The future of AI in radiology is the AI you barely notice

Whenever I speak with health systems about the latest AI innovations in radiology, I hear the same question time and time again. It’s no longer “Will this help us?” or even “Will it be worth the investment?” Instead, they now ask: “How fast can you implement it?”

It’s a dramatic shift in attitude toward AI compared to just a few years ago. What was once met with healthy skepticism is now fueled by a sense of urgency. There’s a shared sense we must innovate faster to meet the sheer volume of imaging demand radiology departments are facing. Gone are the days of debating whether AI will replace radiologists. Today, there’s growing recognition that augmenting the human expertise of radiologists with AI is the only way to keep pace with ever-rising volumes of medical imaging studies.

Our 2025 Future Health Index underscores this shift. Most radiologists (85%) are optimistic that AI technologies could improve patient outcomes, even more so than healthcare professionals in general (78%). In fact, they see many areas where AI could make a difference in their department – automating repetitive tasks, expanding their capacity to serve more patients, and helping improve triage, wait times, and patient throughput. 

Yet despite this enthusiasm, one major hurdle remains: integrating AI seamlessly into clinical workflows. Many of the tools and algorithms are already there. But getting them to truly work for radiologists – in the complex, dynamic, and sometimes messy world of clinical practice – is another challenge, and one that’s critical to solve.

Making radiology AI work in clinical practice

The message I’m hearing from radiologists is clear: they want AI that works quietly and unobtrusively in the background, built into their existing systems. What they don’t want is yet another application to manage. Imagine a pilot’s cockpit, where all the instruments are visible in one pane of glass, working together in real time. That’s how radiologists want to work as well: everything in one view, seamlessly connected, without having to constantly shift attention.

But that vision is still far from reality in many radiology departments. Our Future Health Index shows that 60% of radiologists already spend too many clicks just to access the patient information they need, with 1 in 3 losing more than an hour(!) during a typical shift. We also know from research that when AI is bolted on without considering workflow integration, it can actually increase workload instead of reducing it [1].

The good news? When we design AI with radiologists – embedding it directly into the systems they already use – it becomes a natural extension of their workflows. The best AI feels like a co-traveler in the interpretive process, quietly supporting radiologists within their existing systems. You don’t have to think about it. It’s just there, working alongside you.

Turning time savings into better patient care

We are already seeing the value of seamless workflow integration in action through our collaboration with the Hospices Civils de Lyon in France. As one radiologist noted, every minute of drawing measurement circles on images is a minute that cannot be spent focusing on interpretation or interacting with patients. Together, we developed AI solutions that drastically reduced the time their team spends on image contouring – all within their existing visualization workspaces, so there’s no need to juggle applications.

The same principle applies at Fundação Instituto de Pesquisa e Estudo de Diagnóstico por Imagem in Brazil, where radiologists can access AI-generated insights directly within their diagnostic viewer. This integration improves efficiency by automatically prioritizing abnormal cases and reducing turnaround time. The time savings are significant, translating directly into better patient care. For example, with AI now supporting chest X-ray processes, clinicians receive results in as little as two minutes. This enables them to make faster, more confident decisions to help ensure patients get the care they need.

These are both examples of how designing innovations with radiology departments – not just for them – can turn AI into a genuine time-saver in daily clinical practice. And with the rise of agentic AI, I’m convinced we are only scratching the surface of what’s possible.

Looking ahead: the age of agentic AI in radiology

The first wave of AI in radiology focused on the interpretive moment: helping radiologists read images faster and more accurately, with algorithms trained to spot fractures, bleeds, lesions, and so on. These AI algorithms, often powered by deep learning, will continue to evolve and further improve diagnostic precision.

But the interpretive moment represents only a small portion of what radiologists do. Much of their day is spent managing everything that happens before and after: prepping the case, gathering prior studies, prioritizing worklists, reviewing reports, and making follow-up calls. It’s those thousand little tasks that chip away at their time and focus. When we start to reimagine the entire diagnostic imaging workflow through the lens of AI, even micro-gains in efficiency can compound into department-wide improvements in productivity.

That’s where I see the next wave of opportunity: agentic AI that can help take on much of the pre- and post-interpretive work, allowing radiologists to focus on what they were trained for: the interpretive moment. Because that’s what many radiologists want most: technology that lifts the burden of the work they were never meant to do.

Over time, these agents will learn from user preferences to optimize protocols and orchestrate workflows in ways that support the radiologist – anticipating what’s needed next, surfacing the right information at the right moment, and guiding their attention to the most urgent cases or findings.