The demand for radiological reports outstripped capacity for reporting throughout the network. Examinations were handled in order of arrival, such that clinically important findings were delayed behind routine screening exams.

Improve the worklist, not the diagnosis. The AI flagged the studies based on their likelihood of significant findings, but the radiologists remained entirely responsible for the diagnosis, reporting, and clinical decision-making. Viewing the solution as worklist prioritization, not autonomous diagnosis, allowed the clinical governance team to approve it.

Ingestion of DICOM files from the existing PACS; AWS inference; prioritized worklists written back into the existing PACS/RIS system; extensive audit logging for each model prediction; thresholds determined by the clinical governance team and approved by them.