🫁 LyNoS 🤗

A multilabel lymph node segmentation dataset from contrast CT

LyNoS was developed by SINTEF Medical Image Analysis to accelerate medical AI research.

Brief intro

This repository contains the LyNoS dataset described in "Mediastinal lymph nodes segmentation using 3D convolutional neural network ensembles and anatomical priors guiding". The dataset has now also been uploaded to Zenodo and the Hugging Face Hub enabling users to more easily access the data through Python API.

We have also developed a web demo to enable others to easily test the pretrained model presented in the paper. The application was developed using Gradio for the frontend and the segmentation is performed using the Raidionics backend.

Dataset

Accessing dataset

The dataset contains 15 CTs with corresponding lymph nodes, azygos, esophagus, and subclavian carotid arteries. The folder structure is described below.

The easiest way to access the data is through Python with Hugging Face's datasets package:

from datasets import load_dataset

# downloads data from Zenodo through the Hugging Face hub
# - might take several minutes (~5 minutes in CoLab)
dataset = load_dataset("andreped/LyNoS")
print(dataset)

# list paths of all available patients and corresponding features (ct/lymphnodes/azygos/brachiocephalicveins/esophagus/subclaviancarotidarteries)
for d in dataset["test"]:
  print(d)

A detailed interactive demo on how to load and work with the data can be seen on CoLab. Click the CoLab badge to see the notebook or alternatively click here to see it on GitHub.

Dataset structure

└── LyNoS.zip
    ├── stations_sto.csv
    └──  LyNoS/
        ├── Pat1/
        │   ├── pat1_data.nii.gz
        │   ├── pat1_labels_Azygos.nii.gz
        │   ├── pat1_labels_Esophagus.nii.gz
        │   ├── pat1_labels_LymphNodes.nii.gz
        │   └── pat1_labels_SubCarArt.nii.gz
        ├── [...]
        └── Pat15/
            ├── pat15_data.nii.gz
            ├── pat15_labels_Azygos.nii.gz
            ├── pat15_labels_Esophagus.nii.gz
            ├── pat15_labels_LymphNodes.nii.gz
            └── pat15_labels_SubCarArt.nii.gz

NIH Dataset Completion

A larger dataset made of 90 patients featuring enlarged lymph nodes has also been made available by the National Institutes of Health, and is available for download on the official web-page. As a supplement to this dataset, lymph nodes segmentation masks have been refined for all patients and stations have been manually assigned to each, available here.

Demo

To access the live demo, click on the Hugging Face badge above. Below is a snapshot of the current state of the demo app.

Development

Docker

Alternatively, you can deploy the software locally. Note that this is only relevant for development purposes. Simply dockerize the app and run it:

docker build -t LyNoS .
docker run -it -p 7860:7860 LyNoS

Then open http://127.0.0.1:7860 in your favourite internet browser to view the demo.

Python

It is also possible to run the app locally without Docker. Just setup a virtual environment and run the app. Note that the current working directory would need to be adjusted based on where LyNoS is located on disk.

git clone https://github.com/raidionics/LyNoS.git
cd LyNoS/

virtualenv -python3 venv --clear
source venv/bin/activate
pip install -r ./demo/requirements.txt

python demo/app.py --cwd ./

Citation

If you found the dataset and/or web application relevant in your research, please cite the following reference:

@article{bouget2021mediastinal,
  author = {David Bouget and André Pedersen and Johanna Vanel and Haakon O. Leira and Thomas Langø},
  title = {Mediastinal lymph nodes segmentation using 3D convolutional neural network ensembles and anatomical priors guiding},
  journal = {Computer Methods in Biomechanics and Biomedical Engineering: Imaging \& Visualization},
  volume = {0},
  number = {0},
  pages = {1-15},
  year  = {2022},
  publisher = {Taylor & Francis},
  doi = {10.1080/21681163.2022.2043778},
  URL = {https://doi.org/10.1080/21681163.2022.2043778},
  eprint = {https://doi.org/10.1080/21681163.2022.2043778}
}

License

The code in this repository is released under MIT license.