Deep Learning Reproducibility Project: A Novel Hybrid Convolutional Neural Network for Accurate Organ Segmentation in 3D Head and Neck CT Images

This repository is an attempt at reproducing the results from, A Novel Hybrid Convolutional Neural Network for Accurate Organ Segmentation in 3D Head and Neck CT Images[1]. This repository is implementation in the pytorch framework. For an elaborate overview of our reproduction and results please see our blog post at, https://thomas100z.github.io/DLgroup79/

Installation

Install required packages:

python3 -m pip install -r requirements.txt

Getting started

Data

Extract the data to the data folder. The data folder should be structured like this:

.
├── ...
├── data                    
│   ├── test_offsite                # test set
│   │   └── data_3D
│   │       └── ...                 # samples
│   ├── train                       # train set 
│   │    └── data_3D
│   │        └── ...                # samples
│   └── train_additional            # validation set  
│        └── data_3D
│            └── ...                # samples 
└── ...

For linux this snippet can be run:

wget https://github.com/prerakmody/hansegmentation-uncertainty-qa/releases/download/v1.0/train_additional.zip
wget https://github.com/prerakmody/hansegmentation-uncertainty-qa/releases/download/v1.0/train.zip
wget https://github.com/prerakmody/hansegmentation-uncertainty-qa/releases/download/v1.0/test_offsite.zip

unzip train.zip -d data/
unzip train_additional.zip -d data/
unzip test_offsite.zip -d data/

rm train.zip
rm train_additional.zip
rm test_offsite.zip

Training model

To train the model simply run the following command:

python3 main.py

Testing model

To test the model on the test set and view the box plot and inferencing results run:

python3 predict.py  # (optionally provide model to test as firts argument, default last model is used): ./models/xxx.pth

Results

Background,training & testing results and discussion can be found in our blog post.

Acknowledgements

We would like to thank Prerak Mody for his help and insight during the project as well as for his pre-procssed data from the MICCAI2015 challenge [2].

Contributors

Thomas Zuiker
Storm Holman
Lucas Veeger
Emma Botten

References

Original paper
A Novel Hybrid Convolutional Neural Network for Accurate Organ Segmentation in 3D Head and Neck CT Images
By: Zijie Chen, Cheng Li, Junjun He, Jin Ye, Diping Song, Shanshan Wang, Lixu Gu, and Yu Qiao
DOI: https://doi.org/10.1007/978-3-030-87193-2_54

Dataset
2015 MICCAI Challenge: Evaluation of segmentation methods on head and neck CT: Auto‐segmentation challenge 2015.
By: Raudaschl, P. F., Zaffino, P., Sharp, G. C., Spadea, M. F., Chen, A., Dawant, B. M., ... & Jung, F.
Obtained from: https://github.com/prerakmody/hansegmentation-uncertainty-qa/releases