Hospitals need a management strategy for radiology AI before investing
As artificial intelligence (AI) increasingly moves into clinical practice, hospitals face new challenges of determining not simply whether an AI tool works, but how it will be maintained and fit into the health system over time.
Anand Singh, MD, FACR, chair of the Radiology Partners National Quality and Safety Committee, associate director of Mosaic Clinical Technologies, chair of the American College of Radiology (ACR) Military Committee, and a member of the ACR Peer Learning Committee, said hospitals should begin by identifying the specific problems they want AI to solve rather than starting with the technology itself. He spoke with Radiology Business in the above video interview.
Those problems can include limited access, capacity constraints, workforce shortages, variability in reporting, patient safety and follow-up management, Singh said. Once the problem is defined, health systems can evaluate which AI solutions may address it.
Just as important, he said, is ensuring the technology does not add friction to clinical workflows.
“For radiologists, I don't want to read more things, have more clicks, have more alerts, or make work harder,” Singh explained.
That consideration becomes increasingly important after implementation. Unlike traditional software, AI systems can evolve and need to be monitored because outputs may begin to drift if inputs are changed, such as a new scanners, software updates or CT slice thickness. This requires hospitals to continuously evaluate their AI performance and impact.
Singh described this as the “continuous life cycle” of AI and said health systems will need a concept of clinical AI stewardship. Hospitals should outline how they will measure an AI system's success and, more importantly, its overall value.
Turnaround time and productivity are important measures, he said, but hospitals also should examine capacity, access, patient safety, consistency, downstream impacts on costs, revenue or patient outcomes, and the broader value delivered to the organization.
Moving beyond a collection of AI applications
Another strategic consideration is how multiple AI applications will be managed. Rather than deploying numerous independent tools, Singh said an AI platform-based approach can help hospitals reduce interoperability challenges and simplify monitoring and governance.
He said the proliferation of individual AI applications can create a downstream management burden. A health system with dozens of AI applications may have to monitor each one, assess its failure modes and manage separate workflows. A unified platform can provide a common architecture for applications ranging from detection to reporting and quality and safety, while also creating a framework for validation, monitoring and clinical governance.
Integration into existing clinical systems is also critical. AI that requires additional logins, alerts or workflow steps can undermine adoption. Singh said a unified architecture can help make implementation and ongoing oversight more seamless.
The need for a clinical AI infrastructure
Singh also emphasized that AI infrastructure should not be viewed solely as an IT issue involving computing power, storage, cloud environments and cybersecurity. Instead, hospitals need a form of clinical AI infrastructure, involving clinicians and other staff stakeholders who can provide feedback, oversee AI outputs, monitor safety signals, and help manage the interaction between humans and AI. That includes validation, monitoring, clinical oversight, workforce training and mechanisms for identifying and addressing potential safety issues.
The need for oversight becomes especially important as AI systems encounter changes in clinical practice. Changes in imaging protocols or acquisition parameters, for example, can affect an algorithm's performance.
Traditional monitoring focuses on whether the AI model continues to perform as expected. But Singh said hospitals increasingly will need to monitor something broader, the performance of the human and AI together. That means evaluating not only whether a model produces accurate results, but also how clinicians interact with those results and whether changes in technology alter clinical decision-making or introduce new risks.
Looking ahead, Singh said hospitals should build today's AI investments around the ability to scale and adapt as new applications emerge.
He said the key strategic questions are straightforward: What problem are we trying to solve? How will we measure success and value? And how will we responsibly scale AI while managing both the technology and the human-AI interaction?
The answers could determine whether AI becomes another collection of disconnected tools, or an integrated component of clinical care and patient safety.
Dave Fornell has covered healthcare for more than 17 years, with a focus in cardiology and radiology. Fornell is a 5-time winner of a Jesse H. Neal Award, the most prestigious editorial honors in the field of specialized journalism. The wins included best technical content, best use of social media and best COVID-19 coverage. Fornell was also a three-time Neal finalist for best range of work by a single author. He produces more than 100 editorial videos each year, most of them interviews with key opinion leaders in medicine. He also writes technical articles, covers key trends, conducts video hospital site visits, and is very involved with social media. E-mail: [email protected]