README.md

Lung segmentation API

API that returns lung segmentations for DICOM chest CT images and calculates lung volumes.

We use the Nvidia Clara lungs 3D semantic segmentation model, available here.

The API takes a CT scan, extracts 32 samples from it, and performs 3D semantic segmentation. It then interpolates the predictions between the samples to produce a segmentation mask for each frame and calculates the lung volume based on the DICOM metadata. The final result is returned as a .json file. To learn more about the structure see the documentation

How to run the API

docker-compose up

API avalible at:

0.0.0.0:8011

FAQ

• I'm getting an error on MacBook docker rpc error code = unknown desc = executor failed running [...].

Your docker settings are limiting the size of the image and cannot install all the requirements.txt. Go to Preferences > Resources > Advanced in your Docker Desktop application and increase the memory limit.

How to test it?

Follow the official tutorial. You need to configure a proxy server that will route requests from the viewer to the model API.

You will need to add the endpoint to models.json file in model-proxy

{
  "1d9508dd-9089-40c3-abdd-15f47120d682": {
    "uri": "http://localhost:8011/segmentation",
    "supports": ["/studies/series"],
    "task": "segmentation"
  }
}