Outcome prediction for elderly head&neck cancer patients receiving chemoradiotherapy

Framework to train and validate an artficial neural network (ANN) for the prediction of progression-free-survival (PFS) or overall survival (OS).

Elia Lombardo and Sebastian Marschner

Department of Radiation Oncology, LMU University Hospital, LMU Munich, Germany

Elia.Lombardo@med.uni-muenchen.de Sebastian.Marschner@med.uni-muenchen.de

Installation

• Download and unzip the repository to a local folder of your preference.
• Build a Docker image based on the provided Dockerfile and requirements.txt and run a container while mounting the chemo_outcome_prediction folder.
• Move your dataset excel file into the data subfolder.

Usage

• Open chemo_outcome_prediction/code/config.py and change path_project to the path inside the container where chemo_outcome_prediction was mounted.
• Modify also other variables such as the excel filename of your data etc. as needed in chemo_outcome_prediction/code/config.py.
• Start a hyper-parameter grid search directly from the terminal by running for instance bash main_grid_search.sh ANN OS or bash main_grid_search.sh ANN PFS for training the ANN for PFS and OS, respectively.
• Perfom inference by setting the endpoint and model in the main_infer_ANN.py script.
  • Trained model weights for OS and PFS can be found under chemo_outcome_prediction/results/training/ANN
  • SHAP explainability analysis included

Publication

If you use this code in a scientific publication, please cite our paper: https:xxx