Dense Regression Activation Maps For Lesion Segmentation in CT scans of COVID-19 patients
the implementation of our dense regression activation map algorithm. The algorithm takes a CT image and its corresponding lobe segmentation as the input, generating a lesion segmentation as the output.
Installation
- Please check
/docker_base/DockerFilefor the required pacakges to build the docker image. Note that there is another/DockerFileis for build a docker image for grand-challenge algorithm. - Regarding DGL library, we suggest you install 0.6.x. 0.4.x cannot be used because of bugs related to the implementation of graph attention networks.
Usage
-
before training, run
prepare_data.pyto generate lobw-wise chunk images for training. -
For training, run
train.py -
For testing, run
process_pipeline.py