Published
September 1, 2026
The AI and radiology conversation has been persistent and largely conducted by people who aren’t radiologists. What hasn’t generated as much mainstream attention is what radiologists themselves have been saying — about the work, about the conditions around it and about what is actually breaking down inside health systems right now, independent of any AI timeline.
Recently, the Rad AI team interviewed 12 radiologists — including residents and department chairs, diagnosticians, interventionalists and teleradiologists — and asked them to describe their work in their own words. What they said reflects the operational reality of modern health systems that most of the AI debate hasn’t come close to touching.
The scope of what radiology actually does in a health system is foundational, which makes the replacement argument moot, but the scope of workflow challenges is difficult to ignore.
"Not any one patient gets through an emergency visit or a hospital visit without radiology being integral to how the patient's path goes. We interact with all of the various specialties, and imaging is a part of almost all diagnosis and treatment these days," said Eric Brandser, MD, a radiologist in Kentucky.
The clinical stakes are just as clear. "Whenever something's going wrong with the patient, there's only so much physical exam and labs can offer. With imaging, you can directly tell the referring physician what's going on with the patient. And that's invaluable, especially in a timely manner," said Divya Kumari, MD, interventional radiologist in Illinois.
Yet the profession is burdened by bottlenecks that reach beyond any individual department. "If I had more time, if people understood how the inefficiency decreases our ability, our output, I think that they would try to provide or at least embrace other ways to help us become more efficient," said Jean Jeudy, MD, a professor and vice chair for informatics in Maryland.
Irreplaceability isn’t the same as being well-supported. The gap between the two is the conversation that radiologists want people to have.
A 2025 survey found that 67% of radiology practices are understaffed amid rising exam volumes. Real reimbursement per patient fell 25% between 2005 and 2021, while volumes climbed. The workforce is absorbing more work for less compensation with deteriorating support infrastructure, and the consequences don’t stay inside the radiology department.
Delayed reads mean delayed clinical decisions. Overloaded radiologists mean the potential for errors and faster attrition. That impact is downstream. Attrition shows up in surgical scheduling delays, in emergency department length of stay when imaging backs up throughput and in stroke outcomes when door-to-needle times slip. The radiology department is where those metrics are frequently won or lost — they just don't always carry radiology's name when they surface in operations reviews.
The same pressure compounds through the workforce in ways that most AI procurement conversations never reach. "AI's biggest challenge isn't the algorithm, it's how to operationalize the process, specifically with the revenue cycle management and peer learning processes. That's where I think AI has the biggest challenge but also the biggest opportunity," said Rhett Smith, MD, a teleradiologist in Utah.
The radiologists in this series were consistent on one point: The value they provide is consultative, not transactional, and no algorithm changes that. A radiologist functioning that way generates value that flows across every service line that depends on imaging. The question for health systems isn’t whether radiologists can be replaced. It’s whether the conditions exist to let them practice at the highest level.
This is part of a series drawn from original interviews conducted with 12 radiologists across career stages, subspecialties and practice settings. Quotes have been edited for brevity and clarity.
Expand Your Expertise
The Readout explores the technologies, challenges and realities shaping where care goes next.
Radiology-First AI
Actually Works