UNet-based segmentation algorithm

This repository contains a UNet-based segmentation algorithm and wrapper code to integrate it into the THRIVE framework. This repository is a work in progress, and the segmentation model is to be published in future. Please feel free to contact Yang Zhao ([email protected]) if any questions.

Unpack data as instructed in THRIVE

Set up environment variables

Within your directory that contains the file docker-compose.yml, create a file called .env that contains:

LOCAL_DATA_DIR=/your/data/root Rt106_SERVER_HOST=localhost

Build the UNet cell segmentation algorithm docker image

First, build the rt106 base image. Go to "containers" directory, build the "thrive20/rt106-algorithm-sdk-focal" image by doing:

$ docker build -t thrive20/rt106-algorithm-sdk-focal .

Second, build the algorithm image. Go to the algorithm directory, e.g., containers/algorithms, build the algorithm image by:

$ docker build -t thrive20/unet-cell-segmentation-focal . replace "unet-cell-segmentation-focal" by your own algorithm image name.

Note that you may need to change the "entrypoint.sh" script depending on the python versions you have: /usr/bin/python3 ./rt106GenericAdaptorREST.py & sleep 3 /usr/bin/python2 ./rt106GenericAdaptorAMQP.py --broker rabbitmq --dicom http://datastore:5106

Finally, add the following to the docker-compose.yml file:

unet-cell-segmentation: image: thrive20/unet-cell-segmentation-focal:latest ports: - 7106 environment: MSG_SYSTEM: amqp SERVICE_NAME: unet-cell-segmentation--v1_0_0 SERVICE_TAGS: analytic Again, replace "unet-cell" with your own algorithm name.

Run docker-compose

Run:

$ docker-compose up

Features of Rt. 106

• Based on Docker for easy setup and configuration.
• Entirely web-based for easy deployment in cloud or on premise.
• Displays DICOM or TIFF images using the Cornerstone image viewer, along with mask overlays.
• An Algorithm SDK makes it easy to add new algorithms.
• The user interface is easily customizable to create custom applications.
• A file-based datastore is provided having a well-defined REST API so that alternative storage mechanisms can be integrated or developed.
• Algorithm executions are tracked in a database for later reference.
• Initial set of automated quality tests.

---