualization_result
for more visualization results. Some examples of the visualization rsults are shown here.
8. Citation
If you find our work useful in your research or publication, please cite our work:
@article{Li2020deep,
title={DeepGF: Glaucoma Forecast Using the Sequential Fundus Images.},
author={Li, Liu and Wang, Xiaofei and Xu, Mai and Liu, Hanruo},
journal={MICCAI},
year={2020}
}
9. Contact
If any question, please contact [xfwang@buaa.edu.cn]