HoloRepository 2023 - Monai neural network container

This repository is for the HoloRepository 2023 project, an update from the 2019 version. The project is deployed as a webapp on Azure. The webapp is a Flask app that uses a Docker container to run a Monai Zoo model. The model is a neural network that segments abdominal CT scans. The webapp will process the inputs from the pipeline webapp and send the result back to the pipeline webapp.

Main changes from the 2019 version:

• The model is now a Monai Zoo model instead of a NiftyNet model, the docker file and model.py is updated to reflect this change, with new files for the model added
• The webapp is now deployed on Azure individually instead of being deployed as a part of the pipeline

Build Docker image from the Docker file

# example
docker build -t monai .

This instruction will create the monai docker image with the flask server endpoint. to run the image you create you need to run the other command

Run docker and connect docker to the port 5000

# example
docker run --rm -d -p 5000:5000 --name="monai" monai

This instruction will link your local port 5000 with the port 5000 in the container and the container has been set to run the server when you run the docker image

API

The container runs on the http://localhost:5000 To send a request for the segmentation, We provide a /model endpoint to do the segmentation.

Here are the requirements for the segmentation request:

• request must be sent to this address http://localhost:5000/model
• request must be a post-request
• request should contain the nifiti file for segmentation
• The format for the nifiti file should end with .nii format

Testing for the flask script

To run the test, you need to make sure that your docker container is running and you can run this command, however, this test is left by the original developer using resources that are no longer available, so some tests will result in failure unless updated

docker exec monai python3 test.py