Welcome to the Head and Neck Tumour Segmentation and Prediction of Patient Survival Project!

This project aims to provide methods to automatically segment the primary gross target volume on fluoro-deoxyglucose (FDG)-PET and Computed Tomography (CT) images and prediction of progression-free survival in H&N oropharyngeal cancer. We participated to the HEad and neCK TumOR Segmentation and Prediction of Patient Outcome Challenge 2021 (HECKTOR 2021) which creates a platform for comparing segmentation methods and predictions of patient survival.

For the segmentation task, we proposed a new network based on an encoder/decoder architecture with attention mechanisms and full inter- and intra-skip connections inspired from UNet3+ to take advantage of low-level and high-level semantics at full scales. Additionally, we used Conditional Random Fields (CRF) as a post-processing step to refine the predicted segmentation maps.

For prediction of patient progression free survival, we extracted relevant clinical, radiomic, and deep learning features using Lasso regression. Our best performing model was a Cox proportional hazard regression.

1. Clone the repo

   git clone https://github.com/EmmanuelleB985/Head-and-Neck-Tumour-Segmentation-and-Prediction-of-Patient-Survival.git

2. Install the packages

   pip install requirements.txt

• For segmentation task

   cd src/Segmentation_Task

To train the model

   python main.py

• For Survival task

   cd src/Survival_Task

To train and evaluate the models, run the notebook Survival.ipynb

Distributed under the MIT License. See LICENSE for more information.