AV-casNet: Fully Automatic Arteriole-Venule Segmentation and Differentiation in OCT Angiography

You can download the datasets from the Baidu Netdisk (access code: n4i4) and decompress it to the root directory. Please go to "./datasets/README.md" for details.

• Please prepare an environment with python=3.8, and then use the command "pip install -r requirements.txt" for the dependencies.
• You should also install MATLAB 2020b or later MATLAB version for running post-processing and evaluation code.

• the whole workflow includes five main steps: training and testing a CNN model, building the graph data based on CNN predictions, and training and testing a GAT model. If you want to run all steps automatically, you can type the following:

python terminal.py --train_cnn=1 --test_cnn=1 --build_graph=1 --train_gat=1 --test_gat=1

• where =1 means execute the process and =0 means not to execute.
• The binaried segmentation results will be saved in the folder GAT_final. If you want to see the probability map and superpixel results, you can go to the result folders in "./Core/UNet/" and "./Core/GAT/" for details.
• If you want to run some steps independently, you can simply set the parameter value of the corresponding step to 1 and the parameter value of the other steps to 0. Take training and testing a CNN model alone as an example:

python terminal.py --train_cnn=1 --test_cnn=1 --build_graph=0 --train_gat=0 --test_gat=0

• This part of the code is implemented by MATLAB 2020b. When you have completed all the above processes, you can perform "./post_processing.mlx" in MATLAB to get the final visual segmentation results and quantitative evaluation results.

@misc{cao2021swinunet,
      title={AV-casNet: Fully Automatic Arteriole-Venule Segmentation and Differentiation in OCT Angiography}, 
      author={Xiayu Xu, Peiwei Yang, Hualin Wang, Zhanfeng Xiao, Gang Xing, Xiulan Zhang, Wei Wang, Feng Xu, Jiong Zhang, Jianqin Lei},
      year={2022},
}