Allen, Dreyer explain why radiologists must be involved in AI development

Radiologists must step up and get involved with the design and development of AI tools relevant to radiology, according to a new column published in the Journal of the American College of Radiology. By just taking a “wait and see” approach, specialists risk being left out of the conversation altogether.

Two prominent AI thought leaders penned the column: Bibb Allen, MD, with the department of radiology at Grandview Medical Center in Birmingham, Alabama and chief medical officer of the American College of Radiology (ACR) Data Science Institute (DSI), and Keith Dreyer, DO, PhD, vice chairman and associate professor of radiology at Massachusetts General Hospital (MGH) in Boston.

Allen and Dreyer opened their analysis by pointing to missed opportunities associated with the implementation of electronic health records (EHRs). “In the initial development of EHRs, physicians, specialty societies, and patients were not meaningfully engaged in initial EHR design and implementation, and although EHRs have been able to enhance the operational and business needs of hospitals and health systems, demonstrable improvements in physician efficiency and health outcomes have lagged behind,” they wrote.

With the development of AI, the authors added, radiologists can avoid making similar mistakes by being more directly involved in the entire process and demonstrating the value they bring to the table.

“Most radiology professionals do not have the data science background or the inclination to be involved in the computer science of creating AI algorithms; however, in the not-too-distant future, all of us will be using AI to augment and enhance the care we provide our patients,” Allen and Dreyer wrote. “To be most effective in clinical practice, use cases for AI algorithms must be designed to impact a specific clinical need, and the output of the model must then seamlessly interface with our clinical workflow and existing resources such as our PACS, EHR, reporting software, and our digital modalities.”

Researchers have already demonstrated that AI algorithms can detect certain findings—breast cancer nodules, for example—and help prioritize critical cases as needed. But there is still a significant amount of work to be done, the authors wrote, and radiologists can play a key role by influencing use case development and providing assistance in other ways as needed.

“The potential for AI to positively impact our practices and make radiology professionals even more valuable to our patients and health systems is enormous, but to be effective the design and development of these use cases must have substantial input from radiologists,” Allen and Dreyer wrote.

The ACR DSI stands as a primary example of how radiology can ensure it has a seat at this particular table. “AI use cases developed by the DSI will find the intersection between the clinical needs of our specialties and those problems that are solvable by AI and will be overseen by panels of practicing radiologists in all of the radiological subspecialties, interventional radiology, and radiation oncology,” Allen and Dreyer wrote. “They will include narrative descriptions and flowcharts that specify the goals the algorithm should meet, the required clinical inputs, how the output of the model should integrate into the clinical workflow, and how it should interface with both human end users and an array of electronic resources, such as reporting software, PACS and EHRs.”

DSI use cases will also assist with annotating data sets, training algorithms and clinical integration, the authors added. And they will be developed so that improvements can be made as needed.

More Radiology Business coverage of the ACR DSI is available here and here.