Becoming the Doctor’s Doctor Again: Radiology’s Clinical Role in the AI Era
RADIOLOGY IN THE AGE OF AI & VLMS | ARTICLE 13 OF 14
There is a (bad) version of the next decade in which AI absorbs the high-volume screening work, productivity gains are harvested by whoever owns the contracts, and radiologists find themselves doing more reads per hour while clinical partners call someone else when they have a question. That version is not inevitable. But it is the default if we radiologists do not make a deliberate choice about what to do with the capacity that AI creates.
The term “doctor’s doctor” has a specific historical meaning in radiology. It describes a time when radiologists were the physicians other physicians called, not just to interpret films but to think through a clinical problem. What study should I order? Does this finding change management? Where does the differential go from here? That consultative function eroded as volume pressure increased and electronic report delivery removed the natural friction that used to produce a phone call. Radiology became faster and more efficient and, for large swaths of clinical medicine, more invisible.
AI may be the forcing function that reverses this, but only if radiologists meet it as an opportunity rather than a threat.
Where the Capacity Goes Is the Question
The data on where the specialty currently stands gives us a starting point. The Philips 2025 Future Health Index found that 43% of radiologists report spending less time with patients and more time on administrative tasks compared to five years ago.¹ The workflow pressure that produced this trend is real, and it will not be resolved by intention alone. But it can be resolved structurally, and that is exactly what AI-enabled workflow redesign makes possible if it is pursued by people who understand both the clinical context and the technology.
When AI handles the routine, the radiologist’s comparative advantage shifts toward the complex and the contextual. The chest X-ray that follows a standard clinical pathway is a different kind of problem than the immunocompromised patient whose chest X-ray does not match any pattern the algorithm was trained to recognize. The AI will produce a report for both. It cannot reliably tell you which one warrants a call to the clinician, which one requires prior imaging correlation that the algorithm did not have access to, and which one deserves a direct conversation rather than a PDF in the inbox. That judgment is not algorithmic. It is clinical, and it belongs to the physician whose name is on the report.
The Role AI Cannot Fill
This is not a nostalgic argument for doing things the old way. It is an argument about where value actually lives in a world where routine interpretation is increasingly automated. The radiologist who positions themselves as the clinical partner for AI governance, the physician who evaluates tools, monitors drift, adjudicates edge cases, and trains referring physicians on what AI can and cannot do, holds a role that no algorithm can occupy. This is the radiologist that every practice needs and that most practices are not systematically developing.
Prof. Annemiek Snoeckx of the University of Antwerp discussed this at ECR 2026. Speaking on chest X-ray in the AI era, she articulated a vision of the radiologist’s expertise that maps directly onto the consultation model: “We will be experts in critically supervising a hybrid system, in integrating tools with clinical responsibility, in selecting tools, in combining imaging with context. We will need to be experts in communication with referrers, in verification, and in knowing when the tool might be wrong, in building learning loops through audits and feedback. The expert teacher will be the one who teaches this new form of expertise.”² This is not a description of a radiologist performing volume reads with AI assistance.
It is a description of a physician whose primary value is metacognitive: knowing when to trust, when to override, and how to make others capable of doing the same.
The imaging stewardship opportunity is equally concrete. AI-powered appropriateness tools are already reducing low-value imaging in settings where they have been deployed with clinical rigor. But appropriateness guidance only improves care if someone with clinical authority is willing to own and advocate for the process. Radiologists are the obvious candidates. The department that positions itself as the institutional home for imaging stewardship, ordering criteria oversight, and AI-assisted appropriateness review is not competing on read volume. It is competing on something harder to automate.
Making the Consultation Model Operational
Participation in multidisciplinary tumor boards, oncology case conferences, orthopedic planning discussions, and other MDT settings has historically been limited by workflow friction. A radiologist reading 60 studies before noon cannot spend an hour on tumor board. The same radiologist whose AI handles 30 of those studies at draft quality, freeing an hour in the afternoon, can. The capacity exists. The question is whether the specialty treats that capacity as a margin to be used for additional volume or as an opportunity to rebuild clinical relationships that have been eroding for twenty years.
A Findings Checklist Workflow, which will be explored in full in an article after this series, is the structural mechanism that makes the Doctor’s Doctor role operational rather than aspirational. Intelligent downstream routing means that a radiologist identifying an unexpected finding in the correct clinical context does not simply generate a report. It triggers a coordinated response: the obstetrician is notified of the placenta previa, the pulmonologist receives the alert on the incidental nodule, the relevant team gets a task rather than a document buried in an inbox. A single radiologist workflow becomes a multi-stakeholder clinical coordination event. The consultation model is not just a posture. It becomes architecture.
What the Next Generation Needs to Learn
The next generation of radiology training will need to reflect this. Residents who train in programs that treat AI oversight as a core clinical competency, who practice the metacognitive skills that Snoeckx describes, who learn to evaluate a model’s behavior rather than just interpret its output, will enter practice with a skill set that their predecessors did not have and that their specialty will need. Programs that train residents to generate volume and assume the AI handles the rest are producing physicians who are well-suited for the version of the future that no one in this specialty should want to build.
The Doctor’s Doctor role was not taken away. It was gradually deprioritized under conditions that AI is now in a position to change. The specialty that chooses to reclaim it will find that AI, properly deployed and properly governed, is not the reason radiology becomes a commodity. It is the reason radiology does not.
Next: Article 14 examines what a radiology interpretation is worth when AI can generate a report, and whether the economics of imaging look different when judgment, not throughput, is the primary output.
Deploying clinical AI is not the hard part. Knowing whether it is actually working in your environment is.
Performance varies across sites, scanners, patient populations, and workflows. Over time, models drift, disagreement patterns shift, and small failures accumulate in ways that are rarely visible in standard dashboards or validation reports. This is where risk and missed opportunity both live.
Veriloop.health is a vendor-agnostic clinical AI observability layer that measures real-world performance after deployment. We quantify agreement, track error patterns, and monitor trust over time so teams can see where AI is reliable, where it is degrading, and how much workload it can safely support.
We do not replace pre-deployment validation or regulatory clearance. We make sure that what was approved continues to perform safely and effectively in practice.
If you are deploying or scaling clinical AI and need visibility into real-world performance, contact:
References
1. AI in Radiology: Three Keys to Real-World Impact. Philips Future Health Index, 2025. https://www.philips.com/a-w/about/news/archive/features/2025/ai-in-radiology-three-keys-to-real-world-impact.html
2. ECR 2026 CXR Session (Rylands-Monk / Harvey / Snoeckx). AuntMinnie Europe, March 11, 2026. https://www.auntminnieeurope.com/resources/conferences/ecr/2026/article/15819005/ecr-does-ai-toll-the-beginning-of-the-end-for-chest-xray-reporting