xmengli/Rotation-oriented-self-supervised

IEEE TMI21: Rotation-oriented Collaborative Self-supervised Learning for Retinal Disease Diagnosis

Pytorch implementation

Paper

IEEE TMI21: [Rotation-oriented Collaborative Self-supervised Learning for Retinal Disease Diagnosis.]

Installation

Install Python 3.7.4, Pytorch 1.1.0, torchvision 0.3.0 and CUDA 8.0
Or Check requirements.txt
Clone this repo

git clone https://github.com/xmengli999/Rotation-oriented-self-supervised
cd Rotation-oriented-self-supervised

Data Preparation

Download Ichallenge-AMD dataset, file_index
Put them under ./data/
The folder should be

./data/Training400/resized_image_320/XXX.jpg

./data/Training400/random_list.txt

Evaluate

Download our models, password: h7z6, and put it under ./savemodels/
cd scripts
Run scripts in eval_fold.sh to start the evaluation process
5-fold cross-validation results (Table I in the paper):

AUC	Accuracy	Precision
75.64%	87.09%	83.96%

Download our models, password: 2juk, and put it under ./savemodels/
train on DR, test on AMD (Table II in the paper) -- this step requires Pytorch 1.6.0:

AUC	Accuracy	Precision
78.11%	87.85%	85.58%

Train

cd scripts
Check scripts in train_fold.sh to start the training process

Citation

If this code is useful for your research, please consider citing:

@ARTICLE{9411868,
author={Li, Xiaomeng and Hu, Xiaowei and Qi, Xiaojuan and Yu, Lequan and Zhao, Wei and Heng, Pheng-Ann and Xing, Lei},
journal={IEEE Transactions on Medical Imaging}, 
title={Rotation-oriented Collaborative Self-supervised Learning for Retinal Disease Diagnosis}, 
year={2021},
volume={},
number={},
pages={1-1},
doi={10.1109/TMI.2021.3075244}}

Note

Contact: Xiaomeng Li (eexmli@ust.hk)

xmengli / Rotation-oriented-self-supervised