From procurement to post-market surveillance: a practical guide to AI governance in radiology
This article outlines practical guidance for governing AI in radiology, emphasising that successful implementation depends as much on governance and integration as on technical performance. Although many AI tools are clinically approved and widely available, relatively few are effectively used in practice, often due to challenges in workflow fit, infrastructure and oversight.
AI governance is presented as a continuous life cycle rather than a single approval step. It begins with clearly defining the clinical problem and determining whether AI is an appropriate solution. Establishing baseline performance is crucial so that benefits and risks can be objectively measured after deployment. Procurement requires more than verifying regulatory approval (eg UKCA/CE marking); organisations must critically evaluate evidence, clinical relevance and suitability for local workflows.
The article highlights the importance of organisational readiness, including technical infrastructure (such as PACS/RIS integration), data governance and compliance with safety standards like DCB0129 and DCB0160. Workforce preparation is equally critical, ensuring clinicians understand the tool’s function, limitations and escalation processes.
Local validation is essential before influencing patient care, as AI performance can vary across populations and settings. Strategies such as ‘shadow-mode’ deployment allow safe real-world testing. A structured standard operating procedure should then guide implementation and ongoing use.
Finally, continuous monitoring and post-market surveillance are key. Using frameworks such as the ‘three Ps’ (people, process, product), organisations can track performance, identify risks and ensure safe integration. The authors conclude that AI governance must be an ongoing, adaptive process extending from procurement through to long-term clinical use.
To download and read this article as a pdf please sign up to our newsletter.
The content on this page is provided by the individuals concerned and does not represent the views or opinions of RAD Magazine.
