Nina Kottler, MD
Radiology AI is entering a new era.
For years, innovation has largely focused on individual findings and isolated tasks. Now, advances in foundation models and intelligent workflow integration are creating opportunities to support more comprehensive image interpretation, streamline clinical workflow and expand the value radiology delivers.
Recently, we spoke with Dr. Nina Kottler, chief medical AI officer at Mosaic Clinical Technologies™ , about this evolution, the importance of bringing AI into clinical practice responsibly, and why she believes the future of radiology is not AI replacing radiologists, but radiologists and AI achieving more together.
What is the biggest shift underway in radiology AI, and why are foundation models so important?
The biggest shift is moving from narrow AI to comprehensive AI. For years, radiology AI has largely been built to answer one question at a time: Is there intracranial hemorrhage? A PE? A lung nodule?
Those tools can be useful, but radiologists don’t evaluate an exam one finding at a time. We assess the entire study, identify and characterize abnormalities, integrate clinical context and synthesize that information into a meaningful report.
Foundation models offer an opportunity to support this broader clinical process. Instead of developing, implementing and maintaining a separate model for every finding, we can build models with broader capabilities that can be adapted to multiple tasks.
Advances in large language models and vision-language models are particularly exciting because they create opportunities to combine image interpretation, clinical information and report generation. These comprehensive capabilities of foundation models support how radiologists actually work.
You describe AI maturity using two dimensions. What are they?
The first is model capability. This dimension represents how broad and clinically meaningful the tasks AI can perform. We are moving from highly specialized, narrow models towards models that can support increasingly complex and comprehensive functions.
The second is system capability: What can AI actually do in the clinical environment? The AI’s capability depends on access to relevant data, integration into clinical workflows, and the ability to take appropriate actions. A great AI model is not enough. If the output does not reach the right person, at the right time, in the right workflow, its clinical impact will be limited.
The greatest opportunity comes when both dimensions advance together, allowing increasingly capable models to be integrated into systems that meaningfully support clinical care.
Why does system integration matter, and how do AI agents advance these capabilities?
Healthcare has no shortage of technology. What we often lack is the ability to bring information and capabilities together in ways that meaningfully support clinical care. Too often, clinicians become the integration layer, navigating between disconnected systems, gathering information and coordinating next steps.
AI creates an opportunity to change that dynamic by embedding intelligence directly into clinical workflows. AI agents represent an important evolution in system capabilities, enabling AI to move beyond generating information to taking action.
An AI agent is not simply a smarter AI model. It is a system that can work toward a defined goal by accessing information, using tools, and taking actions. Unlike traditional AI models that primarily generate a response to a prompt, agents can execute a series of steps, make decisions and adapt their actions base on new information.
In radiology, an AI agent might gather relevant clinical history, retrieve prior imaging, coordinate patient follow-up or manage workflow tasks. Importantly, the degree of autonomy should be appropriate to the task and its clinical risks, with corresponding safeguards and oversight.
How does Mosaic fit into this vision?
Mosaic was built around the idea that radiology needs more than a collection of AI tools. We need an integrated platform that brings together clinical data, imaging, reporting, workflow, AI capabilities and clinical expertise. That is why we developed MosaicOS as a cloud native operating system for radiology. By connecting these capabilities within a common, modern platform, we can incorporate AI directly into how radiologists work rather than requiring them to navigate disconnected applications.
Capabilities such as Mosaic Reporting, Mosaic Capture and Mosaic Drafting* reflect this approach. But the larger opportunity is creating an intelligent clinical environment that can evolve as AI capabilities advance, including more comprehensive models and AI agents.
What has Mosaic Drafting shown so far?
Mosaic Drafting represents a shift from AI that identifies isolated findings toward AI that can generate a more comprehensive initial interpretation of an examination. Rather than evaluating one or a few individual abnormalities, Mosaic Drafting assesses the entire examination, identifying and characterizing multiple findings to generate a preliminary report for the radiologist to review, edit and finalize.
This more comprehensive approach has also demonstrated improvements in radiologist detection of clinically significant findings. For example, in early evaluation of CT abdomen and pelvis drafting, we identified improved detection of true positive findings across every condition studied, including appendicitis, pneumoperitoneum, diverticulitis, pancreatic masses and acute fractures.
But what matters most isn’t just how well the AI performs independently. It’s whether the radiologist and AI together deliver better clinical performance. That’s why our evaluations focus not only on the quality of the AI-generated draft, but also on how it affects radiologist accuracy, consistency and efficiency. Ultimately, our goal is to expand what radiologists can accomplish while maintaining clinical oversight and accountability.
Are radiologists at risk of being replaced by AI?
AI will undoubtedly automate more of the tasks radiologists perform today. But automating tasks is not the same as replacing an entire profession. Radiologists do more than identify findings. We synthesize information, apply clinical judgment, communicate with referring clinicians and help guide patient care. As AI becomes capable of performing more of these functions, the role of the radiologist will also evolve.
I believe the future is radiologists plus AI, combining the strengths of both to improve patient care. But radiologists need to actively shape that future. We understand the clinical challenges, the consequences of diagnostic errors and the needs of the patients and clinicians who depend on our work. Our role should extend beyond simply using AI tools. We need to help determine how they are developed, evaluated and integrated into clinical practice. We are at an inflection point, and radiologists have an opportunity to lead this transformation rather than simply respond to it.
You distinguish between “efficiency AI” and “opportunity AI.” What is the difference?
Efficiency AI helps us do what we already do, but better. It can improve accuracy and consistency, streamline reporting, reduce administrative burden and help radiologists manage increasing imaging demand.
Opportunity AI is where things become truly transformational. It allows us to do things we couldn’t previously do or couldn’t do at scale. For example, medical imaging contains a tremendous amount of information that we don’t routinely extract today. AI creates opportunities to identify patients at increased risk of disease, support opportunistic screening and offer new capabilities such as a non-invasive imaging based or “digital” biopsy.
Instead of simply using AI to improve existing workflows, opportunity AI allows us to rethink what radiology can contribute to disease prevention, diagnosis and personalized care. Efficiency AI helps us deliver today’s care more effectively. Opportunity AI has the potential to redefine tomorrow’s care.
What should hospitals and health systems prioritize as they adopt AI?
Health systems are facing increasing imaging demand, workforce constraints and growing expectations for timely, high-quality care. AI can help address these pressures, but successful adoption requires more than purchasing individual AI tools. Organizations need to think about how AI fits into their broader clinical and technology strategy.
Equally important is clinical governance. We need to evaluate not only how the AI performs independently, but how clinicians and AI perform together. That assessment requires validation before deployment, continuous monitoring of both AI performance and the human-AI system in clinical practice, and clear processes for addressing concerns when they arise.
Ultimately, the organizations that realize the greatest value from AI won’t necessarily be those with the most models. They’ll be the ones that can integrate AI into clinical care, demonstrate meaningful outcomes and govern it effectively over time.
What is the ultimate vision?
The ultimate vision is to elevate radiology’s role in healthcare. AI creates opportunities to turn more of the information contained in medical imaging into meaningful clinical insights, helping us identify disease earlier, better understand individual patient risk and support more personalized care. At the same time, we need AI to help address today’s challenges, improving efficiency, consistency and access to high-quality imaging interpretations.
But we shouldn’t stop at doing today’s work faster. I want us to rethink what’s possible when radiologists have access to capabilities that extend what we can see, understand and contribute to patient care. That’s what excites me about the future of radiologists plus AI. It’s not simply about delivering more of the same. It’s about expanding what radiology can do for patients.
*CAUTION: Investigational device. Limited by United States law to investigational use.
Mosaic Clinical Technologies is the technology services division of Radiology Partners (RP), powering the future of radiology through MosaicOS—a proprietary imaging intelligence platform designed to meet the specialty’s most pressing challenges. A fully cloud-native and AI-native operating system, MosaicOS is where innovation meets impact, seamlessly integrating diagnostic technologies, AI-powered tools and smart workflows into a single scalable solution. Mosaic Clinical Technologies™ supports RP’s national network of affiliated practices and commercial partners across the imaging landscape, redefining what is possible in enterprise imaging. Connect with us on LinkedIn and YouTube. Contact us at info@MosaicClinical.ai.
