COVID-19 Outcome Prediction Guideline

Environment Building

Python Environment

Install a python 3.6 environment using conda

Packages

1. Install gdcm using command

conda install -c conda-forge gdcm

2. Install majority of packages using command pip install -r requirements.txt

3. Install pytorch using command conda install

pytorch-cpu==1.1.0 torchvision-cpu==0.3.0 cpuonly -c pytorch

Or follow this link to install pytorch 1.1.0.

4. Install pyradiomics from source using command

git clone git://github.com/Radiomics/pyradiomics

For unix like systems (MacOSX, linux):

cd pyradiomics

python -m pip install -r requirements.txt

python setup.py install

python setup.py build_ext --inplace

For Windows:

cd pyradiomics

python -m pip install -r requirements.txt

python setup.py install

5. (HELPER) If you run into problem described by this link when extracting features, you can replace the files in

anaconda\envs\<env_name>\lib\site-packages\radioimcs\

with the corresponding files we provided in folder radiomics_patch.

Feature extraction pipeline usage

Input

1. Place unsegmented dicom image series in a folder (e.g. <dcm>), and arrange the dicom series in the following way:

<dcm>
├───<patient_id>
│   └───<study_uid>
│       └───<series_uid>
│           ├───000001.dcm
│    		├───000002.dcm
│    		├───...
│    		└───000333.dcm
├───<patient_id>
│...

2. Convert segmentation mask images into PNG format, named them in a sequence of numbers (e.g. 000.png, 001.png, …, 332.png), and place them in another folder (e.g. <seg>) for segmentation, and arrange them in the following way:

<seg>
├───<patient_id>
│   └───<study_uid>
│       └───<series_uid>
│           ├───000.png
│    		├───001.png
│    		├───...
│    		└───332.png
├───<patient_id>
│...

Execution

1. Open terminal in root directory of the prediction pipeline.

2. Activate the conda environment in the terminal.

3. Extract radiomics features using the following command:

python proc_radiomic_feature.py --dicom_root <dcm> --lesion_mask_root <seg> --save_root <save_root>

Output

1. Extracted features will be saved in the <save_root> directory with file name: “final_merge_feature.csv”.

Prediction Pipeline Usage

Input

1. For radiomics-only models (Radiom), use command

python COVID-19_prediction.py --radiomics_data <save_root\final_merge_feature.csv>

1. For models that include clinical symptoms, demographics and lab test results (RadioClinLab), you can prepare an additional CSV input, according to “example_lab_input.csv” (refer to the “units.txt” for units and meanings of the entries). This file should also be placed into the root directory of the prediction pipeline (<project_root>\<lab_input.csv>).
2. To use the “RadioClinLab” model, use the command line code

python COVID-19_prediction --radiomics_data <save_root\final_merge_feature.csv> --lab_data <lab_input.csv>