Annotation-efficient-learning-for-OCT-segmentation

This repository contains the code for the paper "Annotation-efficient learning for OCT segmentation". We propose an annotation-efficient learning method for OCT segmentation that could significantly reduce annotation costs and improve learning efficiency. Here we provide generative pre-trained transformer-based encoder and CNN-based segmentation decoder, both pretrained on open-access OCTdatasets. The proposed pre-trained model can be directly transfered to your ROI segmeantation based on OCT image. We hope this may help improve the intelligence and application penetration of OCT.

Dependencies

python==3.8
torch==1.11.1
numpy==1.19.5
monai==0.7.0
timm==0.3.2
tensorboardX==2.1
torchvision==0.12.0
opencv-python==4.5.5

Usage

1. Clone the repository：

git clone https://github.com/SJTU-Intelligent-Optics-Lab/Annotation-efficient-learning-for-OCT-segmentation.git

2. Install the required dependencies:

pip install -r requirements.txt

3. Download the pre-trained phase1 model file for weights of encoder and phase2 model file for weights of decoder, and then put them in ./runs/ folder.

4. Edit suitable path and parameters in main.py

5. Go to the corresponding folder and run:

cd Annotation-efficient-learning-for-OCT-segmentation
python main.py

Training on your Dataset

The prepared architecture of dataset is referenced to ./dataset/ folder containing train_fewshot_data and val_fewshot_data. The name index of images is listed in train_fewshot_data.txt and val_fewshot_data.txt.

Citation

@article{
  title={Annotation-efficient learning for OCT segmentation},
  author={Zhang, Haoran and Yang, Jianlong and Zheng, Ce and Zhao, Shiqing and Zhang, Aili},
  journal={Biomedical Optics Express},
  volume={14},
  number={7},
  pages={3294--3307},
  year={2023},
  publisher={Optica Publishing Group}
}