| | SEPTEMBER 20248UTLOOKHealthcare Tech IN MY OPINIONNavigating the Future: Challenges and Promises of AI in RadiologyBy Roger Staff, Head of Imaging Physics, NHS GrampianArtificial intelligence (AI) has garnered considerable attention in radiology, and the potential for `faster, cheaper, safer' makes it a dynamic and exciting field. This is particularly true when it comes to reporting. Tools to aid diagnostic decision-making are in the marketplace, with CE marks and legislative approval. Given this landscape, the possibility of independent diagnostic decision-making AI does not seem far away, which is attractive in environments with limited resources and perhaps a more relaxed legislative environment. Diagnostic accuracy and workflow efficiency improvements ultimately promise to deliver better patient care. However, the introduction is not without its challenges, and radiology grapples with complexities as it navigates the path toward a more AI-inclusive future.The key questions when procuring such tools are familiar, although nuanced in an AI context. Does it work in a population I plan to use it with, and how will you demonstrate that it works? It is unknown if a mammography tool trained using central European breast will work on sub-Saharan African breasts. Moreover, how might a vendor demonstrate that it will work in any given population? It may be that differences between populations in the context of the AI tool are minimal, and the tool's performance does not vary between populations. Still, thus far, this is generally unknown. The next challenge is how the tool might fit into the existing workflows and how the AI's `view' is presented to the current practitioners. Without onboarding these critical stakeholders, gaining traction for any tool would be difficult. Various techniques, such as heat maps or ROIs, are available to highlight areas within an image for special attention to simple yes-no responses. Radiologists are accustomed to a well-established image presentation and interpretation routine, relying on their expertise and experience. Performance gains may Roger Staff
< Page 7 | Page 9 >