AI-Care Binary Classification of Lung Cancer Survival

This repository contains the implementation of a binary classification model for predicting survival outcomes in lung cancer patients from German cancer registry data. The project is part of the AI-Care initiative. This work has been published at the MIE 2025 conference. To cite our paper, please use: @incollection{germer2025lung, title={Lung Cancer Survival Estimation Using Data from Seven German Cancer Registries}, author={Germer, Sebastian and Rudolph, Christiane and Katalinic, Alexander and Rath, Natalie and Rausch, Katharina and Handels, Heinz}, booktitle={Intelligent Health Systems--From Technology to Data and Knowledge}, pages={457--461}, year={2025}, doi={https://doi.org/10.3233/SHTI250379}, publisher={IOS Press} }

Requirements

• After cloning, you need to load the data_import submodule. This can be done using: bash git submodule init git submodule update

• Created with Python 3.12.4

• Used libraries are scikit-learn, catboost, optuna and pandas. You can create such a enviroment with the enviroment.yml or by running via Docker/Podman.

Usage

Via Docker / Podman

After loading the submodule, run podman build -t localhost/aicare-binary-classification:latest . to build the image. Then, run podman run --rm --mount type=bind,source=./data_path,target=/app/data --mount type=bind,source=./result_path,target=/app/results aicare-binary-classification:latest (adjust local paths to your needs!)

Manually

To run the binary classification model, execute the following command:

bash python main.py with the following arguments: --registry, type=str: registry number according to --months, type=int: 'Survival months to binary classify' --inverse, action="store_true": Inverse the binary classification --dummy, action="store_true": Use dummy classifier that always predicts the most frequent class --data_path, type=str: path to your data --entity, type=str: Entity to train on (lung, breast, thyroid, non_hodgkin_lymphoma) --traintestswap, action="store_true": Whether to use register as training data instead of test data

Directory Structure

│ ├── main.py ├── README.md ├── Dockerfile ├── run.sh ├── environment.yml └── data_import ├── data_preprocessing.py └── data_loading.py

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contact

For any questions or issues, please contact sebastian.germer@dfki.de.