AI à la Carte: From Prototype to Production – Governing Composable Automation and AI Workflows

AI a la Carte describes a lightweight innovation layer – automation software plus modular AI API endpoints – that lets radiologists and imaging informaticists prototype and integrate AI-enabled capabilities inside the reporting systems they already use. The companion SIIM26 session showed how that layer is built. This webinar takes up the question that session deliberately left open: once local builders can prototype useful capabilities, how does an institution approve, deploy, and govern them responsibly?

Rather than treating governance as a single approval step, the session presents it as three connected dimensions. First, deployment approval – how a locally built workflow is cleared to run in the hospital environment; even the automation layer, though safer than altering enterprise software, carries risk and needs an approval path. Second, the review body and process – how radiologists, developers, and IT/IS form a committee to review, approve, and set policy, and how that stays agile and responsive without sacrificing safety. Third, the technical governance surface – the four-stage workflow loop (trigger, context capture, AI invocation, output routing) across both the automation and API layers, including the difference between self-hosted API endpoints an institution controls and fixed vendor endpoints it only consumes. Throughout, we map concrete responsibilities – capability approval and maintenance, prompt and model validation and drift monitoring, PHI routing and security, output auditing, and clear ownership – onto these dimensions. The intended audience is radiologists, developers, imaging informaticists and institutional stakeholders responsible for operationalizing AI governance.

Objectives

• Define an approval pathway for deploying locally built automation and AI workflows into the clinical environment.
• Describe how radiologists, developers, and IT/IS form a review committee and policy for these workflows, and how that process stays agile and responsive without sacrificing safety.
• Map the technical governance surface across both layers using the four-stage loop (trigger, context and PHI, AI invocation, output routing).
• Identify lifecycle responsibilities including validation, monitoring, ownership, and maintenance of deployed AI workflows.

CE Credit     

The Society for Imaging Informatics in Medicine has approved this activity for 1.0 hours of SIIM IIP Credits towards certification and re-certification by the American Board of Imaging Informatics (ABII).