A life-cycle approach to AI governance in medical device development

Abstract

Medical devices increasingly incorporate AI-enabled functions whose performance depends on development and evaluation data, statistical modelling, deployment conditions, human interaction, and continuing control of change. These dependencies create a governance problem that cannot be resolved by treating an AI model as an ordinary software component, or by adding a one-time AI assessment to the product file.

At any point in the life cycle, a manufacturer should be able to determine what system is being governed, which uses are authorized, which model and data baselines support each use, which assumptions, risks, and limitations were accepted, what evidence supported approval, and which changes or operating signals require those decisions to be revisited. This paper proposes a vendor-neutral life-cycle architecture for an artificial intelligence management system in medical device development, extending from initial concept and product-realisation planning through design, development, release, deployment, post-market surveillance, and modification.

Contents

• A layered standards architecture
• The proposed entity and relationship model
• Integrated AI and medical-device risk management
• Integration with the ISO 13485 quality management system
• Life-cycle governance and decision gates
• Controlled evolution and PCCP operationalisation
• Application beyond medical-device development