Head and neck progression free survival prediction

Code for our survival prediction approach to the MICCAI HEad and neCK TumOR segmentation and outcome prediction in PET/CT images (HECKTOR) 2021 challenge.

Erratum

Note: Index of $T^{max}$ in eq. (2) should be a $k$ not a $l$, i.e. $T_l^{max} \rightarrow T_k^{max}$

Conda environments

There are two files provided to generate conda environments which allow you to run the code. The environment generated by prep_env.yml is needed for preprocessing and train_env.yml for training.
Run

conda env create --file=<FILE.yaml>

to generate the (prep) and (train) environments.

Data preparation

Put the (zipped) challenge data in the data/ folder and run the utils/data_prep.sh file with the (prep) conda environment. If you are not on Linux, follow the respective steps specified in the file.

Training

In order to train the model you have to run:

python main.py [--gpu <GPU_INDEX>] parameter/par.yml

in the train/ folder with the (train) conda environment.

Pretrained weights

The pretrained weights can be downloaded from here and have to be placed in the data/weights/ folder.

Cite as

@inproceedings{lang2022deep,
        title={{Deep Learning Based GTV Delineation and Progression Free Survival Risk Score Prediction
                for Head and Neck Cancer Patients}},  
        author={Lang, Daniel M. and Peeken, Jan C. and Combs, Stephanie E. and
                Wilkens, Jan J. and Bartzsch, Stefan},  
        booktitle={{Head and Neck Tumor Segmentation and Outcome Prediction}},
        year={2022},
        publisher={{Springer International Publishing}},
        pages={150--159},
}