Federated monitoring of chest X-ray AI proves feasible in two-site NHS in silico study

Real-time federated monitoring of chest X-ray AI was feasible across 2 NHS sites, detecting data drift and changing human-AI agreement patterns without moving patient data.

• In silico feasibility study at 2 NHS Trusts using federated architecture to monitor third-party chest X-ray AI applications.
• Input monitoring found outlier exams from image quality issues; AI inference values remained stable with limited drift events.
• Human-AI agreement increased over time at one site, suggesting possible evolving automation bias.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.