Continuos Delivery

[jordimassaguerpla]

Hi
I understand we can use monai-deploy-app-sdk to package the model into a container. But then, what? How do you envision we deploy that to "production" ? Which CD framework could be used for that task?

[MMelQin]

Good question. Like any other container workloads, there are many ways to host/run a containerized application.

• A MAP can be run as a standalone container processing a batch of files and save results. I have a guide on how this works on this board
• MONAI Deploy with its Informatics Gateway, Workflow Manager, and TaskManager with docker plug-in, can integrate with external systems, e.g. PACS, to receive DICOM studies, trigger MAP as a task, and then export results to PACS in DICOM instances
• NYU's Mecure supports MAP as a predefined processor
• FlyWheel supports MAP as a Gear
• Nuance, Answer Digital/NHS AIDE, also adopted MAP
• MAP can be used and hosted as an Nvidia Inference Microservice

An example list of adoptions and ways to host MAP are here in the repo.