The project repository for Weakly supervised perivascular spaces segmentation with salient guidance of Frangi filter (WPSS). WPSS is a weakly supervised convolutional neural network model which focuses on segmenting perivascular spaces (PVS) in the white matter regions.

If you use WPSS code, please cite the following paper:

Weakly upervised Perivascular Spaces Segmentation with salient guidance of Frangi filter
Haoyu Lan, Kirsten M. Lynch, Rachel Custer, Nien‐Chu Shih, Patrick Sherlock, Arthur W. Toga, Farshid Sepehrband, Jeiran Choupan

For data inquiries, please contact me and Dr. Choupan by emails in the publication.

Both training and inference datasets should be organized as the following structure:

dataset
│
└───subject 1
|   |    epc.nii.gz
|   |    mask.nii.gz(optional)
|   |    target.nii.gz(not required for the inference dataset)
└───subject 2
|   |    epc.nii.gz
|   |    mask.nii.gz
|   |    target.nii.gz
└───subject 3
|   |    epc.nii.gz
|   |    mask.nii.gz
|   |    target.nii.gz
│   ...

Human Connectome Project (HCP) dataset was used for the WPSS training and evaluation.

The Enhanced PVS Contrast (EPC) modality and training target generations are upon request. Please refer to the original publication for additional information of EPC modality.

python 3 is required and python 3.6.4 was used in the study.

Requirements can be found at requirements.txt.

Please use pip install requirements.txt to install the requirements

Use the following command for the model training and inference. Pretrained weights with quality controlled labels for the model inference using either the EPC modality or the T1w modality are upon request.

Training script is at ./src/training

Use the following command to run the training script:

python training.py --trainig_size= --gpu= --num_iter= --patch_size= --dataset= --modality= --logdir= --output= --val_portion=

Inference script is at ./src/predict. The corresponding PVS segmentation results will be saved under the same directory as the inference dataset.

Use the following command to run the inference script:

python predict.py --gpu= --weights= --dataset= --modality=

If you use this code for your research, please be familiar with the LICENSE and cite our paper.

@article{lan2023weakly,
  title={Weakly supervised perivascular spaces segmentation with salient guidance of Frangi filter},
  author={Lan, Haoyu and Lynch, Kirsten M and Custer, Rachel and Shih, Nien-Chu and Sherlock, Patrick and Toga, Arthur W and Sepehrband, Farshid and Choupan, Jeiran},
  journal={Magnetic Resonance in Medicine},
  volume={89},
  number={6},
  pages={2419--2431},
  year={2023},
  publisher={Wiley Online Library}
}