DR Grading Survival Handbook

DR Grading datasets preparation:

Messidor-2: data, label

APTOS: data

IDRID: data

DeepDRiD: data

FGADR: data

Notes: you need to write emails to apply for the download link of this dataset

Papers:

This paper, 'CANet: Cross-Disease Attention Network for Joint Diabetic Retinopathy and Diabetic Macular Edema Grading' could be followed.

This paper is about DR and DME jointly grading. You can only focus on DR grading and follow the format of the baseline.py in CANet.

Or you could follow the paper, 'Learning Robust Representation for Joint Grading of Ophthalmic Diseases via Adaptive Curriculum and Feature Disentanglement'.

Other papers available: 1. CABNet: Category Attention Block for Imbalanced Diabetic Retinopathy Grading

Preparation before training:

To run a deep learning algorithm by GPU, there are something need to be prepared.

1. make sure you have a gpu

2. download the correct version of cuda

link: https://developer.nvidia.com/cuda-downloads usually we download the newest version of cuda

3. download the corresponding version of pytorch

link: https://pytorch.org/get-started/previous-versions/ if your gpu is old, something error may occur. you need to find the corresponding version of pytorch

4. check if the cuda could be used

open your terminal, type 'python' to get into python, then type like this: the output should be true

some hints about DR grading

1. you can learn how Resnet is realized

link: https://zh.d2l.ai/chapter_convolutional-modern/resnet.html the d2l library contains some basic codes about the training process, how to compute the accuracy, how to visualize the output... you can check the codes for reference

2. install the tensorboard and learn how to use tensorboard to visualize the output

link: https://github.com/tensorflow/tensorboard

3. the framework

1. a dataloader to load your DR dataset
2. use parsers to imput as commands
3. load your model (usually Resnet) and use pretrained model
4. the training process
5. compute the accuracy and auc score
6. visualize the output

An example output for reference