Deep learning enables accurate diagnosis of novel coronavirus (COVID-19) with CT images

The source code of Deep learning enables accurate diagnosis of novel coronavirus (COVID-19) with CT images.

1. Our platform

Please visit COVID19-CT images diagnosis. I would like to suggest you to upload a group of CT images (for example, upload 10 images) at once, and you will get a more accurate result.

Demo

• Left: COVID-19 images.
• Middle: Non-COVID-19 images. Actually, these are CT images of bacterial pneumonia.
• Right: Invalid image. At present, our server only serves specific hospitals. If you want to test CT images from other hospitals, please refer to 3. Using your own dataset.

2. Using pretrained model

For local prediction, please download model.ckpt and put it into dir /online_backend/, run:

python run.py DIR

where DIR includes a set CT JPEG images of one person.

3. Using your own dataset

First, you should convert your CT images into JPG format.

We recommand the following file structure:

DATASET_ROOT_DIR
└── INPUT_DIR
    ├── Person 1
    |   ├── image_1
    |   ├── image_2
    |   ├── ...
    |   └── image_n
    ├── Person 2
    ├── ...
    └── Person m

Then, run /local_trainer/pre_cut_lung.py INPUT_DIR OUTPUT_DIR to preprocess the images.