Drift isn’t a bug, it’s the work: post-deployment surveillance for clinical artificial intelligence

Key Points

• This research aims to investigate the importance of post-deployment surveillance in clinical artificial intelligence systems.
• Evaluated the operational infrastructure of clinical AI systems post-deployment.
• Conducted surveillance to monitor for system drift and integrity during operation.
• Assessed the effect of surveillance on the stability and reliability of AI outputs.
• Identified that system drift occurs regularly in clinical AI implementations.
• Demonstrated that surveillance mitigates operational risks associated with drift effectively.
• Highlighted the need for ongoing monitoring to ensure AI systems perform reliably over time.