UltraFAV for Artery Vein Segmentation

The UltraFAV framework provides a robust and efficient solution for the segmentation of retinal arteries and veins.

🚀Installation

1.​ Clone the Repository

Download the repository and navigate to the project directory:

git clone https://github.com/iMED-Lab/UltraFAV.git
cd UltraFAV

2. ​Create a Conda Environment

Create and activate a dedicated Conda environment for the framework:

conda create -n ultrafav python=3.11 -y
conda activate ultrafav

3. ​Install nnUNetv2

Install the nnUNetv2 library, which is a prerequisite for the UltraFAV framework:

pip install nnunetv2

4. Install the UltraFAV Framework

Install UltraFAV in editable mode:

pip install -e .

🚥 Usage

1. Run Training

To start training the UltraFAV framework, use the following command:

ultrafav_train <dataset_id> <configuration> <fold>

Example:

ultrafav_train 88 2d 0

2. Run Prediction

To generate predictions using a trained model, use:

ultrafav_predict -i <input_folder> -o <output_folder> -d <dataset_id> -c <configuration> -f <fold>

Example:

ultrafav_predict -i in_dir -o out_dir -d 88 -c 2d -f 0

Note: The checkpoint with the best performance is automatically selected as the default for prediction.

🔖 Citation

Paper is under review, coming soon

🔗 Links

• PRIME
• AV-WIDE
• LES-AV
• HRF

❤️ Acknowledgements

We would like to extend our gratitude to the following contributors and organizations whose support has been instrumental in the development of the UltraFAV framework:

• The developers and maintainers of nnUNetv2, whose robust segmentation framework inspired this project.
• The research community in retinal imaging and medical image analysis for providing valuable insights and benchmark datasets.