xmengli / self_supervised

TMI20: Self-supervised Feature Learning via Exploiting Multi-modal Data for Retinal Disease Diagnosis

Pytorch implementation

Paper

Self-supervised Feature Learning via Exploiting Multi-modal Data for Retinal Disease Diagnosis.
IEEE Transactions on Medical Imaging, 2020

Installation

• Install Python 3.7.4, Pytorch 1.1.0, torchvision 0.3.0 and CUDA 9.0
• Or Check requirements.txt
• Clone this repo

git clone https://github.com/xmengli999/self_supervised
cd self_supervised

Data Preparation

• Download Ichallenge-AMD dataset, synthesized FFA and file_index
  
• Put them under ./data/

Evaluate

• Download our models in Baidu password: gja3, or our models in Onedrive, and put it under ./savedmodels/
• cd scripts
• Run sh evaluate_fold.sh to start the evaluation process
• 5-fold cross-validation results:

AUC	Accuracy	Precision
74.58%	86.58%	83.2%

Train

• cd scripts
• Run sh train_fold.sh to start the training process
• See train_ablation.sh for ablation study
• See supervised_fundus.py for supervised baselines

Note

• Contact: Xiaomeng Li ([e-mail])

Citation

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

@article{li2020self,
title={Self-supervised Feature Learning via Exploiting Multi-modal Data for Retinal Disease Diagnosis},
author={Li, Xiaomeng and Jia, Mengyu and Islam, Md Tauhidul and Yu, Lequan and Xing, Lei},
journal={IEEE Transactions on Medical Imaging},
year={2020},
publisher={IEEE}
}