Radiologists Aren't Being Replaced by AI. So Why Does It Keep Coming Up?

The debate over AI and radiology has a habit of flaring up whenever someone outside the specialty says something provocative. In March 2026, Mitchell H. Katz, MD, president and CEO of NYC Health + Hospitals, provided the latest spark. At a panel discussion hosted by Crain's New York Business, he said: "We could replace a great deal of radiologists with AI at this moment, if we are ready to do the regulatory challenge."

The radiology community pushed back hard. Mohammed Suhail, MD, a San Diego-based radiologist, told Radiology Business the comments were "undeniable proof that confidently uninformed hospital administrators are a danger to patients" — reflecting, he said, "zero understanding of radiology." The controversy followed similar remarks from Anthropic CEO Dario Amodei, PhD, who had recently claimed AI had effectively taken over radiology's core diagnostic function, a characterization that radiologists disputed in equally forceful terms.

These moments matter. Not because they reflect where the technology actually is but because they reveal how wide the gap remains between the people making decisions about radiology and the people practicing it. Rishi Seth, MD, CIIP, a neuroradiologist and Chief Medical Innovation Officer at Rad AI, addressed the Katz comments directly in a piece published in AuntMinnie. His argument cuts to the center of what the replacement debate keeps getting wrong.

The Replacement Argument Misunderstands What Radiologists Do

"Radiology isn’t image classification," Dr. Seth wrote. "It’s the integration of imaging findings into a clinical narrative. We don’t simply identify abnormalities; we determine what they mean in the context of a patient's history, symptoms, prior studies and clinical trajectory. We weigh uncertainty, assess risk and decide what matters."

That distinction — between detection and judgment — is where the replacement argument collapses. Most radiology AI tools are narrow by design, built to detect a single finding on a single modality. A complex cross-sectional examination requires simultaneous assessment across dozens of organ systems. Vision language models show promise in this direction, but significant gaps remain. Even where detection is strong, judgment and reasoning are still absent.

The liability dimension of that gap is the piece of the argument that cost-focused health system leaders most often miss. "The 'major savings' argument for replacing radiologists is shortsighted," Dr. Seth noted, pointing out that radiologist salaries account for less than 1% of hospital operating costs, compared with administrative overhead, which represents nearly a third of U.S. healthcare spending. 

The more consequential cost is accountability: When the radiologist is removed, the clear chain of responsibility disappears. Current AI systems are approved only for clinical decision support, and vendors explicitly disclaim diagnostic responsibility, creating an undefined liability framework for any hospital that deploys them autonomously.

Andrew Del Gaizo, MD, a body radiologist and Chief Medical Information Officer at Rad AI, made the same point from a different angle in an interview with Diagnostic Imaging following the Katz comments. "Rather than replacing a radiologist, there's going to be a different role for radiologists moving forward," he said. "I think radiologists adapt. I think there's actually new opportunities in what we do."

Even David Lubarsky, MD, MBA, president and CEO of Westchester Medical Center Health Network — one of the panelists alongside Katz, subsequently clarified at Becker's 16th Annual Meeting that his position had been misrepresented. What he had actually said was that AI performs well in a specific and narrow circumstance: "In low-risk patients who have an essentially negative scan, AI can read that correctly 99.97% of the time, as good as a radiologist." That isn’t a replacement argument. It’s a triage argument, and it’s precisely the model that radiologists have been advocating for years.

The Predictions That Didn't Hold

The replacement narrative isn’t new. As most readers of this article will likely know, in 2016, Geoffrey Hinton predicted that training new radiologists would soon become unnecessary. Andrew Ng raised similar concerns about the vulnerability of specialized physicians to displacement. Those predictions haven’t held. Not because AI stalled but because the real-world practice of medicine proved more complex than early assumptions allowed. Both researchers have since revised their views toward collaboration rather than replacement.

Nvidia CEO Jensen Huang put it plainly: "The surprising thing is the prediction that radiologists would be the first jobs to go was exactly the opposite." More radiologists are being hired now, in part because of AI, not despite it.

The fear that peaked around 2016 was never entirely about technology, it was about trust. 

"In the beginning, companies really didn’t take the radiologist's point of view," said Kaustav Bera, MD, a body imaging fellow in Ohio, who spoke with the Rad AI team recently. "They were trying to appeal more to physicians downstream, and that's why the fear came up. Now, companies are involving radiologists from the ground up. Because if you're dealing with images, who better than radiologists?"

A 2024 survey found that 61% of radiologists now view AI as an opportunity rather than a threat. Eighty-four percent consider the final assessment by a radiologist still essential. The replacement debate hasn’t ended, but among practicing radiologists, it has largely moved to the margins.

What AI Actually Does in Practice

The honest account of what AI does in a daily radiology workflow is less dramatic than the headlines suggest — and more valuable for it.

"I already read a lot of studies that have an AI diagnosis on X-rays, and it speeds up the process slightly. I still have to read it," said Rhett Smith, MD, a teleradiologist from Utah, told the Rad AI team. Dr. Smith's framing is useful precisely because it isn’t dramatic. AI isn’t transforming his practice in a single moment, rather it’s incrementally reducing the friction of a workflow that has accumulated too much of it.

Dr. Seth's piece names the applications where that reduction in friction is most meaningful and most ready: identifying and triaging critical findings, reducing turnaround times, tracking incidental findings, assisting in report generation and flagging potential errors. "These tools augment clinical judgment, reduce cognitive burden and improve consistency, accuracy and efficiency," he wrote. "The role of the radiologist isn’t disappearing anytime soon. It is, however, shifting."

The cross-generational dimension of that shift is something the radiologists Rad AI interviewed noted as well. "Residents and junior attendings would be super helpful in showing the older generation how to use new tools," said Dr. Bera. "And they have a huge bank of knowledge we can refer to. It goes both ways." 

What Radiologists Are Building Toward

The most telling indicator of where radiology's relationship with AI is heading isn’t the headlines. It’s what individual radiologists are doing about their own futures.

At the time of his interview with Rad AI, Carlos Anaya, MD, an interventional radiologist from Puerto Rico, was establishing an independent practice. His reasoning was specific: "In case AI gets to that point it's so good that it's going to replace all of us, at least the service or a site where to provide the service will be of importance." Whatever AI can automate, the accountability for interpretation, communication and clinical judgment remains with a physician. Dr. Anaya is building toward that accountability rather than waiting to see what happens to it.

Doing something similar at a different scale is Niketa Chotai, MD, FRCR, a breast specialist radiologist based in Singapore and immediate past president of the Singapore Radiological Society's breast section. Over the past decade, she’s built an educational network reaching approximately 9,000 to 10,000 radiologists across nine countries, and AI in breast imaging is central to what she teaches. 

For Dr. Chotai, AI isn’t a replacement threat, but a mechanism for extending diagnostic reach into regions where radiologist capacity is far too low to meet growing demand. "We’re really working on educating patients, educating society and obviously educating the radiologist to do early diagnosis, which, hopefully, will improve the outcome and reduce the mortality from breast cancer patients."

Dr. Seth's conclusion applies equally to those building at the individual and the global level: "AI hasn’t yet crossed that threshold" of autonomous interpretation and until it does, "removing the physician from the diagnostic process isn’t progress. It’s a shift of risk onto patients without a system prepared to absorb it."

What the Transition Actually Requires

The radiologists navigating this moment share a common mindset. They’re curious about the technology without being credulous about it. They’re willing to adopt tools that genuinely help and willing to say clearly when tools don’t. They’re thinking about what the specialty needs to become, not just reacting to what others say it will be.

The next time a CEO makes a claim about replacing radiologists with AI, it would be worth asking which radiologists they consulted before saying so. The answer, in most cases, will be telling.

This is part of a series drawn from original interviews conducted with 12 radiologists across career stages, subspecialties and practice settings. Additional sourcing includes Rishi Seth, MD, CIIP, writing in AuntMinnie (April 28, 2026); Diagnostic Imaging (April 20, 2026); and Radiology Business (April 22, 2026). Quotes from the Rad AI interviews have been edited for brevity and clarity.