FDU-VTS / DRAC

Team FDVTS_DR's solutions for MICCAI2022 Diabetic Retinopathy Analysis Challenge (DRAC)

DRAC

This repo covers Team FDVTS_DR's solutions for MICCAI2022 Diabetic Retinopathy Analysis Challenge (DRAC).

Dataset

We download the dataset from DRAC2022. For pre-training, we additionally adopt the OCTA-25K-IQA-SEG dataset in challenge 2, and the EyePACS & DDR datasets in challenge 3.

Task 1. segmentation of DR lesions

Please refer to challenge1 package

Task 2. image quality assessment

cd challenge2&3

train the OCTA-25K-IQA-SEG pre-trained vit-s model with mixup and cutmix

python main.py --challenge 2 --model vit --KK 0 [--pretrained True]  [--mixup True]  --visname 2_vit_mix_cut_KK0_pre 

Task 3. DR grading

cd challenge2&3

train the EyePACS & DDR pre-trained vit-s model with mixup and cutmix

python main.py --challenge 3 --model vit --KK 0 [--pretrained True]  [--mixup True]  --visname 3_vit_mix_cut_KK0_pre 

Reference

If you use this code, please cite the following paper [pdf]:

@article{hou2022deep,
  title={Deep-OCTA: Ensemble Deep Learning Approaches for Diabetic Retinopathy Analysis on OCTA Images},
  author={Hou, Junlin and Xiao, Fan and Xu, Jilan and Zhang, Yuejie and Zou, Haidong and Feng, Rui},
  journal={arXiv preprint arXiv:2210.00515},
  year={2022}
}