Nvidia sees major shift in radiology to AI agents and new autonomous imaging systems

Agentic AI was a major shift at RSNA

Early adoption of AI in radiology focused heavily on computer vision and image perception, but Powell said the industry is now moving into agentic AI. She described it as a software paradigm shift, where instead of AI being used to help humans do their jobs, these AI agents can do things autonomously for humans, or oversee specific functions or entire informatics systems without the need for a human interface.

That shift comes as healthcare systems face mounting pressure from staffing shortages and rising imaging demand. Powell said agentic AI has the potential to offload clerical and administrative burdens and even assist with report preparation.

“We need to close that gap by adding additional AI—an agentic AI—that can relieve these amazing doctors who dedicated their lives to saving our lives,” she said.

Powell pointed to the broader cultural shift in AI adoption following the launch of generative AI tools like ChatGPT, which has made millions of people much more comfortable using AI to help them do their jobs on a daily basis.

“This was a moment in time when any human, no matter how young or old, specialized, or tech savvy, became an AI user,” she said. 

Powell added that healthcare professionals are now increasingly comfortable engaging with AI systems embedded directly into their workflows. This includes direct transcription of radiologist dictation that AI structures and places in the correct field in radiology reports. Another example is ambient AI that records physician and patient visit discussions and then creates a full patient visit report the doctor just needs to edit. She emphasized these types of successful medical AI are seeing adoption because the AI feels seamless and supportive.

“AI cannot be a burden like previous generations of technology and software, it has to be a helper out of the box,” Powell said. “It has to feel seamlessly integrated into workflows. It has to shave time off the work that they do, and ultimately it has to provide better clinical outcomes.”

GE Healthcare has launched numerous products that integrate Nvidia AI

One of the most visible examples of Nvidia’s role at RSNA was its expanded collaboration with GE HealthCare. At the meeting, GE HealthCare showcased multiple new imaging technologies built with AI as a core component, including work-in-progress systems such as a photon-counting CT scanner, a new 3T MRI and a 1.5T very low helium, sealed MRI system. In total, Powell said GE HealthCare introduced nine major new products, most leveraging Nvidia technology.

“They had nine new products that were huge introductions. Eight of them integrating Nvidia technology,” she said.

According to Powell, the pace of innovation in imaging hardware is accelerating as AI enables continuous upgrades throughout a scanner’s lifecycle, rather than multiyear revision cycles.

Physical AI will be the next major shift in radiology

Powell highlighted what Nvidia sees as the next major frontier with physical AI. It is physical in that the AI will no longer be limited to a backend software system and will be able to integrate robotics and avatars to interface directly with patients to perform imaging exams.

“It's actually the next generation of AI that we call physical AI,” she said. This concept centers on autonomous, robotic imaging systems capable of acquiring images with minimal human intervention.

Powell said physical AI could address one of the most pressing global challenges in medical imaging access. She noted the massive shortage of imaging systems and technologists to operate them in developing countries, rural areas and lower-income underserved areas. She said even in major metropolitan areas, patients often need to wait months for routine imaging exams.

“Only about a third of the world's population has access to this critical technology,” Powell said, noting that both image acquisition and interpretation currently depend heavily on trained professionals. “What if someday all of these medical devices and imaging instruments could become robotic, more and more autonomous, just like our cars are becoming more and more autonomous.

She described a future in which autonomous imaging systems could be deployed in underserved and rural areas. She outlined a scenario where patients could enter an imaging room, interact with a digital agent for intake, and be guided through an exam without the need for on-site specialists.

This vision aligns with GE HealthCare and Nvidia previously announced plans to co-develop autonomous X-ray and ultrasound systems in March 2025. At RSNA 2025, GE HealthCare gathered feedback from radiologists through early-stage demonstrations of these concepts, which aim to reduce wait times and extend imaging services to areas with limited staffing.

Powell said growing clinician shortages are accelerating acceptance of autonomous systems. She noted her own experience as a patient and being frustrated that she needs to plan imaging exams months in advance because there is just not enough capacity in the health system.

“I live in California where it's one of the richest areas for healthcare services and providers. And I oftentimes am waiting longer than six months to get my routine imaging,” Powell explained.

She added that autonomous AI, digital agents and robotic systems could help close the widening gap between demand and capacity.

“The more we can close that gap through autonomous systems with the digital agents giving us a better experience, and then the physical agents being able to actually deliver some of these services, all the way into robotic surgery, this is where we're going to see this next chapter of medicine be written,” Powell said. 

Will radiologists be replaced by AI? Read more in the article CEO of America’s largest public hospital system says he’s ready to replace radiologists with AI.