CasiaFan / COVID19_multi_class_classification

A project to train a 3-class COVID-19 classification model on CXR and CT images

Here is the tutorial to train a 3-class COVID-19 classification model.

0. Requirements

• Python 3.7
• PyTorch
• TorchVision
• Fire
• OpenCV 4.2.0
• Numpy
• Pandas
• Scikit-Learn
• Pillow

1. Data

• Chest X-ray images: make the COVIDx dataset following this tutorial: https://github.com/lindawangg/COVID-Net/blob/master/create_COVIDx.ipynb train_COVIDx5.txt and test_COVIDx4.txt: train and test files of COVIDx dataset train_sample.txt: toy sample for testing the code.

• CT images: make the COVIDx-CT dataset as: https://github.com/haydengunraj/COVIDNet-CT/blob/master/create_COVIDx_CT.ipynb Note: if we need to reduce the training time, consider reducing the number of the training images.

2. Code

Currently, only CXR datasets are supported (3-classes). data.py: prepare data for training. model.py: network architectures that could be employed for training train.py: code to train the model (Read it and understand the basic steps to train a network using PyTorch) eval.py: evaluate trained model performance: F1 score, precision, recall, sensitivity, AUC

Train: bash run_train.sh, or python train.py --help for help to check how to set the parameters Test: bash run_eval.sh, or python eval.py --help for help.

3. Tasks

1. add dataset function for CT images as shown in data.py
2. adjust the training process in train.py to support multi-modal training
3. support evaluation of multi-modal models in eval.py
4. try to add model network architectures in model.py

4. Reference Repos:

1. https://github.com/lindawangg/COVID-Net
2. https://github.com/haydengunraj/COVIDNet-CT