Concepts Overview

The User Guide and Extension Development Guide describe how to use and extend Kaapana. This section instead explains why some of its core subsystems are built the way they are: the reasoning behind a design, how the pieces fit together, and what is still evolving.

Platform Architecture

• Backend Architecture – how the backend moved from one monolithic service to several narrowly scoped ones.

• Multi-Node Architecture – running Airflow workers, GPU workflows, PACS and MinIO across more than one machine.

• GPU Sharing Strategy – letting several tasks share one physical GPU instead of reserving a whole device per task.

• Monitoring and Resource Allocation – how CPU, RAM and GPU are scheduled, and how cluster/task health is exposed.

Data & Storage

• Cloud Storage – where the data is actually stored on single-node vs. multi-node deployments.

• External PACS – pointing the platform’s DICOMweb ingestion/query path at an external PACS.

• Workflow Storage – how a workflow’s processing pods get and hand off their data.

• File Types – what file types are supported, through which mechanism, and what else is possible.

Workflows & Extensions

• Distributing Extensions – how an extension is packaged and distributed and how it gets into Kaapana.

• Interactive Workflows – how a workflow spawns an application and waits on it for confirmations, annotation, or other human-in-the-loop steps.

• Federated Learning – running a computation across multiple Kaapana instances without moving raw patient data.

Access & Frontend

• Access Control – controlling access to data, workflows and other resources.

• Authorization Flow – how a request is authenticated and authorized on its way to a backend service.

• Calling Kaapana Programmatically – calling Kaapana’s workflow APIs from a script or notebook.

• Connecting Viewers and Applications to Data – why connecting a viewer or application is a matter of configuration, not code.