Leiden University Medical Centre

LUMC processes approximately 23,000 chest radiographs annually, with roughly half yielding normal findings. Prior to AI deployment, each study required 3–4 minutes of manual radiologist time, creating unsustainable workloads—especially for residents on overnight and weekend shifts with limited senior oversight. The absence of AI-assisted triage meant complex and routine studies were queued together, risking delayed attention to urgent cases.

LUMC deployed Oxipit ChestLink, a CE Class IIb-certified autonomous AI tool, to identify and auto-report normal chest radiographs. The solution was fully integrated into the hospital's PACS worklist, eliminating the need for external portals. Oxipit Quality, a secondary-read QA tool, was deployed in parallel to flag discrepancies between AI findings and radiologist reports. Implementation began with retrospective and prospective validation on ~20,000 chest radiographs before full clinical rollout.

ChestLink autonomously reported 15–20% of chest radiograph studies as normal with 99.9% sensitivity, freeing radiologist time for complex cases and reducing cognitive load on night-shift residents. Oxipit Quality flagged several clinically significant discrepancies, including a missed pulmonary nodule and subtle fractures, prompting reassessments. The department maintained or improved turnaround consistency and began piloting regional deployment to satellite hospitals.

• Seamless PACS integration—eliminating external portals—was the single most important factor in clinician adoption and workflow continuity.
• Beginning with binary triage (normal vs. abnormal) rather than full autonomous reporting was a pragmatic stepping stone that built clinical trust before expanding AI autonomy.
• Structured change management, including transparent dialogue about liability and including radiologists in deployment decisions, was as critical as algorithmic performance.