Effective Deep Learning Approaches for Predicting COVID-19 Outcomes from Chest Computed Tomography Volumes

This project provides code to train semantic segmentation models for pulmonary lesions segmentation from Computer Tomography (CT) scans.

Getting Started

Data preprocessing and Dataset creation

1. Mask Generation: Generate the segmentation mask for the different classes.
2. Random Split: The final dataset is saved in three folders: train/ test/ val/
3. Generate stats: Generate statistics of the input image channels: returns keys for means and stds accross the channels in input slices. Note: Not implemented yet.

Software Dependencies

🐍 Python 3.x

• torch==1.4.0
• torchvision==0.5.0
• tifffile==2020.2.16
• opencv-python==4.2.0.32
• matplotlib==3.1.3
• pandas==1.0.1
• pyarrow==0.16.0
• imutils==0.5.3
• scikit-image==0.16.2
• scikit-learn==0.22.2
• tqdm==4.46.1
• xmltodict==0.12.0

Install Dependencies with Conda

conda create --name ct_segmentation python=3.6
conda activate ct_segmentation
conda install pytorch torchvision cudatoolkit=10.2 -c pytorch
pip install -r requirements.txt

Note: Make sure to intall a version of cudatoolkit supported by your NVCC version

Build and Test

To run all preprocessing and create dataset:

python3 -m data.preprocess_dataset.py --raw_data_dir=/path_to_the_directory_where_raw_data_is_located_at --output_dir=/path_to_the_directory_where_covid_ct_datadset_will_be_located_at/

This will generate a dataset in the format:

output_dir/covid_ct/
└── train/
    └── images/
    └── masks
└── val/
    └── images/
    └── masks/
└── test/
    └── images/
    └── masks/

Training multihead segmentation Network

To train the network with default values and see training output on the command line run:

python3 train.py --data_dir=/dir_to_datasets_directory/ --verbose

Note: Make sure the path to your code is added to th PYTHONPATH, so modules can be recognized: export PYTHONPATH=$PYTHONPATH:/path_to_project_code_folder/

Contribute

You can contribute to the project by submitting bugs and feature requests.

License

Copyright (c) Microsoft Corporation. All rights reserved.

Licensed under the MIT license.