PET/CT Head and Neck tumor auto-segmentation

This is a repository for competition of MICCAI 2021: HECKTOR - head and neck gross tumor volume(GTV) segmentation.

Task 1: Segmentation of GTV:

PET Normalizations to Improve Deep Learning Auto-Segmentation of Head and Neck Tumors in 3D PET/CT

Task 2: Treatment outcomes prediction:

Comparing deep learning and conventional machine learning for outcome prediction of head and neck cancer in PET/CT

Prerequisites:

HECKTOR public code for resample images to istropical 1mm grid with bounding box(144x144x144):

git clone https://github.com/voreille/hecktor
cd hecktor/src/resampling/
python resample.py

We use nnUNet as the baseline model for the development of GTV multimodality segmentation network. install nnUNet:

git clone https://github.com/MIC-DKFZ/nnUNet.git
cd nnUNet
pip install -e .

convert data: We've provded a script to convert sampled image to nnUNet format. Please change the your file location for both downloaded_data_dir and downloaded_data_dir_test

if convert train set only:

python data_conversion.py

if conver both train and test set:

python data_conversion.py --test=True

Evaluation

Evaluation on 5-folds cross validation(CV)

Run evaluation on 5-folds CV using following command for task id XXX:

nnUNet_find_best_configuration -m 3d_fullres -t XXX --strict

For more details, please reference to:

Ren, J., Huynh, B. N., Groendahl, A. R., Tomic, O., Futsaether, C. M., & Korreman, S. S. (2021, September). PET Normalizations to Improve Deep Learning Auto-Segmentation of Head and Neck Tumors in 3D PET/CT. In 3D Head and Neck Tumor Segmentation in PET/CT Challenge (pp. 83-91). Springer, Cham.

@inproceedings{ren2021pet,
  title={PET Normalizations to Improve Deep Learning Auto-Segmentation of Head and Neck Tumors in 3D PET/CT},
  author={Ren, Jintao and Huynh, Bao-Ngoc and Groendahl, Aurora Rosvoll and Tomic, Oliver and Futsaether, Cecilia Marie and Korreman, Stine Sofia},
  booktitle={3D Head and Neck Tumor Segmentation in PET/CT Challenge},
  pages={83--91},
  year={2021},
  organization={Springer}
}

and

Huynh, B. N., Ren, J., Groendahl, A. R., Tomic, O., Korreman, S. S., & Futsaether, C. M. (2021, September). Comparing deep learning and conventional machine learning for outcome prediction of head and neck cancer in PET/CT. In 3D Head and Neck Tumor Segmentation in PET/CT Challenge (pp. 318-326). Springer, Cham.

@inproceedings{huynh2021comparing,
  title={Comparing deep learning and conventional machine learning for outcome prediction of head and neck cancer in PET/CT},
  author={Huynh, Bao-Ngoc and Ren, Jintao and Groendahl, Aurora Rosvoll and Tomic, Oliver and Korreman, Stine Sofia and Futsaether, Cecilia Marie},
  booktitle={3D Head and Neck Tumor Segmentation in PET/CT Challenge},
  pages={318--326},
  year={2021},
  organization={Springer}
}

BBQtime / HNSCC-ct-pet-tumor-segmentation