What Radiology Demonstrates about Medicine, AI and ‘Replacing’ Physicians
Radiology is the specialty most often named as the first one AI will replace, and its record points the other way. Digitization, teleradiology, narrow AI, and foundation models each arrived with the potential for upheaval, and at every step the specialty grew, its relevance increased, and it needed more radiologists. Radiology is still struggling to find enough radiologists for the work. Dr. Ezequiel “Zeke” Silva III is a diagnostic and interventional radiologist who has seen that history from the inside. He has chaired the AMA RVS Update Committee for the last five years, helped establish and lead the AMA Digital Medicine Payment Advisory Group, and chaired the ACR Task Force on Teleradiology Practice, where he was lead author of the ACR White Paper on Teleradiology Practice. He has 77 peer-reviewed publications and has spent more than a decade working on how coding, payment, and coverage work for physicians.
What Radiology’s Record Shows About AI and Physicians
Radiology was among the first specialties to be digitized. DICOM allowed images to move between hospitals, across the country, and internationally, so commoditization pressure reached the field early. Narrow AI followed, and radiology had large digitized data sets that radiology reports had effectively annotated. Foundation models came next, and agentic AI is now moving from generating information toward influencing and making clinical decisions.
Zeke pointed out that any one of these steps could have been predicted to disrupt the specialty and create upheaval within it. At every one of them the specialty only grew and its relevance only increased. From a workforce perspective, the changes led to radiology requiring more radiologists to meet the demand they created, always with a focus on proper and superior patient care. For other specialties hearing that AI will replace physicians, radiology is an existing example with the longest record, and the conversation about what replacement would mean elsewhere can start from that record.
Zeke does not treat the next step as settled. Agentic AI is arriving faster than many expected, and the payment and regulatory systems probably need to catch up. His view is that radiology’s history gives the field a basis for meeting that change, and he attributes his own contribution in these discussions to a physician first lens that he brings to every conversation with innovators, policymakers, and the RUC.
The ACR Teleradiology Task Force
The task force began around 2010, as teleradiology spread from local practice to national and international scale. The impetus was a perceived threat to traditional on-site practice. The group did not find a way to limit the expansion of teleradiology, and early in the effort, with credit going to his task force members, it concluded that the technology was ready for prime time and growing and that radiology needed to embrace it while keeping a patient-centered, physician-centered focus.
The group then turned to how teleradiology could be done in the best way possible. That work covered technical standards and clinical guidelines, along with the teleradiologist’s reading room, the regulatory space, and licensure. The paper was published in 2013 and tied to existing ACR standards so that it could evolve over time. Many in the industry expected a self-protective effort from the ACR, and the teleradiology industry praised the paper for acknowledging the technology and supporting its growth with patient primacy at the center.
Coding, Payment, Coverage, and Adoption
Zeke treats payment as relevant to the diffusion of innovation and expects developers to think about it. He also describes a great technology as insufficient to reach clinical care on its own. A CPT code describes a service and allows it to be reported for billing, and having a code does not equal payment. The RUC values services and makes recommendations to CMS, and payment does not equal coverage. Coverage determinations by Medicare and large private payers carry their own weight, and coverage by itself does not produce diffusion or clinical meaningfulness for a technology.
Each step can be a bottleneck, and each hurdle is not always easy to cross. Zeke’s approach is to view them as individual hurdles in one continuous race and to build that mindset very early in the process. The AMA Digital Medicine Payment Advisory Group works on this question and wants to hear from physicians and organizations about how the codes are working in practice.
Physician Engagement During Development and Deployment
Zeke drafted and presented his first CPT code in the mid-2000s, for fiducial marker placement in the chest. He was performing the procedure in the outpatient setting and learned that no code existed and that his patients were receiving explanations of benefits and going uncovered. He sees the technologies that succeed as rooted in a clinical opportunity, a problem looking for a solution.
From a practicing physician’s view, the arrival of a technology in clinical care is step one, and for physicians that step is the easy part. The questions that follow concern clinical care, workflow, and the ability to deliver better, more efficient, high-quality care. If a physician does not believe in a tool, it goes no further. Zeke describes physician hesitation as quite practical in nature, and he expects that asking physicians early how they view a technology improves the chance of it scaling once many physicians are using it with many patients. Physician autonomy sits underneath this, since clinical decisions rest on clinical data and also on each patient’s needs, beliefs, and fears.
Payment and Regulation as AI Makes Decisions
Payment systems such as the Medicare Physician Fee Schedule and the hospital-based systems were built around what physicians and other qualified health professionals do, because humans were the only ones providing those services. Software, and in the near future agents, now provide some of them. Zeke sees an opportunity to evolve these systems to embrace a machine and human pairing while keeping the focus on patient care, outcomes, and value. The systems have adapted to new technology before, as with laparoscopic surgery, endoscopy, and interventional radiology.
Shared vocabulary is also part of this. Stakeholders describe the same software in different terms. Zeke pointed to the assistive, augmentative, and autonomous framework and to the risk-based work of the International Medical Device Regulators Forum, and he noted that radiology standardized its own language through DICOM, which supports consistent interpretation of images across systems.
Advice for Physicians
Zeke encourages physicians to ask the most basic questions when a new product arrives, to learn what the technologies are doing and understand the taxonomy, and to keep their physician first language. He suggests asking what a tool means for the patient in front of you and reporting how it worked on a real patient last week.
Lessons for Healthcare Leaders
Radiology has met digitization, teleradiology, narrow AI, and foundation models, and at each step it grew and needed more radiologists. Use that record when the conversation turns to AI replacing physicians.
Expect agentic AI to arrive faster than planned, and update payment and regulatory frameworks as AI moves from generating information to making decisions.
Plan for coding, payment, coverage, and adoption as four separate hurdles in one continuous race.
A great technology does not reach clinical care on its own, and each step along the way can become a bottleneck.
Start from a clinical opportunity, a problem looking for a solution.
Engage physicians during development and again during deployment, and ask early how they view a technology.
Tie governance documents to standards that can evolve, as the ACR teleradiology paper did, and develop shared vocabulary now.
Encourage physicians to ask basic questions and report their experience with tools on real patients.
Links to our Full Episode: Watch on YouTube, listen on Spotify or Apple Podcasts or visit Ardexia Insights Webpage
Follow Dr. Zeke Silva on LinkedIn
Related Resources
Episode 13: You Cannot Mandate Your Way to Change and Adoption with Dr. Ethan Booker
Episode 12:Building Digital Health Adoption from the Inside out
Episode 11 Why Most Digital Health Programs Never Cross the Threshold
Follow us on those outlets to hear our next episode, drops every month.
Dr. Aditi Joshi is the CEO of Ardexia and host of the Ardexia podcast. She's an emergency physician who has built multiple digital health programs across three continents and specializes in turning failed digital health implementations into measurable clinical and financial success.