Interoperable clinical data integration for distributed medical AI: The Open Medical Inference Gateway

Background: Integrating AI inference services into clinical workflows remains a persistent challenge, primarily because inference backends and healthcare data standards operate with fundamentally different assumptions. The EU AI Act further requires transparency and traceability for high-risk AI systems in healthcare. Methods: This paper presents the Open Medical Inference (OMI) Gateway, an interoperability layer that mediates between FHIR- and DICOM-based clinical data environments and heterogeneous AI inference backends. The gateway was implemented in Python using FastAPI with asynchronous job processing via Celery and RabbitMQ. The architecture follows a service-oriented plugin model that separates shared interoperability concerns from service-specific transformation logic. Input and output data are validated against FHIR profiles and DICOM specifications at multiple pipeline stages, producing OperationOutcome resources as a machine-readable audit trail. Results: End-to-end pipeline latency was measured across 100 sequential inference requests, yielding a median latency of 263 ms and a 100% success rate. The design was evaluated against interoperability requirements from German Medical Informatics Initiative (MII) Data Integration Centers. The system is deployable with a single command and available as open-source software. Discussion: The results indicate that embedding validation and structured artifact handling directly into the inference execution path yields a more governable and auditable integration model than ad-hoc service adapters.