Official Repository for DRG-Net

Tusfiqur, Hasan Md, Duy MH Nguyen, Mai TN Truong, Triet A. Nguyen, Binh T. Nguyen, Michael Barz, Hans-Juergen Profitlich, et al. "DRG-Net: interactive joint learning of multi-lesion segmentation and classification for diabetic retinopathy grading. arXiv preprint

• Contains the training scripts for classification and segmentation models for FGADR and EyPACS datasets
• Scripts for Additional models and other datasets will be added soon.

Contact: Hasan Md Tusfiqur Alam ([email protected])
Licenced under CC BY-NC-SA 4.0

Installation

Recommended environment:

• python 3.8+
• pytorch 1.7.1+
• torchvision 0.8.2+
• tqdm
• munch
• packaging
• tensorboard
• scikit-learn
• opencv-python
• pillow < 7
• segmentation-models-pytorch

To install the dependencies create a virtualenv and run:

$ pip install -r requirements.txt

Datasets

Datasets used in this paper can be accessed from the following links:

• Indian Diabetic Retinopathy Image Dataset - IDRiD
• FGADR Dataset - Look Deeper into Eyes.
• EyePACS Dataset

For classification

• The 'dr_classification' folder contains scripts for the dr grading classification task.
• Follow readme.md inside 'dr_classification' for instructions.

For segmentation

• The 'dr_segmentation' folder contains scripts for the dr segmentation task.
• Follow readme.md inside 'dr_segmentaion' instructions

Citation

If you use this code or results in your research, please cite:

@article{tusfiqur2022drg,
  title={DRG-Net: interactive joint learning of multi-lesion segmentation and classification for diabetic retinopathy grading},
  author={Tusfiqur, Hasan Md and Nguyen, Duy MH and Truong, Mai TN and Nguyen, Triet A and Nguyen, Binh T and Barz, Michael and Profitlich, Hans-Juergen and Than, Ngoc TT and Le, Ngan and Xie, Pengtao and others},
  journal={arXiv preprint arXiv:2212.14615},
  year={2022}
}