Why Unified Platforms are the Foundation for Meaningful AI

Radiology is at a crucial point in its history. Thanks to continued advancements and a rise in preventive imaging programs, utilization continues to grow rapidly, with outpatient volumes projected to increase by roughly 10% over the next decade, and advanced imaging modalities such as CT and PET growing even faster. Medical imaging is no longer a support function (and radiologists are no longer just the “doctor’s doctor.” This surge is driven by earlier disease detection, minimally invasive therapeutics, the prevalence of chronic diseases, precision diagnostics, and more. Yet, the infrastructure supporting these workflows was built for a very different era. The mismatch between demand and infrastructure is becoming economically unsustainable.

This tension set the stage for this panel discussion on unification as the foundation for profitability and meaningful AI deployment, featuring voices from Stanford, Mount Sinai, and Sirona Medical. The central message was consistent; AI can’t deliver value if it’s deployed into fractured workflows.

As moderator Andrew Colbert put it, radiology’s reality today is “real pressure” everywhere you look. Volumes continue to rise, and IT environments are more complex than ever. AI, while highly promising, has often produced more fragmentation than impact.

A Disjointed Experience

Dr. Curt Langlotz walked through how radiology workflow evolved into today’s complex, disjointed system. Systems developed over time as separate stacks, and that pattern continued as speech recognition, worklists, advanced viewers, and AI arrived in separate waves.

The result, he said, is that radiologists now sit in front of “disparate tools from different vendors” that “have grown up separately” and as a result, “our experience is disjointed.”

Even when systems claim to be integrated, the lived experience often tells a different story. Dr. Langlotz shared a familiar workflow failure, “I take an exam off the worklist… I interpret… I dictate… I sign it… and then I go back to the worklist, and it’s still sitting there!”

When asked where radiologists experience the greatest friction, Dr. Langlotz didn’t hesitate. He said, “One I hear about a lot is display protocols… the system doesn’t know where to put anything. I end up spending a lot of time just moving stuff around.” Instead of reading, radiologists often become desktop managers.

He also noted how many reporting and speech interfaces still reflect design choices from decades ago: “We are using user interfaces that were designed a very long time ago.”

Dr. Alex Kagen presented the perspective from within a large and growing health system. His word for the day-to-day reality was simple, “Inconsistency.” He said that with an aging population of baby boomers, and the change of cancer management from a fatal disease to a chronic disease, facilities are going to have more imaging than ever.

He noted that as systems expand and merge, it takes time and budget cycles to harmonize tools, and radiologists bear the brunt of the friction. He said, “That context switching causes cognitive load and burnout.” And, “The more integrations we have, the more downtime risk, and it involves different vendors.”

A Fundamentally Broken Infrastructure

Ken Kaufman framed the infrastructure problem bluntly: radiology’s core platforms weren’t built as modern infrastructure, yet they’re still running the show. He described an industry where thousands of PACS systems remain “in the basement of hospitals,” aging in place, consuming budget and attention just to keep running. “There is a large investment in keeping these old systems alive,” he said. Meanwhile, innovation slows because IT teams are overloaded with operational maintenance. Radiologists frequently have to work across multiple stacks. “40% percent of radiologists today are working on two systems, and in the teleradiology world, it’s 80%, said Ken. ”

Why “Cloud-based” Isn’t Enough: the Difference Between Hosted Legacy and Cloud-Native

The panel challenged the common industry claim that radiology is already “in the cloud.” Kaufman argued that much cloud messaging is essentially legacy systems hosted elsewhere. “You can store images in the cloud, but that doesn’t solve the fact that what you have are very fragile systems,” he said.

He drew a sharp line between running legacy systems in virtual machines and building a true multi-tenant, cloud-native platform. In modern architecture, shipping new functionality is instant. “When we launched it (Sirona), you pressed a refresh of your web browser. There is no software on the workstation,” he said. That ability to deploy quickly becomes critical as AI evolves weekly.

AI’s Trajectory

Dr. Langlotz provided a clear history of the trajectory of radiology AI. The first wave, he said, focused on what was easiest to bring to market: detection and triage. Now, radiology leads the broader medical AI market, “More than three-quarters of FDA-cleared AI products today are radiology-focused,” he said. Yet, adoption hasn’t matched the hype, in part because many tools create additional work. Dr. Langlotz said, “It’s like, ‘hey, that’s kind of interesting, kind of cool, it might help me, but it might slow me down.”

The panel anticipates radiology entering a second wave of tools that reduce burden, draft content, and automate multi-step tasks. They discussed examples such as report drafting and impression suggestions, multi-agent workflow orchestration, and the automation of tasks such as notifications and follow-up.

Dr. Langlotz shared a striking projection: with these tools, radiologists could find a sustainable way to handle rising demand. He said, “It will help us with the incredible amount of work, the burnout, and improve our efficiency.”

What Meaningful AI Requires

The panel’s definition of meaningful AI was multimodal intelligence—systems that can reason across images, priors, reports, clinical history, and workflow state.

Mark Longo demonstrated what it looks like when AI operates inside Sirona’s unified platform. He described unification at three levels:

• architecture: cloud-native, multi-tenant
• application layer: worklist, viewer, reporting in one connected system
• data layer: standardized concepts through an ontology so the system can reliably interpret messy inputs

Then he introduced an “AI sandbox”/agent concept: “You can think of it as a real-time AI co-pilot,” he said. He highlighted examples such as reviewing AI outputs and incorporating relevant findings into the report, interpreting embedded images/annotations, and commenting on findings.

In the closing discussion, the panel aligned around a future that’s less chat with AI and instead more “AI does the work automatically.” Ken Kaufman summed it up, “Don’t even ask for it—just have it done.”

The panel’s conclusion was simple: agentic AI will only be as powerful as the underlying infrastructure. In radiology, meaningful AI requires a unified workflow, unified data, and a platform that can deliver change instantly and safely at scale.