Introduction

This algorithm is for the stenosis detection task in ARCADE Challenge, which was held at MICCAI 2023. We are ranked ${\textsf{\color{red}3rd}}$ !

Our publication: StenUNet: Automatic Stenosis Detection from X-ray Coronary Angiography Arxiv

Please refer to MICCAI-ARCADE for the segmentation detection task.

Installation - original

python>=3.9 and torch>=2.0.0

  conda create -n stenunet_env python=3.9
  conda activate stenunet_env
  git clone https://github.com/HuiLin0220/StenUNet.git
  cd StenUNet
  pip install  -r ./requirements.txt

Installation - rsz

  python -m venv env
  source env/bin/activate
  git clone https://github.com/szuhow/stenunet
  cd StenUNet
  pip install  -r ./requirements.txt

Prepare data

• Rename and put the training images in this folder "./nnNet_training/Raw_data/"

The training folder structure is like this:

     Raw_data/Dataset_train_val/  
      ├── imagesTr
      │   ├── sten_0000_0000.png
      │   ├── sten_0000_0001.png
      │   ├── ...
      │   ├── sten_0001_0000.png      
      │   ├── sten_0001_0001.png      
      │   ├── ... 
      │   ├── sten_0002_0000.png
      │   ├── sten_0002_0001.png
      │   ├── ...
      ├── labelsTr
      │   ├── sten_0000.png
      │   ├── sten_0001.png
      │   ├── sten_0002.png
      │   ├── ...

Train - optional

  python training_planning.py #Planning hyper_parameters

  CUDA_VISIBLE_DEVICES=0 python training.py 0
  #CUDA_VISIBLE_DEVICES=X python train.py fold_ID(0,1,2,3,4)

Inference

1. If not properly named, do rename and put the test images in this folder'./dataset_test/raw';

2. Run

     python inference.py
   

3. Shareing StenUnet's weight (Google drive) - put file in ./model_folder directory.

4. You will get the preprocessed images, raw prediction after StenUNet, and post_prediction after postprocessing.

You can integrate your own preprocessing/postprocessing strategies in preprocess.py/post_process

The inference folder structure is like this:

  daset_test/
      ├── raw
      │   ├── sten_0000_0000.png
      │   ├── sten_0001_0000.png
      │   ├── ...
      ├── preprocessed
      │   ├── sten_0000_0000.png       # prerpocessing method0
      │   ├── sten_0000_0001.png       # prerpocessing method1
      │   ├── sten_0000_0003.png       # prerpocessing method2
      │   ├── ... 
      │   ├── sten_0001_0000.png
      │   ├── sten_0001_0001.png
      │   ├── sten_0001_0003.png
      │   ├── ...
      ├── raw_prediction
      │   ├── sten_0000.png
      │   ├── sten_0001.png
      │   ├── ...
      ├── post_prediction
      │   ├── sten_0000.png
      │   ├── sten_0001.png
      │   ├── ...

References

nnunet

Citation

Please cite the following paper when using SteUNet:

  @article{lin2023stenunet,
    title={StenUNet: Automatic Stenosis Detection from X-ray Coronary Angiography},
    author={Lin, Hui and Liu, Tom and Katsaggelos, Aggelos and Kline, Adrienne},
    journal={arXiv preprint arXiv:2310.14961},
    year={2023}
  }

Contact Us

Feel free to contact me at huilin2023@u.northwestern.edu

To-do List