Why Health Systems Need to Worry About Radiology

When a hospital's surgical volume slows, finance notices. When emergency department throughput backs up, operations notices. When a high-value service line underperforms, the C-suite notices. What rarely gets traced back to its source is the radiology department — the diagnostic infrastructure that every one of those functions depends on, and the one health systems have historically underinvested in.

The radiologists who spoke with the Rad AI team recently weren’t talking about the future. They were describing conditions that exist right now, with direct consequences for the clinical and financial performance of the health systems they serve. 

The Problem Is Already in Your Metrics

"Triaging is really difficult. Everything now is ordered stat. They want a read within 30 minutes," said Divya Kumari, MD, an interventional radiologist in Illinois. "And if you're bogged down by a list you have to do in chronological order, you may miss what's most abnormal and most urgent."

That isn’t a description of an overwhelmed radiologist. It’s a description of a broken triage system — one in which the physician responsible for identifying the critical findings is prevented from doing so by the volume of non-critical ones. When that system breaks down, the consequences don’t stay in the reading room. They show up in surgical scheduling bottlenecks and ED length of stay.

AI triage systems deployed in acute care settings have been reported to get flagged studies to radiologists for review 20 to 30 minutes faster than standard worklist order. Cutting imaging workflow friction has been associated with nearly a full hour reduction in ED length of stay per patient. These are system metrics, but radiology is where they are won or lost.

The workforce behind that system is under compounding pressure that no single initiative will fix. Imaging utilization among Medicare beneficiaries grew 13% per patient between 2005 and 2021. Radiologist attrition rates more than doubled between 2014 and 2022. The Health Resources and Services Administration (HRSA) projects radiology at roughly 90% workforce adequacy by 2038, a gap the training pipeline can’t close on its own. These aren’t radiology department problems. They’re system design problems with system-level consequences.

What You’re Leaving on the Table

Radiologists functioning as consultants before imaging is ordered, not just after it arrives, can meaningfully reduce the estimated 30% of imaging studies ordered in current practice without a clear clinical indication. That is cost, radiation exposure and false positive risk with no clinical gain. But consultative capacity is exactly what disappears first when volume overwhelms bandwidth. The radiologist buried in a chronological worklist isn’t available to advise on appropriateness, flag unnecessary studies or redirect a clinical workup before it goes in the wrong direction.

The same constraint limits a specific and measurable revenue opportunity. Health systems that have built structured workflows around incidental findings — the coronary calcium discovered on a trauma CT or the lung nodule found on a pneumonia scan — are generating an estimated $2 to $4 million in additional contribution margin annually through appropriate downstream care. The radiologist is the physician positioned to surface those findings, track them and route them into follow-up pathways. Without the infrastructure to support that function, the value stays buried in the report and the patient stays at risk.

"As opposed to just a simple service, like a lab value, a lot goes into what we do," said Jean Jeudy, MD, a professor and vice chair for informatics in Maryland. "Our close clinicians understand that. They refer to us for consultation because they know we can help them with our tools." 

AI Is a Necessity. So Is a Radiologist’s Perspective

There are now more than 1,000 FDA-cleared AI algorithms approved for clinical use in radiology. Only around 30% of radiologists currently integrate AI into their routine workflows. The gap isn’t a technology problem, rather it’s an implementation and culture problem and closing it requires health system leadership to be in the room.

The radiologists who are paying attention have moved past the replacement debate. "It's not a matter of if, but when," said Carlos Anaya, MD, an interventional radiologist in Puerto Rico. "I want to learn about it as much as I can. The more efficient we can become, the better we can do for our patients and the entire radiology community." 

The next generation entering the field has made AI readiness a criterion for choosing where to work. "It would be prudent to ask about a practice's AI strategy," said Jay Gupta, MD, a radiology resident in Ohio. "I want to find practices that are more forward AI-thinking or technology-driven." 

But investment in AI alone won’t solve the problem. A 2023 study published in JAMA Network Open found that frequent AI use was associated with an increased risk of burnout, particularly among radiologists with high existing workloads or low AI acceptance. AI that generates alert fatigue, requires constant verification of low-specificity flags or is layered onto an already overwhelmed reading environment makes the problem worse. 

The implementation has to be designed around the radiologist's actual workflow, which means understanding what that workflow looks like, what breaks it and what would actually relieve it. That requires asking radiologists how AI needs to function, but many health systems haven’t given them a seat at the table.

What Changes When Radiologists Have More Time

"The AI is going to get pretty good at saying: This is a normal exam that may only need eyes on by a mid-level. So, we'll be focused on the more complex exams, which will require more communication with the clinical teams," said Elizabeth Bergey, MD, Chief Clinical Officer at Rad AI and a former pediatric radiologist. AI handling normal exams isn’t a future scenario, it’s a near-term operational necessity, and the health systems that treat it as one will have a structural advantage over those that don’t.

What changes when a radiologist has margin matters as much as what gets automated. A radiologist with more time picks up the phone. "I talk to our referring clinicians every day," said Taylor Pomeranz, MD, a neuroradiologist in Ohio. "The old-fashioned stuff still matters." That call — the one that catches the finding before the patient is discharged, the one that changes the treatment plan — is what a stretched radiologist may not have time to make. The cost of that missed call doesn’t appear on the radiology P&L. It appears elsewhere, later, in ways that are harder to trace and more expensive to fix.

"The more clear and accurate the radiology report is, the easier it is for our ordering providers to grasp and make use of the information," said Jeff Chang, MD, Co-Founder and Chief Product Officer of Rad AI and a former radiologist. Report quality, not just speed, determines the downstream value of every imaging study a health system orders. It’s also the lever that most health systems have barely begun to pull.

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.