Cancer-segmentation

Getting started

To clone via ssh configure your local system for ssh-access as described in the github-docs.

To clone this repository, type:

git clone git@github.com:Krebs-Busters/zone-segmentation.git

If outgoing SSH is disabled in your network, download the repository using your web browser, using the download ZIP button. After extracting the zip file set up the Python environment by running:

pip install -r requirements.txt

Automatic setup of the Prostate-X data for training and testing.

Navigate into the data folder and run python download.py to set up the Prostate-X data-set.

Training & Validation

For training, run

PYTHONPATH=. python scripts/train_prostate_X.py

. In case of training for the first time, change the variable load_new to True. Once the training is done, weights are saved as a pickled file in ./weights.

Model

src/networks.py implements a 3D U-Net flax-model as specified by Meyer et al. as well as our own implementation of a softmax_focal_loss.

To test the model, run

PYTHONPATH=. python scripts/sample_prostate_X.py

Citation

Should you use this work in an academic context please cite:

@inproceedings{meyer2019towards,
  title={Towards patient-individual PI-Rads v2 sector map: CNN for automatic segmentation of prostatic zones from T2-weighted MRI},
  author={Meyer, Anneke and Rakr, Marko and Schindele, Daniel and Blaschke, Simon and Schostak, Martin and Fedorov, Andriy and Hansen, Christian},
  booktitle={2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)},
  pages={696--700},
  year={2019},
  organization={IEEE}
}

@software{wolter2024stability,
  title={On the Stability of Neural Segmentation in Radiology},
  author={Wolter, Moritz and Wichtmann, Barbara},
  url = {https://github.com/Krebs-Busters/zone-segmentation}
  year={2024},
}