My takeaways from RSNA 2025: Personalization, interoperability and the rise of agentic AI

RSNA 2025 kicked off in true Chicago fashion—with a snowstorm. Despite whiteout conditions across the Midwest, which left thousands scrambling due to travel delays, RSNA still welcomed 50,000 attendees and 650 exhibitors from around the globe.

This year’s theme was “Imaging the Individual”, as reinforced in President Umar Mahmood’s opening address. While his remarks focused on the clinical side of precision imaging, they set the tone for something broader:

Personalization in medicine can only scale when the operational systems behind it evolve and keep pace.

And that’s what made this year’s RSNA feel different. There was a noticeable shift in various plenaries, educational sessions and show floor conversations that agentic AI and intelligent automation aren’t coming to healthcare “someday”. They are here now, reshaping how radiology teams work, how systems communicate and how patients experience care.

Here are the takeaways that made it clear that the next era of radiology isn’t only about better imaging technology—it’s about investing in the operational backbone that makes better care possible.

Takeaway 1: AI’s most transformative impact in radiology isn’t just clinical—it’s operational

While many sessions still focused on the emerging clinical applications of AI to improve clinician productivity, well-being and diagnostic accuracy, the most dramatic shift was the emphasis on using AI for the non-clinical operations.

From interactive customer panels with URMC and St. Luke’s to AI theater presentations like Radiology Reimagined: AI, Innovation and Interoperability in Practice, speakers emphasized the need for AI to streamline operational workflows across departments, systems and care settings.

And it’s no wonder: In North America alone, labor costs make up more than 50% of a hospital’s expenses, with 57% of healthcare leaders reporting that administrative tasks, like time and attendance processes, remain manual. Data entry errors can further compound the issue, requiring even more manual work to fix.

When AI automates non-clinical tasks that slow radiology teams down, it not only reduces burnout and administrative overhead but also enables more resources to focus on what actually moves patient outcomes—personalized clinical care at every touchpoint. Contrary to popular belief, AI isn’t replacing human connection; it’s helping protect it.

Takeaway 2: Agentic AI is the next generation of interoperability

This shift toward AI isn’t just a technical evolution—it’s a practical one, rooted in the everyday realities radiology teams face. As demonstrated in the RSNA Associated Sciences Consortium session, the time has come for imaging departments to seriously evaluate how they can future-proof with AI.

Today, countless clinical and non-clinical steps are held together manually by staff—toggling between portals, reentering data and chasing down info in other systems. While standards help, variability at the edges persists, leaving staff to fill gaps when APIs fall short.

This is precisely where agentic AI steps in.

Unlike traditional automation, agentic AI doesn’t wait for the perfect integration. As I explain in a recent op-ed with Aunt Minnie, agentic AI moves across systems the same way a staff member would—using agents to navigate unstructured inputs (e.g., fax, portal, or voice), complete complex tasks and apply decision logic safely and consistently. The RSNA Deep Learning Lab echoed the importance of agentic AI in its educational session on using these multi-agents to better support radiology workflows.

In my RSNA AI theater talk, Scaling Radiology’s Hidden ROI Through Workflow Automation, I also highlight how agentic AI uses coordinated agents to convert manual, non-clinical workflows into scalable, automated processes—and its practical and immediate impact. Not only does it relieve staff burden, but it also drives measurable ROI—helping health systems cut costs, improve access and refocus human energy back on the patient. And when teams spend less time clicking and more time with patients, care becomes more human.

Takeaway 3: AI safety and governance are non-negotiable

Some of the most forward-thinking conversations at RSNA centered on the ethics of AI, as demonstrated by techno-sociologist Zeynep Tufekci’s talk: Everyone is Having the Wrong Nightmares: AI’s True Threats.

A leading voice on the intersection of technology, society and science, she offered a more pragmatic view of AI and—when left unchecked—how it could negatively affect public health, from patient surveillance to data privacy to care discrimination. She reinforced the crucial importance of ethical oversight and governance of AI, especially at scale across large populations.

Because of these risks, strict governance policies and safety protocols in data-driven medicine are crucial. It’s up to human teams to set the guardrails that keep AI behavior in check, as explored in the educational session: Best Practices for Human-AI Collaboration.

As agentic AI takes on more of the administrative load, having these guardrails in place becomes even more critical—because patients and care teams need to trust that automation is improving the care experience, not undermining it.

This is why role-based access control (RBAC), policies-as-code, auditable trails and clear human-in-the-loop escalation paths aren’t optional—they’re what make AI reviewable, reversible and safe at scale.

The future of AI is here

One pressure continues to define radiology: Administrative costs are rising faster than operating budgets can absorb. If this trend continues, the clinical experience itself—the part of healthcare that matters most—is at risk.

Solutions like PocketHealth’s Conductor show how agentic AI can help mitigate this, especially when applied to the right problem at the right time with the right guardrails. Entire non-clinical workflows are automated end-to-end, freeing up staff from manual tasks and restoring the capacity needed to deliver more connected, personalized patient experiences.

This is why RSNA felt very timely this year. The consensus wasn’t “AI will transform radiology someday.” It already is—and teams that modernize their operations now will be the ones ready to focus on the next era of personalized care.

Radiology doesn’t need more point solutions. It requires a smarter operational backbone—one that scales, adapts and safely takes work off the plates of the people who enable care. RSNA 2025 made it clear that the future of radiology isn’t just more advanced imaging. It’s more intelligent operations. And that future has already begun.