Medical device industry continues to turn to AI

SAN DIEGO — Medtech executives are incorporating artificial intelligence into medical devices, but they are also wading into using the technology for administrative tasks. 

At AdvaMed’s The MedTech Conference, company leaders said they expect to incorporate AI into more of their devices. With the recent rise of generative AI, companies are also considering the technology for internal tools, but progress has been slower.

Here’s how four medtech executives are using AI today: 

Stryker CEO Kevin Lobo

Lobo said every one of Stryker’s businesses has a digital or AI component. He also expects that Stryker will have a digital division in the future, following the company’s acquisitions of Vocera Communications and Care.ai.

“Five years ago, I would have never thought I’d be buying an AI company,” Lobo said.

With Vocera, Stryker has a technology that provides badge alerts to healthcare workers based on what’s happening in a patient’s room. For example, if a patient is lowering a guardrail to leave their bed, it can send an alert to prevent falls.

Stryker is also developing AI technologies in surgery. The company has a system, called Blueprint, that advises surgeons on the type of implant they should use in an operation and provides a surgery plan.

Projects that use innovation to solve a problem for Stryker’s customers are easy to prioritize, Lobo said, adding that teams “have no trouble finding the money” for these projects.

Figuring out how to use AI for internal productivity is more difficult.

“Frankly, we’re a little bit behind where I’d like to be,” he said, noting that the company is focusing on areas such as customer service and regulatory work. 

Insulet CEO Ashley McEvoy

McEvoy, who became CEO of Insulet in April 2025, said the diabetes technology company has used OpenAI’s foundation models for customer care and retention. Insulet uses the technology to provide customer care agents access to company knowledge, which can help with call wait times. 

Looking to the future, McEvoy hopes to use Insulet’s data in a precision medicine play to build more personalized algorithms for the company’s insulin pump users. 

From a culture perspective, McEvoy said she is working with company leaders to “embrace the good parts of [AI] and not be naive to some of the darker sides,” but still have the freedom to experiment.

Hologic CEO Stephen MacMillan

MacMillan said AI has been helpful with the rise of 3D mammography, which creates more images than 2D mammograms. Looking through so many images can be fatiguing for radiologists. The company uses machine learning to distill the process down to “go look at slide 3 in the upper left quadrant instead of looking at hundreds of slides,” MacMillan said. 

The company has started using AI in its breast imaging segment, but plans to roll it out to the rest of its businesses. For example, Hologic has developed a newer tool to support pap smears, which are still often done by cytologists looking at individual slides through microscopes. The company has digitized the process and uses machine learning to flag objects of interest for cervical cancer. Hologic received the FDA’s de novo authorization for the digital cytology system in 2024. 

Olympus Chief Strategy Officer Gabriela Kaynor

Four years ago, Olympus started focusing on investing in its data and AI capabilities, but more geared toward the company’s medtech products, Kaynor said. That has culminated in a recent product launch for AI to support colorectal polyp detection, as part of a broader platform to use AI in endoscopy. 

In the past two years, Kaynor said, Olympus has explored AI more for enterprise productivity. For example, Olympus built an internal AI tool to help with language translation, which can be a friction point. 

Olympus is headquartered in Tokyo, and work is done in Japanese, English and German. Being able to translate data — which might be sensitive — without it leaving the company was a good use case for the technology, Kaynor said. Olympus has been piloting the tool, and the company is getting ready to launch it internally. 

“This has been something that I think has reached, really, every corner of our company,” Kaynor said.

As more medical device companies incorporate artificial intelligence, chip designer Nvidia is playing a central role. The company has struck partnerships with top medtech firms including Medtronic, Johnson & Johnson, GE HealthCare and Philips.

The uses span a breadth of technologies. In imaging, Nvidia has partnered with GE HealthCare around autonomous X-ray and ultrasound solutions, and is working with Philips to develop foundation models, a type of model that can be used for a wide variety of tasks, for MRI machines. In robotics, J&J’s Monarch platform for bronchoscopy uses Nvidia’s computing platform. Nvidia also debuted a new developer framework for healthcare robotics in March 2025, called Isaac for Healthcare. The system features three computers to generate synthetic data to simulate workflows, create virtual environments where robots can safely learn skills, and a platform to deploy applications and for real-time sensor processing. 

MedTech Dive spoke with David Niewolny, director of business development for healthcare and medical at Nvidia, about the company’s partnerships and the future of AI in medical devices.

This interview has been edited for length and clarity.

MEDTECH DIVE: What’s the current state of AI in medtech? 

DAVID NIEWOLNY: In the last 18 months, we’ve seen this really fast progression in terms of healthcare and medtech adopting generative AI. That’s starting the idea of creating — drafting clinical notes, generating synthetic data for training. Devices are now looking at how can they begin bringing agentic AI into these applications. You’re seeing digital agents being an assistant to the healthcare provider or the patient with automating workflows, driving more context-aware support for clinicians. 

Then you get to where we see the future. Where a lot of the AI and innovation is happening is in the idea of physical AI. And that brings us into this role of robotics. The easy one to think of is surgical AI. But there’s a huge number of applications. One specific use case that I talked about in Taiwan [at GTC Taipei] is with some of these more operational robotics. Think of a nurse assistant in terms of delivering medication, bringing different supplies around a hospital, making sure different areas of the hospital are stocked.

You recently announced a partnership with GE HealthCare around autonomous imaging. How does that work? 

That was around completely transforming the way that you would look at doing medical imaging in the future. We took two initial use cases, one being the idea of an autonomous X-ray. Think of a future world where you no longer have the X-ray tech. You now have a digital agent that's essentially checking you in. You walk into a room where there’s another robot, it could be a digital agent, that’s providing you with all of the guidance for where to stand, when to hold your breath and how to position yourself. You stand in one spot, and then the actual machine positions itself. 

You can also look at some of the generative AI applications, where it can hand a doctor a full report on everything that it saw in terms of its clinical findings.

In that particular case, now you’re expanding the access to care because you essentially have these fully autonomous systems that are doing medical imaging. 

X-ray is one that we announced, and the other one is around ultrasound. In each one of these cases, GE HealthCare is working with us and collaborating on a methodology and tools to build these robotic systems. 

What opportunities does Nvidia see in AI in medical devices?

Everything is about building an ecosystem. You look at all of the great work that we can do from accelerating a lot of these applications with AI and now robotics, going from Nvidia direct to a healthcare provider, there’s just too big of a gap. They don’t have the developers in-house to begin building this. 

So, then you work backwards in that ecosystem, and you realize it’s the medtech companies that are building out all of these devices and solutions. What we’re looking at doing is, how do we bring all of those components of the ecosystem together, building on a common platform? Computers were not mainstream until Windows opened the door for this huge influx of software. 

A lot of this learning we took from another industry: automotive. We essentially needed to create a fully simulated environment, train on that simulated data, take those algorithms and move them down to the edge.

We took those learnings and said, “What other areas are ripe for disruption?” Medtech, specifically, has the most to benefit from this opportunity of having a common platform. But at the same time, it had a big hurdle to jump over. There wasn't any single company in that ecosystem, as we saw it, that could essentially build out that platform.

I’m hearing more people talk about Agentic AI, a type of AI that can perform autonomous tasks. Why are medtech companies interested in this technology?

Agentic AI has a whole number of applications. A partner of ours, Abridge, leveraged a lot of our technology to do clinical documentation. They're integrating with major EHRs, and they're continuing to get more and more hospital users. 

You also have some of these agents actually working right alongside a surgeon, where now you have an assistant in the room, where you can begin asking questions, pulling up the patient's medical records, adjusting some of the devices in the room. 

One of our partners, Moon Surgical, actually downloaded their entire instructions for use manual into an agent that a doctor or surgeon can reference for things like setup. Instead of referencing a 1,000-page manual, you can just ask the robot where to be set up, how to be set up and what are the best practices?

Have you faced concerns from clinicians about AI taking their jobs?

Yes, people do get concerned. The key piece is, in almost every one of these cases, we’re augmenting the team members that are already there. There's a shortage in place. This is actually improving care, as opposed to the narrative around taking people's jobs. We take a lot of those workloads that the staff sees as either busy work or mundane, and augment them.

There's always going to be a surgeon involved here, but we can do sub-task automation and actually make some of those tasks easier for a surgeon.

Sponsored content

By RQM+

Risk-based monitoring (RBM) was designed for a different era of clinical trials. The model of periodic site visits, manual source data verification, and centralized trend analysis assumed that the most important information would arrive in batches, to be reviewed after the fact. That assumption no longer holds.

Connected devices, remote sensors, and decentralized trial designs have changed the data environment. Sponsors now receive continuous streams of real-world performance data instead of periodic snapshots. The question facing clinical operations executives is whether their RBM infrastructure is capable of using it.

From snapshots to continuous streams

The shift matters because the risk profile of modern trials has changed in kind, not just degree. In hybrid and decentralized clinical trial (DCT) designs, risk moves away from site-level administration and toward patient device usage and remote data submissions. Fixed monitoring schedules designed around pre-set site contacts cannot adequately address the deviations that show up in real-time data between visits.

Real-time access allows sponsors and CRAs to detect irregular patterns like missed submissions, incorrect device usage, and anomalous sensor readings as they occur, allowing them to direct intervention resources precisely where they are needed. Under both FDA guidance and EU MDR's increasingly stringent post-market clinical follow-up (PMCF) requirements, that kind of continuous, adaptive oversight is no longer optional.¹

AI as a force multiplier — with limits

AI and machine learning tools offer the scalable oversight necessary to process high-volume continuous data streams and flag signals that would be normally be lost in manual review. Analysis suggests that AI-enabled monitoring has strong potential to redirect resources toward high-performing centers and accelerate site-level decision-making.² However, AI can make mistakes (e.g., false positives), so the most defensible approach is human-guided automation: AI detects irregularities and potential safety issues, but no action is taken without clinical expert confirmation.

Governance is the real constraint

For most organizations, the barrier to modernizing RBM is governance. AI tools require high-quality, structured data to function reliably. Teams that lack clear protocols for human oversight, defined accountability at the investigator level, and infrastructure to handle continuous data streams will find that adding technology compounds the problem rather than solving it.

Before live deployment, leading sponsors use three approaches to stress-test their frameworks without exposing sensitive patient data:³

• Synthetic data modeling — simulated datasets allow teams to test monitoring strategies and filtering algorithms before live integration
• Data filtering — separating signal from noise in real time ensures monitoring efforts focus on meaningful deviations, not data volume
• In silico optimization — computer modeling predicts anatomical responses and adverse event timing, helping sponsors refine inclusion criteria

The executive takeaway

RBM is no longer a clinical operations function. It is a core component of life cycle management, PMCF strategy, and regulatory compliance. Organizations not treating it as such are carrying risk they may not see until it reveals itself in a regulatory finding or a trial that cannot generate defensible evidence.

Modernizing RBM means addressing four organizational realities:

• Study oversight has evolved — your infrastructure needs to handle modern hybrid and decentralized models, not just periodic site visits
• Manual review is a signal for change — extra manual steps or bypassed remote data opportunities suggest your data management strategy needs a refresh
• Governance is nonnegotiable — AI and ML drive efficiency, but clinical context and data quality are the investigators' responsibility
• Partnership matters more than software — you need partners who understand MedTech device complexity, not generic platforms built for pharma

References

¹ FDA Guidance. (2024). Digital Health Technologies for Remote Data Acquisition in Clinical Investigations.
² Applied Clinical Trials. (2025). Modernizing Clinical Oversight: The Shift to Adaptive Monitoring.
³ Journal of Clinical Innovation. (2025). The Role of In Silico Modeling in Modern Trial Design.

The Food and Drug Administration released two final guidance documents in January 2026 that would loosen regulations for certain types of wellness and software products. 

FDA Commissioner Marty Makary announced the changes at the Consumer Electronics Showcase. Makary, in a video posted on X, said the changes would “promote more innovation with AI in medical devices.”

Wellness exemptions for blood pressure, blood glucose

The first guidance clarifies the FDA’s thinking on what constitutes a wellness device. It offers broader leeway to wearables that provide readings around heart rate, blood pressure and blood glucose, so long as they are intended solely for wellness purposes.

In examples provided in the guidance, the FDA said a wrist-worn wearable that tracks metrics including sleep, pulse rate and blood pressure would fall under a general wellness claim, provided the product has validated values for blood pressure.

The guidance appears to contradict a warning letter the FDA sent to wearable company Whoop in 2025 for rolling out a blood pressure feature without authorization. At the time, the agency said that blood pressure is inherently related to a medical diagnosis.

A spokesperson for Whoop applauded the new guidance in an emailed statement, saying it clarifies that Whoop can provide blood pressure insights and wellness metrics when designed for non-medical purposes. The spokesperson said the changes should “help resolve long-standing uncertainty about the boundary between providing wellness insights and medical diagnosis and treatment.”

Another example was for a wearable intended to provide blood glucose estimates to monitor nutritional impacts, and is explicitly contraindicated for use by people with diabetes and pre-diabetes. According to the guidance, this would fall under a general wellness claim, but would not count as a low-risk product if the device used microneedle technology to provide the estimates. 

In 2024, the FDA warned consumers not to use smart watches or smart rings that claim to measure blood sugar without piercing the skin. 

“In the past, the FDA has permitted some measurements like pulse rate and O2 saturation in some wellness products, but not others,” Makary said in a speech at CES. “It didn’t always make a lot of sense from the outside.”

Tom Hale, CEO of Oura, maker of a wearable ring that is working on a blood pressure feature, welcomed the changes in a LinkedIn post.

“As wearables continue to evolve, modern regulation that recognizes the difference between early awareness and medical diagnosis is critical,” Hale wrote.

Less regulation of clinical decision support software

In a separate guidance, the FDA unveiled significant changes to how it regulates clinical decision support tools. The biggest change is to a section describing how the FDA interprets whether software is providing recommendations to healthcare providers. Software that provides a sole medical recommendation can now be exempt from regulation. Under a previous guidance, it would have been considered a medical device. 

For example, according to the new guidance, software that predicts a patient’s risk of future cardiovascular events based on their weight, smoking status, blood pressure and lab tests would be exempt from FDA enforcement. However, if the same test was intended to predict the risk within 24 hours, or used genomic data, it would be considered a medical device.

Another example considers the use of software to summarize radiologists’ findings, a concept that firms are testing using generative artificial intelligence. A software function that analyzes a radiologist’s report to generate a summary with specific diagnostic recommendations would be exempt. However, if the software directly analyzed an image to create the report, it would still be regulated by the FDA. 

The guidance also spells out what information device makers must provide to clinicians so they can review the basis of the software’s suggestions. Software intended for critical, time-sensitive tasks would still be considered a device, and companies should provide a description of the underlying algorithm, validation and required input information, the FDA said.

The pair of guidance documents offers some insight into the Trump administration’s approach to AI. In past months, the administration has pushed for a deregulatory approach, but with few details. The Department of Health and Human Services issued a request for information to the healthcare industry in December 2025, asking for feedback on how the department can help speed adoption of AI.

Both documents were published without a public comment period, an approach the Trump administration has recently taken for major policy changes.

Michelle Tarver, director of the Center for Devices and Radiological Health, provided an update on digital health and artificial intelligence policy at the Association for the Advancement of Medical Instrumentation’s neXus conference.

Tarver said the CDRH plans to issue final guidance on AI lifecycle management, and told attendees to watch for the center’s thinking following two advisory committee meetings on generative AI.

The Food and Drug Administration published draft guidance on AI lifecycle management in January 2025. The document outlines best practices to ensure AI-enabled devices are safe and effective, and how developers should address transparency and bias. 

All AI algorithms are required to be trained on data that reflects the intended use population and should be validated in the intended use population, Tarver said.

The draft guidance also addresses postmarket monitoring after an AI-enabled device is deployed in the real world, knowing that this technology can hallucinate, bias can come into play and models may drift, or become less accurate over time.

“All of that can impact the output that those technologies are generating and impact patient outcomes as a result,” Tarver said.

The FDA is currently looking at comments from the draft to issue a final guidance, Tarver said.

The CDRH director also addressed generative AI, a technology that can generate text, images, software code and other outputs.

The CDRH’s Digital Health Center of Excellence held two advisory committee meetings on the topic: one in 2024 on how the agency should regulate generative AI, and another in 2025 on digital mental health devices, including chatbots that may or may not be overseen by a clinician.

“We are taking all the feedback that we heard, and I encourage you to look toward the end of this year for some initial thoughts,” Tarver said.

The update provided clarity as the Trump administration has focused on AI adoption and deregulation, but with little information on what that means for healthcare.

In 2025, after President Donald Trump was sworn in for his second term, the CDRH issued just one final guidance related to AI, on predetermined change control plans. Meanwhile, the agency grappled with staff cuts, which affected experts in AI and digital health, although some employees were reinstated.

In 2026, the new year started off with two unannounced changes to digital health guidance documents that exempted some types of wellness features and clinical decision support tools from device regulations.

Amid a deregulatory push by the Trump administration, the Food and Drug Administration is scrutinizing its digital health policies. The agency suddenly issued a pair of guidances in January 2026, intended to clarify its approach to wellness devices and medical software. 

The updates reflect changes to the agency’s thinking about what counts as a wellness device, but also raise new questions and pose challenges to consumers, experts said. 

FDA Commissioner Marty Makary announced the pair of guidances — issued without any prior notice or public comment period — at the Consumer Electronics Show in early January 2026. Makary said the agency has 27 different guidances that deal with software and digital health, and he aims to cut that number by at least half, while updating them to be more clear, modern and consistent.

Despite Makary’s framing, attorneys viewed the updates as less of a major change to regulations, and more as tweaks and examples. 

“He was talking about cutting red tape and deregulating, and that’s not really what these are,” said Amanda Johnston, a partner at Gardner Law. “The law itself has not changed.” 

Wellness guidance clarifies requirements for wearables, but muddies the water for patients

The FDA’s guidance on wellness products had the most significant changes of the two, experts said. The document was focused on wearables, “which aligns with what we’ve heard from this administration more broadly,” said Scott Danzis, a partner at Covington. Health and Human Services Secretary Robert F. Kennedy Jr. has repeatedly emphasized wearables as part of his agenda, calling for everyone in the U.S. to wear one.

The most notable change to the guidance is that it specifies products derived from certain measurements, such as blood pressure or blood glucose, don’t have to be regulated as medical devices if they are used for wellness. The position is a reversal from a warning letter the FDA sent to wearable company Whoop in July 2025, saying that blood pressure measurements are “inherently associated with the diagnosis of hypo- and hypertension.”

“This guidance, to me, pretty clearly signals that that is not their position anymore,” said Claire Davies, a shareholder with Polsinelli. 

The change provides helpful clarity for developers, Davies added, but will likely result in more products coming to market without FDA review.

For patients, differentiating between products that have been authorized by the FDA and those that fall under the wellness exemption can be challenging. 

“They’re counting on savvy consumers to be able to understand the difference, where I don’t think that’s always the case,” Gardner’s Johnston said.

It will be important for patients to look at the labeling or marketing claims of devices, said Priyanka Shah, a principal project engineer in device safety at ECRI, a nonprofit focused on healthcare safety. If the label at any point says wellness, Shah added, the device should not be used for diagnostic purposes. 

Shah compared it to the use of pulse oximeters during the COVID-19 pandemic to track blood oxygen saturation. Many pulse oximeters sold in stores or online are not medical grade or FDA cleared. Concerns about the accuracy of unregulated, over-the-counter pulse oximeters came up in a 2024 advisory panel.

“They can just mention in their labeling that this is for sports or recreation purposes, and that will completely take that product off that FDA regulatory pathway,” Shah said. 

When can wearables tell people to see a doctor?

Another significant change in the wellness guidance was that wearables can now direct users to seek an evaluation by a healthcare provider if they get a reading outside of ranges appropriate for general wellness use. The act of telling someone to see a doctor wouldn’t make a product a medical device. 

“I think it’s important that under this guidance there is an opportunity to say to the user, you should get checked out, but how you go about doing that is a tricky question,” Covington’s Danzis said. 

With the changes, for example, if a person had a low oxygen saturation reading, their device could tell them to talk to a clinician, Danzis said.

The challenging part is that the FDA does not define the limits around what ranges are appropriate for a general wellness use, Polsinelli’s Davies said. Additionally, the guidance says that for products to fall under the wellness exemption, they cannot name diseases or conditions, or characterize an output as abnormal or concerning. 

“One of the open questions is how do you do this in a way that is appropriate but also useful?” Danzis said. “If you or I were using a tool, and you get a notification saying that you should talk to your doctor, about what?”

Questions about clinical decision support

The FDA also published guidance in January 2026 that makes some changes to how it regulates clinical decision support software. While the base criteria for what counts as clinical decision support, or CDS, has not changed, the FDA made a few tweaks to when CDS falls under enforcement discretion.

Under previous guidance, software could not provide a sole recommendation, such as a drug or a diagnosis for a patient, and be exempt from medical device regulations, Davies said. Under the new guidance, it is ok to provide only one recommendation if it’s clinically appropriate. 

Davies said the change was a useful clarification, but it’s unclear what the standard is for there only being one clinically appropriate option.

One example provided in the guidance says that software that predicts a person’s risk of future cardiovascular events based on their weight, smoking status, blood pressure and test results would fall under enforcement discretion. However, the same device, if predicting a person’s risk of a cardiovascular event over 24 hours, would fall under FDA oversight.

This timing element is not new, Danzis said, but the way the FDA has framed it in the latest guidance raises some questions. If the software were to make a prediction for the next week, or the next year, he asked, would it be regulated as a medical device?

Questions about future AI regulations

When Makary announced the two guidances in a post on X, he said they would “promote more innovation with [artificial intelligence] in medical devices.” The documents themselves don’t say anything new about AI, Davies said, although CDS is relevant for people who are developing AI products.

The industry is waiting for an update after two digital health advisory committees discussed generative AI and the FDA received comments on how the agency should monitor the real-world performance of AI in medical devices.

“There is also just this huge waiting game for when and how FDA is going to regulate or authorize for marketing a gen AI-based product,” Davies said.

The Food and Drug Administration received more than 100 comments after seeking responses on how to monitor the real-world performance of artificial intelligence in medical devices. 

The feedback diverged, with patients calling for stronger postmarket reporting and medical groups saying reporting should be the responsibility of manufacturers. Device companies, meanwhile, called for the FDA to use its existing regulatory frameworks instead of introducing new requirements.

The FDA’s emphasis on real-world performance comes as the agency considers how to regulate increasingly complex technologies, such as generative AI, and how to ensure the performance of AI models doesn’t degrade over time. 

Industry groups oppose universal postmarket requirements

Medtech lobbying groups and individual companies called for the FDA to use existing quality metrics and a risk-based approach rather than implementing universal postmarket monitoring requirements.

AdvaMed, a medical device industry group, recommended that the FDA use existing regulatory requirements, such as those outlined in the Quality Management System Regulations, adding that they provide “robust mechanisms” for design validation and postmarket surveillance.

“Duplicative or prescriptive new requirements for performance monitoring of AI-enabled devices risks undermining both patient safety and innovation,” the trade group wrote in comments.

AdvaMed instead called for a risk-based approach built on QMS and international consensus standards, adding “there is no one-size-fits all approach to performance monitoring for AI-enabled devices.” 

The Medical Device Manufacturers Association also called for a risk-based approach, adding that special monitoring requirements should be used only in special circumstances. The lobbying group said that locked AI models, which don’t change autonomously over time, may carry lower risk and not require postmarket monitoring. 

“In contrast, continual machine learning models that update autonomously based on new data may introduce additional complexities and risks, which could call for specific monitoring mechanisms beyond standard controls,” wrote MDMA CEO Mark Leahey. 

Olympus Corporation of the Americas also called for the use of existing quality management structures, and Masimo supported a risk-based approach. 

Healthcare providers say monitoring should be manufacturers' job

Hospitals and medical groups see a need for postmarket monitoring of AI devices, but they said that work should be manufacturers’ responsibility. Comments emphasized the growing number of AI tools, but also noted that many hospitals don’t have the resources to evaluate or monitor these technologies.

The American Hospital Association wrote in comments that hospitals are expanding their use of AI applications. Although the technology is mostly used for administrative tools, facilities are also deploying AI-enabled medical devices.

“The potential for bias, hallucinations and model drift demonstrates the need for measurement and evaluation after deployment,” wrote Ashley Thompson, the AHA’s senior vice president of public policy analysis and development. 

Thompson said the FDA should update adverse event reporting metrics to include AI-specific risks. The AHA also recommended that the FDA add monitoring requirements for manufacturers, from periodic revalidation to ongoing surveillance, depending on a device’s risk. The hospital lobbying group suggested that the FDA focus its efforts on higher-risk areas related to the diagnosis of conditions or the treatment or mitigation of disease, and not clinical decision support or administrative tools. 

“The ‘black box’ nature of many AI systems can make it more challenging for hospitals and health systems to identify flaws in models that may affect the accuracy and validity of an AI tool’s analyses and recommendations,” Thompson wrote. “As such, post-market measurement and evaluation standards should be developed for vendors.”

Thompson added that some hospitals — particularly rural and critical access facilities — may not have the staff or resources to support AI governance and ongoing monitoring. 

The American College of Surgeons offered similar comments, supporting postmarket monitoring but delegating that responsibility to vendors instead of surgeons or other clinicians.

Patients call for transparent monitoring

Patients wrote to the FDA calling for transparent monitoring and better performance metrics that reflect people’s lived experiences. 

“When an AI-enabled device misfires, patients experience it as extra visits, extra tests, confusion about why their care plan changed, or mental distress when an automated output contradicts what they know about their own bodies,” wrote Andrea Downing, co-founder and president of the Light Collective, a nonprofit advocating for patient rights in health tech. 

These types of burden rarely appear in traditional reporting metrics, Downing wrote, adding that device evaluations should include considerations such as additional appointments, delays in care, confusion about the AI’s role, and emotional or mental distress.

Dan Noyes, a healthcare AI strategist, mentioned his personal experience living with a chronic health condition in comments to the FDA. He called for disclosures to patients about how AI tools are involved in their care decisions, as well as transparency for when models are updated, and testing across diverse populations to ensure equitable performance. 

“From the patient perspective, approval is not the finish line — it is the starting line,” Noyes wrote. “Patients need confidence that these devices will continue to perform safely and fairly after deployment.”

At AdvaMed’s annual medical device conference, big tech companies touting artificial intelligence have had a growing presence. This year’s event featured speakers from Google Cloud, Microsoft and NVIDIA, as well as a keynote on the future of AI in healthcare. 

MedTech Dive spoke with company leaders at the conference to hear what’s on their minds when it comes to AI in medical devices. They emphasized focusing on outcomes, privacy and a shift in the type of AI being used.

Here are five takeaways on AI from AdvaMed’s The MedTech Conference:

1. Focus on outcomes over flashy tech

In a keynote discussion, medtech firms emphasized the importance of purpose-driven innovation over flash-in-the-pan AI projects. 

Todd Mersch, general manager of Microsoft’s U.S. life sciences business, said he noticed a lot of “sprawl” during the first few years of companies transitioning to using more AI tools. Companies would pick up use cases based on inspiration or excitement, but there wasn’t a great connection back to the business.

“It was like a bunch of candles with no cake underneath of it,” Mersch said.

During the panel discussion, company leaders described the challenge of picking which AI pilots to pursue. 

“I think we’re still working on that a little bit,” said Lisa Earnhardt, Abbott’s group president of medical devices. Earnhardt said Abbott, being a large company, can have several projects in the works, while knowing that it might not want to scale up all of them.  

Resmed CEO Mick Farrell said he wants to see more discussions around AI focus on results, and for companies to publish those results. Farrell would prefer to see fewer conversations focused on how much companies are investing in technology.

“Nobody cares,” he said. “What they care about is outcomes.”

2. Privacy standards must be stronger

Abbott’s Earnhardt said that while there’s a lot of interest in AI, the company is taking a step back and thinking about issues such as guiding principles, standards and transparency. She said it’s important to bring the same standards and trust to the technology as to Abbott’s other medical products. 

Many tech companies, such as Google, Meta and Amazon, view their users as the product, but medtech should have a line separating companies from that approach, Resmed’s Farrell said. That line should include privacy protections, cybersecurity and interoperability. 

“I know a lot about 33% of someone's life — when they went to bed, when they wake up, every breath they took,” said the CEO, whose company sells equipment including CPAP machines.

People using a medical device should know they’re not the product, he added, and they should be able to access their data.

3. Companies are closely watching global AI regulations 

As the U.S. retools its approach to AI under the Trump administration, and the European Union implements its first AI Act, medtech companies are working to keep pace with global regulations. 

“We’re trying to just keep a pulse on how policy is moving forward in general,” Shweta Maniar, Google Cloud’s global director of healthcare and life sciences, said in an interview. 

In the U.S., the Trump administration pulled back the Biden administration’s AI plan, instead calling for a deregulatory approach and to “sustain and enhance America’s AI dominance.” What that means for the healthcare industry, and medical devices in particular, is still unclear.

Based on its latest priority list of guidances, the Food and Drug Administration’s device center seems to be building on previous work. The center is prioritizing a final guidance on predetermined change control plans, a framework for allowing certain changes to devices after they’re on the market, and a final guidance on lifecycle management considerations for AI-enabled devices, following a draft version published in January 2025. 

“Generative AI …. is still relatively new for the entire industry and also for the regulators,” Roland Rott, GE HealthCare's CEO of imaging, said in an interview. 

Rott said the FDA is very interested in understanding the potential and risks of these capabilities, and companies are working closely with the agency on issues such as what safeguards are needed around data quality, transparency and bias for medical devices. 

“What you see is the regulators, both FDA [and] also in Europe, have stepped up their guidances and, ultimately, expectations when it comes to medical devices over the last years, and they will keep doing so,” Rott said.

4. The type of AI companies use is shifting

Some of GE HealthCare’s first AI tools used machine learning. Then, the imaging company started using deep learning, a type of tool that uses neural networks to find patterns, for image reconstruction. These initial models, which comprise the more than 100 devices GE HealthCare has taken through the FDA, were focused on targeted use cases, such as finding lesions in a structure or segmenting the brain. 

This approach fits well within the current framework for medical devices because it can be validated and documented for a specific use case, Rott said. But it is also limited because there are so many different disease types and different variations in humans that might not be encompassed in these individual models. 

GE HealthCare, and other medtech companies, have started testing foundation models, another type of framework where large datasets are used to build a model that can adapt to new problems.

For example, GE HealthCare is working on a full-body 3D MRI foundation model built on more than 173,000 MRI images from more than 90,000 studies, Rott said. The company is working on a similar project with X-ray images, based on 1.2 million anonymized datasets. 

5. Big tech’s role in healthcare is changing

In the past, big tech companies were always on the outside, trying to play a role in the healthcare industry, Google Cloud’s Maniar said. That dynamic has shifted. 

“Now, I feel that technology is the underlying foundation of the healthcare ecosystem,” Maniar said. 

Tech firms have been increasingly partnering with medical device companies. Google Cloud worked with diabetes tech firm Dexcom on a weekly insights feature for its over-the-counter glucose sensors that uses generative AI. NVIDIA has been striking AI partnerships with companies including Medtronic, Johnson & Johnson and GE HealthCare. Amazon Web Services has also partnered with AI companies in the field. 

“Medical devices is a very traditional industry and it is ripe for innovation,” Maniar said.

More medical device companies are promoting the use of “foundation models,” a type of artificial intelligence that can be adapted to a wide variety of tasks. However, there are still questions about use of the technology in the medtech industry.

GE HealthCare has promoted an MRI research foundation model, Philips announced plans to work with Nvidia to build a foundation model for MRIs, and several abstracts at the 2024 Radiological Society of North America meeting focused on how to assess and improve foundation models. The Food and Drug Administration also updated its database of AI-enabled medical devices to note that it is exploring ways to identify and tag devices that incorporate foundation models. 

However, the definition of what counts as a foundation model is unclear, experts said, and it’s difficult to know if the tools available today are helping radiologists and patients.

What are foundation models? 

Magdalini Paschali, a postdoctoral scholar at Stanford’s Department of Radiology, said foundation models have a few key characteristics: They’re trained on large datasets, which mostly consist of unlabeled data. They can process multiple types of data, such as images, text, medical history and genomics. And finally, they can tackle a wide variety of tasks, such as a model detecting a disease it hadn’t seen during training.

Paschali published a paper in 2025 in RSNA’s Radiology journal looking to define the technology more clearly.

In practice, “everything can really be defined as a foundation model,” said Akshay Chaudhari, an assistant professor of radiology and biomedical data science at Stanford.

Chaudhari said the term “foundation model” was first coined at Stanford in 2021. In healthcare, one of the first versions was the debut of Google’s Med-PaLM in late 2022, a large language model designed to answer medical questions.

Foundation models started becoming more prominent at RSNA in 2023, Chaudhari said. 

Traditional deep learning models used in radiology, such as those for detecting pneumonia, are focused on a specific health condition and rely on labeled data. For example, radiologists would go through images, circling instances of pneumonia, or highlighting it in the text report, said Nina Kottler, associate chief medical officer for clinical artificial intelligence at Radiology Partners.

For foundation models, which are trained on millions of images instead of thousands of images, requiring labeled data simply isn’t practical, Chaudhari said. 

Are foundation models more accurate?

Some medical device developers claim foundation models are more accurate than narrow AI models. For example, Aidoc, which makes triage software for radiology, says foundation models allow for faster development of more accurate AI tools. 

Experts said the devices’ accuracy depends on how they’re built. Stanford’s Paschali said a foundation model used straight “out of the box,” without any specialization or additional training, may perform worse than a more specific AI tool, but it could work better once it has seen some examples and context.

Because foundation models are trained on such a vast amount of data, they can perform better at finding rare events, such as brain aneurysms, Kottler said. 

“When you only have a small number of people that have something, finding that thing is like a needle in a haystack,” Kottler said. “You need a very accurate model to be able to do that.”

Another area where foundation models can shine is allowing for faster development of other AI models. Building a traditional narrow AI model, it might take six months to clean the data, label the data and go through training, Kottler said. Different iterations of a foundation model can be built in a matter of weeks. 

In practice, however, it’s difficult to know if those advantages have translated to real benefits for patients or their care teams. For foundation models that have gone through the FDA, there’s little public information backing companies’ claims, Stanford’s Chaudhari said. 

“The best we can do is look at the summary statements that come from the FDA clearance documents,” he said, “and at least on the market, we haven't really seen the benefits of these foundation models yet.” 

Evaluating foundation models

Currently, the foundation models that have been authorized by the FDA are designed to solve a specific task, such as Aidoc’s rib fracture triage tool that is built on the company’s foundation model. Aidoc received 510(k) clearance for its model using an older version of the rib fracture triage tool as the predicate. The 510(k) process requires that manufacturers demonstrate substantial equivalence between their device and a predicate device that can be marketed in the U.S.

For broader models that incorporate language and images or video, there are no guidelines yet, Chaudhari said.

Some hospitals have come up with systems for evaluating AI models, but they’re not perfect. For example, hospitals will identify a need, such as a model that can identify pneumonia in X-ray images or a model to draft X-ray reports. 

“Then they'll compile 1,000 images from their site where they know the labels, and then they'll basically hold a competition to see which vendor maximizes the performance on that dataset,” Chaudhari said. “It’s rudimentary, but that's really the best of what we have, because that's the only way we can assess that local performance will actually be ideal for a given task.”

This might be doable for larger academic medical centers that have a data science team, but “a lot of hospitals will just deploy models and get questionable quality data out of it,” Chaudhari added.

To test a foundation model for accuracy, it’s important to first define metrics and tasks for the model based on claims about what it can do, Paschali said. Hospitals should also test how the model performs in different subgroups of patients and different scanner types. Finally, hospitals should “stress test” a model to identify potential issues that could arise, such as around very rare diseases. 

“That's why it's very important to work closely with radiologists, because they can help us design the stress testing in a very thorough manner,” Paschali said. “Because they have seen so many cases, and they know when even tasks are difficult for them.”

The hope is that foundation models, being trained on much larger datasets spanning different states, types of hospitals and imaging machines, could require less evaluation before deployment.

“I don't think we've gotten there, at least there is no evidence out in the world,” Chaudhari said. “But that's the allure of what these foundation models can provide.” 

Another goal is to free up radiologists’ time, amid an ongoing shortage of radiologists in the U.S. and a growing number of images. 

“If we're asking them to verify the output, to doublecheck everything,” Chaudhari said, “is that actually fulfilling the promise that these models have?”

Medtech firms in radiology are introducing some of the first artificial intelligence devices based on foundation models, even as a regulatory pathway for the technology remains unclear. 

Foundation models are a type of AI that are trained on large datasets and can provide multiple outputs, such as detecting several health conditions in a scan. Large medtech firms like GE HealthCare, technology firms like Nvidia and Amazon Web Services, startups, and radiology practices are introducing their own models.

The radiology sector has been an early adopter of the buzzy technology, with the promise of greater accuracy and faster model development, but experts said its benefits still have yet to be proven.

Khan Siddiqui, CEO and co-founder of the AI startup HOPPR, previously worked on healthcare AI and medical imaging at Microsoft. Siddiqui, a trained radiologist who founded HOPPR in 2019, noticed rising radiology workloads amid an ongoing shortage of clinicians.

“My spouse is also a radiologist — we started seeing people burning out,” Siddiqui said.

AI models have been used in radiology since the late 1990s, when the first tools were authorized by the Food and Drug Administration. As of May 2025, the agency has greenlit more than 950 AI-enabled devices in radiology alone, making up a majority of all authorized AI devices.

Many of these models have been one-off solutions, such as models to help flag hemorrhage in scans. However, this doesn’t always align with how patients go through a radiology practice.

“A patient comes in, and has a headache. A headache could be hemorrhage, could be cancer, could be nothing, could be trauma,” Siddiqui said, adding that covering all of the possible conditions would require buying many applications. 

Siddiqui wanted to build a large, pre-trained model that could be used across multiple health conditions, patient demographics and scanner types. To accomplish that, he needed a lot of data, which meant building out protections around privacy, security and consent.  

Building a foundation model is a “very long, very complex and very costly process,” Siddiqui said. “We’re talking about tens of millions of images.” 

HOPPR trained a foundation model on a large dataset of chest X-rays, spanning hundreds of health conditions, Siddiqui said. But the company isn’t taking that model directly to the FDA. Instead, it’s offering it to other medtech companies that want to use HOPPR’s foundation model as the basis for their own fine-tuned models. 

HOPPR’s data practices are an important part of that, because if a company wants to sell an AI model for commercial use, they must know where the data came from and have rights to use the data. A lot of research data restricts commercial use, and it’s difficult to know the source of images scraped from the internet.  

“Because we source data directly from health systems and large radiology practices, we know exactly where the data came from and can provide that to the FDA as needed,” Siddiqui said.

DeepHealth, a subsidiary of outpatient imaging provider RadNet, is one of HOPPR’s first commercial partners to adopt the technology, Siddiqui said. 

Taking a foundation model through the FDA

Aidoc, like HOPPR, sees foundation models as a launching point for more specific radiology tools. The company started working on its own foundation model about two and a half years ago, after the launch of ChatGPT, CEO Elad Walach said. 

“It was a massive effort all through the years, but we had a really big breakthrough about six months ago of the first working model,” Walach said. “It took us a lot of trial and error, a lot of resources spent on it.” 

At the 2024 Radiological Society of North America conference, Aidoc debuted its CARE foundation model, a vision-language model built using CT and X-ray images, and supporting clinical information such as notes, labs and vitals. 

Walach said Aidoc has since received FDA clearance for two derivative models based on the technology: a rib fracture triage tool and another to detect aortic dissection, a serious condition caused by a tear in the aorta.

“The way the FDA works today, it’s still disease by disease,” Walach said. “The accuracy … has to be super high to actually help physicians.” 

Walach said the company’s CARE model is more accurate and comprehensive than the company’s past AI tools. For example, using the technology for abdominal CT scans, which contain dozens of diseases, “for each of them, it was better than every one of our prior models,” Walach said. 

Like HOPPR’s Siddiqui, Walach also said the foundation model has reduced development time for derivative models. His goal is to cover more than 100 diseases with AI models in the next three years, “which was truly unimaginable a year ago.”

Aidoc also plans to open its foundation model to other developers in the future.

Radiology firm pilots foundation model 

Radiology Partners, a national practice encompassing more than 3,400 sites, is taking a different approach. The company has made two different models that it is piloting with radiologists.

The first tool, called Mosaic Reporting, uses large language models and voice recognition to structure radiology reports and reduce time spent dictating notes. Radiology Partners has rolled out the model to about 316 radiologists so far, said Nina Kottler, associate chief medical officer for clinical artificial intelligence. Because the tool only uses language, it doesn’t fall under the FDA’s purview.

The second model, Mosaic Drafting, is more complex. It combined a large language model and a large vision model to interpret images and pre-draft X-ray reports. A radiologist then reviews, edits and signs off on the findings. 

Because the model interprets images, it must go through FDA review. Radiology Partners is testing the tool under an institutional review board, Kottler said, and so far 57 radiologists have access to it.

Feedback so far has been “unanimously positive,” Kottler said, while adding that the company is moving slowly with the rollout.

“A lot of people think of the technology, because it's so capable, that … someone could just buy it and plunk it in and maybe do a little training. It's actually far from the truth,” Kottler said. “There is a lot of work that is required for training and change management, and understanding how it fits in the workflow, and making sure you get feedback and continuous monitoring.”

Practices are still figuring out how to handle many of these aspects.

Kottler sees foundation models as the future of AI, adding that Radiology Partners hasn’t looked at a narrow AI solution for about a year. 

“People, if they're worried about it, or they just hear about hallucinations and they're afraid, I would tell them … get involved,” Kottler said. “This is going to be a part of your future. And you, especially if you're a clinician or you're involved in clinical care, need to be the person that is driving it.”

As artificial intelligence proliferates across the medical device sector, the industry is seeing a shift. At the end of 2024, the Food and Drug Administration had reviewed more than 1,000 AI devices, most designed to detect or triage specific health conditions. Now, medtech companies are talking about the use of broader AI tools that can analyze images, text and other types of data across multiple contexts. 

At the Radiological Society of North America’s conference in 2024, more speakers focused on foundation models, a term for models pre-trained on massive datasets that can be adapted to a variety of tasks. And at the beginning of 2025, AI experts at medtech and radiology firms interviewed by MedTech Dive said their focus has shifted to foundation models. 

Rowland Illing, Amazon Web Services’ global chief medical officer, discussed the trend and how AWS is partnering with companies around AI, including Illumina, Johnson & Johnson MedTech, Medtronic and Abbott. 

This interview has been edited for length and clarity.

MEDTECH DIVE: Tell me about your background and how you got started at AWS.

ROWLAND ILLING: I'm an academic interventional radiologist by background. I trained in surgery originally, then did research into image-guided cancer therapy using medical devices. I took a medical device end-to-end through the regulatory process, and then retrained as a radiologist. 

I was part of the largest imaging intervention network in Europe. We had over 300 medical centers across 16 countries. The only way you can do that kind of scale play is using cloud. And so that's where I first got to know cloud and how to implement AI on top. 

I was working as chief medical officer for the Affidea Group, and realized that to try and work with 300 different medical centers, all with different IT platforms, doing things slightly differently — not being able to integrate that data — was really difficult. The best way to deploy AI really is at cloud level. Because having to implement AI on a center-by-center basis is really difficult — to deploy it locally and to manage it locally, and then fix it locally if it goes wrong.

That's really where I first got to know AWS, because all of the AI that we were adopting across all the countries was built on AWS. 

What types of AI are you working with right now? Is most of it generative AI?

We're seeing a massive explosion of generative AI in use cases. It doesn't stop all of the other AI that's been happening for ages [from] going on. There are over 1,000 applications now that have FDA approval that contain AI. Most of that is narrow AI, and has been quite well established. A company like Icometrix doing brain imaging at scale, looking at scarring for multiple sclerosis, they can do a really good job of brain imaging and segmentation. That’s just good old machine learning.

So a whole bunch of use cases are still there, but I think we’re seeing an absolute explosion of generative AI cases, especially with building foundation models.

A lot of the imaging foundation models we're seeing [are] being built out on AWS today. GE [HealthCare] has an imaging foundation model around MRI. We’ve got Harrison.AI in Australia, we’ve got Aidoc out of Israel, HOPPR in the U.S. The interesting piece being that it's not just large language models; they’re large data models and with multimodal inputs. So DICOM imaging, they’ve got biological foundation models using genomics as well as language. The integration of all the different data types is really interesting in terms of extracting extra information. 

How are you approaching generative AI with the FDA?

We're also working with the FDA. The FDA platform is leveraging generative AI to synthesize information being given to them by drug and medical device companies in order to make sense of their applications. 

They’ve got a platform called FiDL, which is a platform we’ve been working with them on for a number of years. 

What does your work with foundation models look like in the medtech field?

We want to build the best infrastructure on which foundation models get built in general. Our view of the foundation model piece is there’s actually going to be hundreds, if not thousands, of different foundation models, each with very specific use cases. There will be very specialist models that are built to address specific tasks, imaging being one of those things. If you can have a very large data model with a lot of different imaging types, and it doesn't look at a very narrow piece of the imaging, it looks at the imaging in total.

At the moment, radiologists, when they look at a scan, there's tons of data that the human can't see. So the really interesting thing about foundational models is actually what is in there that potentially goes beyond the ability of humans to interpret. And so we're working with GE and Phillips and HOPPR to ingest vast amounts of data, and with the reports against those scans, to say, “If you put in any type of scan, how do you get a report out of it?” So just a base model for imaging you can start using out of the box. And then how can you start building those into new applications? How do you securely manage that foundation model and mix it with your own data?

So once the likes of GE have built their foundation model, they'll actually be able to surface it, and then that will be able to be used by third parties to then build the next generation of imaging applications.

What kinds of applications can companies build using foundation models?

It could be MRI or ultrasound or plain film or CT, so the different types of imaging scans. I spend a lot of my time as a radiologist drawing around lumps. An example of narrow AI is, [for] lots of liver scans, you draw around a lump in the liver, and you basically point the AI to it and say, “this is a lump.” So you’ve got a really well trained AI that can identify a lump in a liver, but it couldn't necessarily then identify a lump in a bone on the same scan.

And so the benefit about these foundational models, they’re trained on millions of images with the full text report. The models, in the end, will be able to look at the scan in its entirety, they'll be able to look at the bones, the muscles, the liver, the lungs, the kidneys, and be able to have a comment about all of it.

Often when a radiologist looks at a scan, they are directed. Maybe there's liver pain in the right upper quadrant, so I say, “I’m going to look at the liver.” You may not, as a radiologist, be looking at the bone. AI can look across the entirety including the bone. I think that's a very interesting attribute to some of these foundation models. 

Now they're not perfect. There still needs to be a human in the loop, and also it needs to be fine tuned on whichever dataset you're looking at, because a CT scan from one company or from one country may not look similar to another one, but having that base model trained to be quite accurate out of the box and then fine-tuned on the data from a very specific center or region will improve the accuracy again. So I think that's the direction that we’re seeing.

Artificial intelligence-enabled medical devices with no clinical validation were more likely to be the subject of recalls, according to a study published in JAMA Health Forum. 

The study looked at 950 AI medical devices authorized by the Food and Drug Administration through November 2024. Sixty of the devices were associated with 182 recall events. 

The most common causes of recalls were diagnostic or measurement errors, followed by functionality delay or loss. About 43% of all recalls also took place within one year of FDA authorization. 

Tinglong Dai, lead author of the study and a professor at the Johns Hopkins Carey Business School, said the “vast majority” of recalled devices had not undergone clinical trials. For the majority of AI-enabled devices, which went through the FDA’s 510(k) pathway, clinical studies are not required. 

“Unfortunately, it's not required, and so people don't do it,” Dai said in an interview. “So, that's why we believe it is one of the most important drivers of the recalls.”

By comparison, the study found that devices that had gone through retrospective or prospective validation were subject to fewer recalls. 

The study also found that publicly traded companies accounted for disproportionately more recall events, with public company status associated with a nearly 6 times higher chance of a recall event. Publicly traded companies accounted for about 53% of AI-enabled devices on the market, but they were associated with more than 90% of recall events in the study and 98.7% of recalled units. 

Public companies also had a lower rate of clinical validation compared to private companies. While about 40% of recalled devices from private companies lacked validation, by comparison, about 78% of devices from larger public companies and 97% from smaller public companies had no validation. 

Dai was surprised by this finding, saying that “this fundamentally has something to do with the 510(k) clearance pathway.” 

The results raise concerns about the devices’ post-market safety and reliability. Dai and his co-authors recommended requiring human testing or clinical trials before a device is authorized, or incentivizing companies to conduct ongoing studies and collect real-world performance data. The pre-market and postmarket data could also help manufacturers identify and reduce device malfunctions and errors. 

Dai also suggested a process where clearances may be revoked after five years if a device has no public clinical data, postmarket validation or proof that it is effective in the real world.

In 2023, the FDA issued three draft guidances to improve the 510(k) program, including recommendations around choosing appropriate predicate devices and when clinical data may be needed to demonstrate substantial equivalence. However, the guidance documents still have not been finalized. 

Researchers at the Johns Hopkins Carey Business School, the Johns Hopkins Bloomberg School of Public Health and Yale School of Medicine contributed to the study. It was funded by an award from Johns Hopkins University.

ECRI has named insufficient governance of artificial intelligence in healthcare as one of its top 10 patient safety concerns for 2025.

The report warned AI could lead to misdiagnosis and outlined steps healthcare providers should take to mitigate risks posed by the technology.

ECRI included other medtech-related threats on its list of patient safety concerns, noting that cybersecurity breaches can delay care and diagnostic errors can lead to improper treatment.

ECRI, a nonprofit focused on healthcare safety and quality, publishes a report of the top 10 threats to patients every year. The scope of the report extends beyond the medtech industry, with the top concern this year being the risks of dismissing patient, family and caregiver concerns. It also covers topics that affect manufacturers of medical devices and diagnostics. 

For 2025, ECRI put insufficient AI governance second on its list of the top patient safety concerns. The nonprofit said AI has been present in healthcare for years and has the potential to improve outcomes and reduce costs. Yet, ECRI is also concerned about the potential for AI to endanger patients. 

“Medical errors generated by AI could compromise patient safety and lead to misdiagnoses and inappropriate treatment decisions, which can cause injury or death,” ECRI said. “Staff may also have difficulty determining when events are attributable to AI, making such errors harder to track.”

ECRI provided recommendations for mitigating the risks posed by AI. If implemented, some of the advice could have implications for medtech companies with products that use AI. ECRI is advising healthcare providers to form multidisciplinary committees to evaluate new technologies that use AI and regularly assess safety and clinical outcomes related to practices impacted by AI.

The nonprofit put medical error and delay in care resulting from cybersecurity breaches fourth on its list of the top patient safety concerns. The ranking reflects the potential for breaches to cause patients to “experience poor outcomes from delays in tests and procedures, longer lengths of stay, more complications from medical procedures, more transfers and higher mortality rates,” ECRI said.

ECRI has advised healthcare providers to regularly assess cybersecurity risks and their adherence to best practices. Reviews should cover evaluations of threats and vulnerabilities, which can include connected medical devices, and the effectiveness of the organization’s existing security measures. 

Diagnostic error placed seventh on the list of patient safety concerns. ECRI focused on the misdiagnosis of cancers, vascular events and infections, reflecting malpractice claims data that suggest they account for most serious errors. The nonprofit’s analysis of why errors happen focused on human mistakes and organizational shortcomings, rather than the accuracy of diagnostic tests.  

Other threats to safety listed in the report include the wide availability of medical misinformation, the growing threat of substandard and falsified drugs, and inadequate communication and coordination during discharge.