Medical device industry continues to turn to AI

The number of medical devices with artificial intelligence technology has risen sharply in the past decade. 

The Food and Drug Administration has authorized 950 AI or machine learning-enabled devices as of Aug. 7, 2024, according to the agency’s database. While the FDA authorized the first AI-enabled device in 1995, the number of submissions has spiked in recent years.

In 2015, the FDA authorized six AI medical devices. In 2023, the agency authorized 221 devices, according to data reviewed by MedTech Dive.

The trend has been driven by more connected devices, more investment into AI and machine learning and growing familiarity with how software is regulated as a medical device, experts said in interviews.

“We’re definitely seeing huge increases in investment. There’s no doubt about that,” said Jennifer Goldsack, CEO of the Digital Medicine Society, an industry group for digital health. 

Large medtech firms including GE Healthcare, Siemens Healthineers and Medtronic are incorporating AI into equipment and building standalone software tools. Startups such as Aidoc, RapidAI and Butterfly Network are creating targeted solutions to help flag health conditions and improve ultrasound imaging. 

Companies outside of the medical device space are also getting involved. Apple has developed features that use data from its watches to detect heart arrhythmias. Chipmaker Nvidia has partnered with medical device companies including Medtronic and Johnson & Johnson to help build out their use of AI.

AI-enabled medical devices can be used to improve the quality of images, reduce scan times and help clinicians make a diagnosis or prepare for surgery.

MedTech Dive analyzed the FDA’s list at the end of September 2024. The data show a steep increase in the number of AI-enabled devices going through the regulator over the past 10 years. The data also show a field dominated by imaging, but companies are expanding into other specialties. 

These graphs will be updated as new data becomes available.

The number of submissions for AI/ML-enabled devices to the FDA grew almost at an exponential rate between 2015 and 2023. 

As of August 2024, the FDA has authorized 107 devices, which is on track to meet 2023’s mark. David Niewolny, Nvidia’s director of business development for healthcare, said he has been working on connected medical devices since 2007. At the time, connected medical devices were a “vision,” but people were still figuring out how to get the data out of the devices and where it should be stored, he said. 

“We're now at a point where all of these devices do generally have some form of connectivity. Everyone is feeling the pressure to innovate faster,” Niewolny said, adding that this is the “first time in my career where I see all the technology pieces are put together.”

More than three-quarters of the AI-enabled devices authorized to date were in radiology. They include features to improve the quality of images, help position patients for scans, optimize radiation dosing and flag potential health conditions.

On the FDA’s list, 55 devices were described as radiological computer-assisted triage and notification software and 24 were systems for planning radiation therapy treatment. 

Companies are developing more AI-enabled devices in other areas. For example, cardiovascular was the second most prevalent specialty, with 98 devices on the FDA’s list. Some examples of AI-enabled heart devices include electronic stethoscopes and software that uses electrocardiogram data to detect heart arrhythmias or signs of heart failure. 

“These are specialty areas that are confident in these tools and have been using them for a while,” Goldsack said. “What's going to be interesting is as we start to see more of these products in specialty areas that perhaps don't have 15 years of experience with these products … when are we starting to see new groundswells in new therapeutic areas?”

Nvidia’s Niewolny sees robotics as an emerging field for AI applications. Before surgery, software could pull information on a patient’s medical history and other relevant data. During a procedure, augmented reality applications could help a surgeon visualize imaging data on the patient in front of them, he said. After surgery, AI could also read video footage to provide analytics or create a report. 

“What started in the radiology suite is now migrating to other areas of the hospital,” he said. 

GE Healthcare and Siemens Healthineers have topped the list with the most AI-enabled devices since MedTech Dive first analyzed the FDA’s data in 2022. 

GE Healthcare had 81 AI devices authorized as of Aug. 7, 2024. One of its flagship AI products, Air Recon DL, was launched in 2020. Jan Beger, GE Healthcare’s head of AI advocacy, said the algorithm can enhance image quality and reduce the time for MRI scans by up to 50%. The company has scanned more than 34 million patients using the software as of October 2024.

Beger broke down GE Healthcare’s approach to AI into three groups: products like Air Recon DL that are used to make imaging more efficient, AI that pulls data from a variety of sources to help clinicians make decisions and enterprise-level systems for planning.

Sometimes, these features are built into the imaging devices the company sells. In other cases, GE Healthcare sells them as a separate subscription. 

“Overall, the idea with AI is really, how can we automate tasks which are redundant, repetitive, mundane?” Beger said. 

GE Healthcare has also purchased companies making AI products, including Caption Health, which makes guidance software for ultrasound imaging; MIM Software, which makes software for cancer treatment and radiation dosing; and BK Medical, which makes ultrasound machines used to guide surgeons through procedures.

Beger said his interest is not in point solutions, such as software to triage for certain conditions, but more around “foundation models” that can be tweaked for a specific purpose. 

Siemens Healthineers had 70 AI-enabled devices on the FDA’s list. Peter Shen, Siemens Healthineers’ head of digital and automation for North America, said the company’s AI products include features built into imaging machines and diagnostic algorithms. For example, the company makes a feature to help position a patient on an MRI scanner for an optimal image, and it also makes algorithms to help clinicians diagnose conditions such as coronary artery calcification. 

Radiation therapy treatment planning is a big area of focus for Siemens Healthineers, Shen said. The goal is to target a cancer tumor while avoiding exposing the surrounding healthy tissue to radiation. AI can help clinicians find the best contour to target the radiation, allowing them to build a treatment plan faster, Shen said. 

Looking to the future, Shen is most interested in multimodal AI, which brings together different pieces of data, such as imaging, laboratory results and a patient’s medical history. This could help support decisions, such as whether to do a biopsy on a tumor or how much radiation to give a particular tumor, Shen said.  

“It's not anything that replaces any sort of decision, but really respects that relationship that the physician has with his or her patient, and is there to provide more information or data so that he or she can make that proper decision for the patient,” Shen said.

The FDA has cleared an overwhelming majority of AI devices through its 510(k) pathway, which is less rigorous, faster and cheaper than the agency’s other market authorization options. About 97% of AI-enabled devices on the list were 510(k) cleared as of August 2024. 

The 510(k) pathway, for moderate-risk devices, means applicants have to prove their device is “substantially equivalent” to a predicate that has already been authorized by the agency. 

Twenty-two devices went through de novo classification, which is for low- to moderate-risk devices with no predicate. Just four AI-enabled devices received premarket approval, the most rigorous pathway for high-risk devices. 

The FDA says all devices on the list must be validated and must include an evaluation of appropriate study diversity based on the device’s intended use and technological characteristics. Still, patient advocates have called for stronger controls. 

Many AI tools are also exempt from FDA regulations. The agency does not regulate software intended to help with administrative tasks such as scheduling, managing inventory and processing finances, according to the Pew Charitable Trusts. 

Software intended to help clinicians make decisions has been a gray area. The FDA issued guidance in 2022 clarifying that AI intended to make specific recommendations around a diagnosis or treatment —  such as flagging potential cases of sepsis using patient information — should be considered a medical device, while software that matches patient data with current treatment guidelines for common illnesses would be exempt. 

For nearly 13 years, countries have collaborated on medical device guidance to develop a common set of standards for regulators around the world. The work has been part of the International Medical Device Regulators Forum (IMDRF), which will meet next week in Washington, D.C.

Members include the European Union, Japan, Canada, Australia and the U.S., and the World Health Organization is an official observer. The group has developed guidance documents and discussed topics ranging from early questions such as what is a medical device to how countries can regulate new technologies like artificial intelligence.

Melissa Torres, the Center for Devices and Radiological Health’s associate director of international affairs, spoke to MedTech Dive about the IMDRF’s push to get more countries involved in the group, helping countries with newer device regulatory systems develop, and the challenges around overseeing fast-moving technologies like AI and software as a medical device.

This interview has been edited for length and clarity.

MEDTECH DIVE: What is a top priority or topic that IMDRF will focus on either at the meeting next week or throughout 2024?

MELISSA TORRES: As soon as the FDA took over chairmanship in 2024, one of the big areas that we wanted to prioritize was increasing engagement within IMDRF. Last year, we created a new membership category called the “affiliate membership category.” Not-so-well-developed regulatory authorities — for example, a regulator that’s thinking about developing a medical device regulatory system — can apply to IMDRF and be part of the group. They can engage and understand what IMDRF is doing. Because many of these countries are thinking, “How do we implement some of these documents in our jurisdiction?” IMDRF has documents as simple as the classification of a medical device or the definition of a specific medical device. All of that is encompassed within a lot of the IMDRF documents.

We wanted to think about a mechanism for these countries to be able to engage and understand what's going on within the group. This new membership category gives them an opportunity to do so. Before this category, it was really difficult to become a member. You had to have a fully developed regulatory system, you had to be able to do pre-market and post-market reviews. It didn’t offer the opportunity for these countries to engage. We’re trying to continue what the European Union [IMDRF’s 2023 chair] had started and increase engagement and membership.

It's important for these countries to not reinvent the wheel. They may have three people at their regulatory authority, and they think they could do it all. We tell them to think about ways they can rely on these other regulators that are doing premarket reviews, for example, and how can they rely on our decisions already and take that at face value.

What is IMDRF’s AI working group focused on, and what is IMDRF looking at regarding the proliferation of AI in healthcare?

We’ve recently done some multilateral work with the U.K. and Health Canada on good machine learning practices (GMLPs). We issued our first document on GMLPs, and we’ve done predetermined change control plans as well. 

IMDRF is building upon that work. The first step was looking at GMLPs. We've taken that work — those basic principles we identified — and brought that work into IMDRF to create a more detailed document with the organization. That's what the AI working group is currently working on.

In January, FDA Commissioner Robert Califf spoke about the challenges around AI, such as the complexity of oversight and the lack of manpower as the use of the technology skyrockets. Do other countries have similar challenges?

It's all of it, right? I think we're all struggling on how to fit it into the regulatory framework. It's so fast evolving. We've heard that same challenge from other regulators that are at the table. Even countries that are just developing these regulatory systems ask, “How are you doing this? How are you handling AI?” They don’t even know how to classify medical devices yet, and they’re struggling with what to do about AI. It’s certainly a challenging topic.

I was just in Korea. We had a joint workshop we did between the FDA and Korea’s Ministry of Food and Drug Safety, the country’s medical device regulator, and it was all focused on AI and medical product development. The workshop not only focused on devices, but it also talked about drugs and biologics — we kind of covered the gamut. But it was the same messaging we heard across the board from many regulators: the challenge of manpower, and how to fit AI into regulatory systems that have existed for X number of years and are not really accounting for the fast-changing and evolving technologies right now that we're dealing with.

Is IMDRF interacting with groups like Advamed or Medtech Europe as it looks at AI, for example, to hear how companies are thinking about using the technology and developing it into devices?

Some of the working groups are only open to regulators, and then other groups are for regulators and industry. AI is actually one of the groups where we have key stakeholders plus the regulators involved because of the vast nature of that topic. We need to have their feedback in the process as well.

What are the biggest challenges IMDRF faces going forward?

I think it is in these areas, AI and software as a medical device. Some of the same challenges we were just talking about. How do you create harmonized documents internationally in these topic areas that are so fast evolving? Because while we're creating concepts that are sort of at a higher level, they still become outdated so quickly. How do we keep up on this? And how do we think about it continuously? It is a challenge — everybody's facing that same challenge — but it does give a great opportunity for us to be able to discuss it with all the right stakeholders at the table. I do see these as being the two areas that are going to be the most challenging as we move forward.

Sponsored content

By IQVIA MedTech

MedTech has seen a burst of innovation over the past year, with more than 20 innovative devices approved by the FDA in 2024. At the same time, MedTech companies face a spate of financial pressures as investors continue to push for profitable growth, prompting developers to seek new strategies to manage the costs associated with clinical development.

Already, developers across MedTech have turned to AI for support in administrative tasks, such as finance and HR roles. However, AI is poised to play a vital role in clinical development, as well.

Here are five ways AI-assisted technology enhances MedTech development, helping developers improve efficiencies and streamline processes.

1. Accelerating patient recruitment

Attraction and recruitment remain significant cost drivers for MedTech innovation. This is only poised to worsen as MedTech innovators develop increasingly complex and personalized devices targeted at highly specialized, geographically diverse participants.

Against this backdrop, AI-driven clinical development platforms are essential to streamlining enrollment. For example, predictive analytics of electronic health records (EHR) and claims data can help sites identify potential study participants based on high-level inclusion/exclusion criteria. In return, sites can generate eligibility lists to help facilitate and expedite patient recruitment.

In addition, innovators can also use predictive analytics to identify medical centers where ideal candidates are likely to be located so they can focus their recruitment efforts on reaching the patients most likely to benefit from the device.

2. Streamlining clinical trial design

The same data analysis mentioned above can also play a central role in optimizing clinical trial design. Leveraging patient demographic, EHR and claims data allows innovators to facilitate site selection to efficiently reach patients over the trial's lifespan.

Innovators can also leverage AI for the trial design itself. By scenario-mapping multiple designs, developers can identify the optimal approach—and plan for— any challenges that are likely to arise. For example, AI can identify the most common hurdles for enrollment based on study specific enrollment criteria. With this information, innovators can strategically pinpoint inclusion/exclusion criteria that are necessary but will not restrict enrollment.

3. Keeping patients engaged in the trial

High patient attrition rates can exponentially increase a trial's recruitment costs, delaying study completion and possibly causing studies to re-open enrollment to ensure they have proper follow up rates, but developers can use AI to minimize dropouts.

As a baseline, machine learning technology can analyze data to identify the top reasons for attrition within a specific patient population so MedTech can optimize its trial design accordingly. Once the trial is underway, AI-assisted technology can also help identify patients at risk of attrition, creating opportunities for sites to offer additional support. For example, AI can identify factors, such as lack of study specific education related to length of screening activities, related symptoms and study responsibilities, which all lead to study dropout.

Finally, AI-assisted software can provide trial participants with detailed study information and notifications to help improve engagement and adherence to clinical trial activities.

4. Finding deeper insights within clinical data

AI-assisted technologies excel at data collection and analysis, providing study teams crucial insights throughout the trial while optimizing resources. These tools allow trial sites to focus on supporting patients, tracking regulatory requirements, and providing a holistic look at the data trends in the study. This can help proactively manage safety risks across the study.

5. Enhancing post-market monitoring

AI's benefits extend far beyond approval and play a key role in enhancing post-market activities. The right AI-driven tool can leverage data to assess performance and safety once a new device reaches the market, helping generate real-world evidence to facilitate negotiations with payers and support regulatory submissions in additional markets and expanded claims.

Tracking and analyzing health outcomes also helps developers quickly identify and report on adverse events to support patient safety.

AI-assisted insights can optimize clinical trial strategies across device development — but only if you have quality data

Overall, AI can play a central role in allowing MedTech developers to manage clinical development costs while enhancing performance. By helping to inform optimal trial design, maximize patient retention and reduce the administrative burden associated with manual data collection and entry, AI helps study teams drive efficiencies before, during and after clinical trials.

However, the quality of your analysis is only as reliable as the quality of your data. Access to comprehensive and up-to-date data is critical to gaining the most value from AI-driven technology.

IQVIA MedTech is here to help with robust data sets to inform device development, strategic expertise and regulatory support. Learn more.

A growing number of medical devices include artificial intelligence or machine learning features, and the Food and Drug Administration has started tracking ones that have gone through the agency’s approval. The FDA updated its list of AI/ML-enabled devices in October 2023, showing that nearly 700 have now received marketing authorization. 

The update added a total of 171 new devices, the majority of which were cleared between August 2022 and July 2023. 

In 2022, when the FDA first unveiled the list, MedTech Dive found that the majority of cleared AI/ML devices were used in radiology and went through the agency’s 510(k) clearance pathway. That continued to be true, based on an analysis of the newest additions.

Here are five takeaways from the updated list: 

1. The rate of FDA submissions slowed in 2021 and 2022 but has started to pick up again

The number of AI/ML device submissions the FDA has received has increased every year since 2015. The agency saw the biggest increase between 2019 and 2020, when the number of submissions increased by 39%. The rate of submissions slowed in 2021 and 2022, but was expected to surpass 30% again in 2023, the agency said in a summary provided with the list.

2. GE HealthCare had the most cleared AI/ML devices

GE HealthCare and Siemens Healthineers led the list with the most cleared AI/ML devices between August 2022 and July 2023, according to an analysis by MedTech Dive. GE HealthCare has had the most authorized AI/ML devices to date. 

Other companies that topped the list include Aidoc, a Tel Aviv-based company that makes triage systems for radiologists, and Brainlab, a Munich-based firm partnering with ZimVie on surgical navigation for spine procedures. 

3. Radiology still accounts for the majority of AI/ML devices

In 2022, MedTech Dive found that the majority of AI/ML devices authorized to date have been in the field of radiology. That trend has continued, with radiology accounting for 87% of new devices last year, and 79% of new devices in the first half of 2023. 

Part of the reason radiology might be so represented on the list is because of the early digitization of imaging data compared to other specialties. The 21st Century Cures Act also specifies anything that analyzes a medical image should be regulated as a medical device. The FDA sought to clarify through a final guidance that certain types of software should be regulated as devices, such as those that use information from electronic medical records to flag potential cases of sepsis. 

4. None of the devices on the list used generative AI

Despite rising interest in large language models, such as ChatGPT, this type of technology was not used in any of the cleared devices. As of Oct. 19, no device has been authorized that uses generative AI or artificial general intelligence or is powered by large language models, the FDA said. 

The types of models featured in the list vary in complexity. Some were “shallow,” having less than two hidden layers between the algorithm’s input and output, while others used deep learning, the FDA said. Typically, the models used a combination of approaches to achieve a result, such as using one model to generate features and another model to do classification. 

Some examples of the types of devices cleared by the FDA in 2023 include an algorithm developed by Edwards Lifesciences to help clinicians predict the risk of a global hypoperfusion event in patients receiving advanced hemodynamic monitoring, and a radiation planning assistant developed by MD Anderson Cancer Center to help plan cancer treatment.

5. Most devices still went through the 510(k) pathway

The vast majority of AI/ML devices cleared between August 2022 and July 2023 went through the FDA’s 510(k) pathway. Just two devices received de novo clearance during that time period. This is consistent with previous years: 96% of cleared AI/ML devices went through the 510(k) pathway between 1995 and 2022, according to an analysis by MedTech Dive.

The Food and Drug Administration is grappling with a surge in the number of medical devices that contain artificial intelligence or machine learning features. The agency had authorized 882 AI/ML-enabled devices as of March 2024, and many other devices include AI features that don’t require regulatory review. 

Currently, AI is most often used in medical devices in the radiology field, although the technology is also used in pathology, for appointment scheduling and in clinical support tools that pull in a variety of metrics. 

The influx of AI in medical devices has raised questions from lawmakers and patient safety groups. Medical device industry group Advamed responded to a Congressional request for information on AI in healthcare in May 2024. Advamed said the FDA’s authority is “flexible and robust enough” for AI/ML in medical devices.

On the other hand, patient safety nonprofit ECRI listed insufficient governance of AI in medical technologies among its top health technology hazards for 2024. CEO Marcus Schabacker said AI has the potential to do good, but the technology also can do harm if it provides inaccurate results or magnifies existing inequalities.

MedTech Dive spoke with Schabacker about what developers, regulators and hospital administrators can do to ensure devices that use AI are safe and effective.

This interview has been edited for length and clarity.

MEDTECH DIVE: The FDA finalized guidance in late 2022 on clinical decision support tools, and which ones should be regulated as medical devices. What are your thoughts on that? 

MARCUS SCHABACKER: We think the FDA is taking a much too laissez faire approach here. They should be much more stringent, particularly with decision support tools, and have clear regulations. 

If they don't want to regulate it upfront, they need to put post-market surveillance programs in place at a minimum. But we believe they should be way more forceful in the regulatory pathway and get more evidence from the companies that their tools are truly supportive instead of influencing decisions in a more severe way. 

The FDA’s always in a tough spot. On the one hand, they want to make sure that the population stays safe. That's their mission. That’s our mission too. On the other hand, they want to make sure appropriate innovation can happen and don’t want to be seen as a blocker. 

What prompted ECRI to list insufficient governance of AI in medical technologies as a top concern for 2024? 

We’re concerned from three areas: One is development. Typically, it’s not medical people who developed the actual tool. It’s IT or software developers, who might or might not understand the work environment in which the tool will be used and rely on limited input from medical or clinical people. Therefore, the application of that might or might not be appropriate when it’s then put into actual clinical use. 

Two, AI is a self-learning tool. It will be trained on a subset of people. If that subset is skewed or biased, then what the algorithm spits out will be skewed or biased. 

We have made great strides, for example, when we design clinical trials to make sure that there’s a representative population in the testing of innovations from a gender, race and age perspective. 

While this is still not a slam dunk, at least we are aware of it now and taking care of it in clinical trials. I don’t think that is to the same extent happening in AI development. That concerns us greatly, and we’ve already seen that AI can be misused to exclude or disadvantage certain populations. For example, there was a report about a tool that favored the scheduling of people the AI expected to be able to pay, versus people who weren’t able to pay. Those biases can get aggravated and accelerated if a tool is not carefully developed. 

The last aspect when we think about AI is around how it gets implemented. Is the user aware of it? Is the clinician aware of it? Does the hospital have good processes and procedures in place? 

How are these AI/ML systems regulated as medical devices? 

Some circumvent regulation by saying we’re a decision support tool, which doesn't require regulatory oversight. Tools that could reduce dozens of inputs or variables to a manageable number could be potentially helpful. But our fear is that quickly becomes a decision-making tool, where particularly less experienced clinicians would take what the tool suggests as a decision and act upon that.

We’ve seen some of that can be problematic. One of the sepsis tools wasn’t very good at predicting sepsis.

We think that all regulatory agencies — but particularly in the U.S. — are way behind the 8-ball. The U.K.’s Medicines and Healthcare products Regulatory Agency has a better approach. They have created a virtual sandbox for those AIs. They need to demonstrate that they’re safe and effective in that virtual sandbox. 

The other point is, How do you then do postmarket surveillance on the safety and effectiveness of the use of AI? It’s going to be often difficult to find out if the AI contributed to an adverse event or if it was the human who did, or if it was the actual device. Often it’s not clear if AI is included in the device, so someone would not think about that necessarily. We’re thinking about if we can do something like a nutrition label for AI. If AI is somewhere in there: What is the AI? How was it developed? What’s the testing pool? So that users are more aware of the AI.

What is working about how the U.K. is approaching AI regulations? 

Our understanding is that before the AI is actually allowed to be used in a clinical setting, there's a virtual simulation of the application. It needs to demonstrate that it’s effective and safe to use. That’s a pre-requirement for clinical approval.

Currently, the majority of AI tools in the U.S., if they actually go through registration at all, go through the 510(k) route. And the 510(k) route, as you know, is a very abbreviated process where you essentially just need to demonstrate that you are equivalent to a device that is already in existence. That has not proven to be a very effective safety tool. 

How are hospitals approaching AI tools? 

We know of one large system that seems to have a relatively robust approach to this. And so we applaud that. The majority of what we see out there — it’s hard to grasp for an administrator. Often these companies made great promises in terms of cost savings, revenue acceleration, or patient safety. To me as an administrator, that sounds great. Sure, why wouldn’t I do that? And not necessarily understanding that a lot of these tools do not undergo any rigorous evaluation process upfront. 

There needs to be a way more thoughtful approach from administrators in saying, “AI is here. AI will stay. AI can be helpful. What do we need to do to make it safe?” 

In light of the lack of regulatory oversight and the enormous interest from a commercial perspective, and the increasing capabilities, it behooves administrators to take a more risk-averse approach.

More than 40% of all artificial intelligence tools cleared or approved by the Food and Drug Administration lack clinical evidence, according to an article published in Nature Medicine. 

Researchers at the University of North Carolina School of Medicine found that 226 AI-enabled medical devices authorized by the FDA between 1995 and 2022 were missing clinical validation data. The figure represents about 43% of all 521 devices the agency authorized during this period. 

“The FDA and medical AI developers should publish more clinical validation data and prioritize prospective studies for device validation,” the study authors wrote in the Aug. 26, 2024, article. 

Most AI medical devices have been cleared through the FDA’s 510(k) process, a less rigorous regulatory pathway where manufacturers must demonstrate their products are safe and effective compared to a predicate device.

“A lot of those technologies legally did not need [validation data], but the argument we made in the paper is it could potentially slow adoption if we’re not seeing evidence of clinical validation, even if the device is substantially equivalent to a predicate,” Sammy Chouffani El Fassi, the study’s first author, said in an interview. 

Additionally, small changes to an algorithm “can really change the effect” of a device after it is implemented, meaning the best way to know how a device will work is to test it in a clinical setting, said Chouffani El Fassi, an M.D. candidate at UNC.

The researchers categorized AI medical devices based on whether they had any clinical validation, meaning they were tested on real patient data for safety and effectiveness. In cases where validation data was available, they also tracked what type of study was done. 

Of the 292 devices with validation data, 144 were studied retrospectively, which the researchers defined as testing before the devices were implemented in patient care or using data from before the start of a study. The other 148 were validated prospectively, tested in patient care or with data collected after the start of a trial.

Of the devices tested prospectively, just 22 underwent randomized controlled trials. 

Value of prospective studies

Chouffani El Fassi said prospective studies can provide a better understanding of how a device works in real life. The researcher was part of a team at Duke University validating an algorithm that detects cardiac decompensation using electronic health record data. The team, which included the Duke Heart Center and Duke Institute for Health Innovation, validated the algorithm with a retrospective study. They also ran a prospective study where cardiologists used the algorithm and noted whether they agreed with its diagnosis of cardiac decompensation in patients. 

“Prospective validation was particularly helpful because we could see what needed to be improved in the device,” Chouffani El Fassi said, adding that the group learned the user interface needed to be improved so the device was easier to use and more efficient. 

Testing an algorithm prospectively can also reveal potential confounding variables. Chouffani El Fassi provided the hypothetical example of a device that reads chest X-rays: After the start of the COVID-19 pandemic, some people’s chest X-rays would start to appear different. Testing the device on pre-pandemic data wouldn’t reveal that new information. 

Prospective studies don’t have to be complicated or burdensome, Chouffani El Fassi added. For example, it could be as simple as a physician using an ultrasound probe optimized with AI technology on a patient, and rating on a scale of 1 to 5 how useful the tool was in clinical practice. 

“It’s really basic, but we would consider that a prospective study because the confounding variables are there. Let’s say a doctor has a hard time using it because it has a bad user interface; he or she is going to give it a 1 out of 5,” Chouffani El Fassi said. “You learned something new; you saw how that device actually works in real life. We see that as the most valuable kind of data.”

Dive Brief:

Food and Drug Administration officials have outlined how lifecycle management could address the challenges of generative artificial intelligence in healthcare.

In a blog post published in July 2024, leaders from the FDA’s Digital Health Center of Excellence (DHCoE) shared their work to map the AI lifecycle and identify key activities at each step in the process.

The authors see the model as a playbook for helping to assess standards and a launchpad for other projects, such as the development of a checklist for the systematic creation and evaluation of AI.

Dive Insight:

Troy Tazbaz, director of the DHCoE, and John Nicol, digital health specialist at the center, co-authored the blog post. In the post, Tazbaz and Nicol said lifecycle management has been essential to creating reliable software since the 1960s. Today, software development lifecycles embody the principles to provide a structured framework for each step from planning to retiring programs.

DHCoE has worked to map the phases of the traditional lifecycle onto the specific steps involved in AI development. The work generated a model that shows the AI lifecycle as a wheel that starts at planning and design. After seven total stages, such as model building and real-world performance evaluations, the wheel returns to the planning and design step, showing an ongoing development cycle.

The illustration also expands each of the seven stages of the AI lifecycle to show details of activities that take place at each step. Planning and design, for example, features activities such as problem definition, algorithm selection and feature engineering.

Tazbaz and Nicol said the model could serve as a guide or a playbook to help assess standards, tools, metrics and best practices for the considerations in each phase of the lifecycle. Applied to considerations about data suitability, the model could help “identify the relevant standards and applicable metrics, like data quality, population coverage and provenance,” and explore tools for tasks such as bias detection. 

DHCoE also sees potential for the model to support other projects, such as the AI creation checklist and “adoption of a harmonized approach to unify development strategies, techniques and discipline.” Tazbaz and Nicol are encouraging the healthcare community to “engage with, iterate and refine these concepts” and consider the lifecycle model “for standards development efforts in AI.”

Medtech companies are rapidly integrating artificial intelligence into cardiovascular medicine, from imaging and electrocardiology to genetics and patient monitoring. Best practices are still evolving, however, and few AI tools have been shown to improve cardiovascular and stroke care enough to be widely adopted, according to a statement from the American Heart Association.

Medtronic is among the device makers working to support earlier disease detection and treatment through AI-enabled technologies. The company told MedTech Dive that better algorithms and sensing technologies and an expansion of available data are improving accuracy and access across multiple specialties. For example, AI is now being used to enhance detection of polyps in real-time during colonoscopies, eliminate finger sticks for calibrating continuous glucose monitors in diabetes management and optimize screw alignment in spine surgery planning.

MedTech Dive recently spoke with Stacey Churchwell, general manager of Medtronic’s cardiovascular diagnostics and services business, to discuss the company’s integration of AI algorithms for detecting abnormal heart rhythms into its insertable cardiac monitors (ICMs). The miniature devices — the latest version is about a third the size of a triple A battery — are implanted under the skin in a patient’s chest to help diagnose abnormal heart rhythms such as atrial fibrillation (AFib).

This interview has been edited for length and clarity.

MEDTECH DIVE: How did Medtronic overcome some of the inherent challenges in developing AI, such as disparities built into algorithms that may reflect bias? 

STACEY CHURCHWELL: With regard to AccuRhythm AI, we utilized a large global data repository for training and validation, including not only standard ECGs, but also rare and edge case scenarios, helping us to mitigate biases and ensure the AI's appropriate behavior when deployed in the market. We took active steps to develop additional permutations of data that account for common differences observed across populations.

To prevent the AI from re-learning undesirable habits or biases, we locked its training once our scientists were satisfied with its performance. Before launching the algorithms, we rigorously validated the AI's decisions with human adjudicators to ensure consistency with human decisions.

What is the goal of incorporating AI into the ICM?

We started our journey by improving the specificity of our devices so that we can remove false positives or false alerts that would otherwise create burden for clinics.

An ICM looks for anything abnormal, and it's really designed to be highly sensitive. It's looking for things that don't appear to be normal sinus rhythm, and it's flagging it. Before AI was introduced, there was a potential for things to kind of slip through that may not be clinically relevant or actionable and that would require the care team to have to sift through and look at those alerts to determine and adjudicate whether in fact they're real or not. We've got some great technology on board the device itself, but it's not perfect. 

We saw that as an opportunity, and what we have done with AI is to implement it in the cloud. As those signals come in that are deemed to be suspect signals that could be an arrhythmia, that information gets sent up to the cloud where AI sits, into our CareLink network and that AI, which has been trained using millions of data points.

How is AI improving the ability of the device to do its job?

We've been doing this for 26 years. We've got a lot of data on board. We've used that data to train the AI to understand and look for those discrete patterns. It goes through a process of adjudicating those signals to say, “I have high confidence that this is real, I have high confidence that it's not real, or maybe it's borderline.”

When it gets through that filter, we immediately eliminate [false alerts] that we know with high confidence aren’t real. And so the physician, the clinician team, never sees that. We're going to always send forward [signals] that we know are real, but also those things that may be suspect. We're always going to put that in front of the clinician to make the final judgment on. But because the AI is so good, it does a fantastic job of weeding out a lot of that noise that otherwise, before the AI was activated, would have slipped through, and that burden would have been left on the care team to try to figure out.

How well is the AI performing?

We've been really, really thrilled with the results. We hear it from our clinicians that it's making a difference. We’re committed to not compromising on our ability to identify true events. That sensitivity, we try to keep in that 99% range, so that we can say that when we call something AFib, you can take to the bank that it's AFib. You're not going to miss those events that you could have otherwise missed had we compromised on our sensitivity capability.

What we've seen with AccuRhythm AI 2.0 — we launched 2.0 last summer — is that the AFib false alert reduction went from 74.1% to 88.2%. We're taking out even more of that false data.

What is the benefit of being cloud-based?

One, we get to take advantage of the computational power of the cloud, versus burdening the device itself, which would have a battery longevity consequence. The other great benefit is that we get to impact and improve devices that were implanted years ago.

When we flip the switch in the cloud, every device that's still active gets the benefit. Whether we implanted it today, two days ago, two weeks ago, two months ago, two years ago — if that device is still active, it's going to see the benefit of that AI. That clinic is going to see that relief.

Where will you be focusing your AI efforts next?

We think our next frontier is prediction, and one of the things that we have an opportunity to do is to demonstrate that Linq can play a role in facilitating heart failure management.

We've got a big clinical trial underway right now called Alleviate-HF that's exploring that very question, and we'll have results back in about a year’s time that will give us the green light to move forward in commercializing a solution and algorithm that will leverage AI for prediction. Then we can give our customers the heads up that a patient may be headed toward hospitalization, which would give them time to take action and adjust meds to perhaps keep that patient out of the hospital and drive costs out of the system.

When is it best for a heart patient to get an insertable monitor versus a wearable one?

Wearable solutions are considered as a first step. Holter monitors or external patches are generally used for a time period ranging from, say, seven days to 30 days. One of the big hurdles that those technologies have is around patient compliance.

For patients who have more infrequent symptoms, and these could be episodes that happen every two or three months or even six months, you really want to consider something more long term. The great thing about an ICM is it's set it and forget it. It's looking 24/7, and you have up to several years of a window to find what you're looking for.

There's a recent study [Stroke-AF] that showed that 78% of patients who had AFib would have been missed if you only monitor them for 30 days. We know without a doubt that 30 days just isn't enough. That's why we tend to think that ICM is a great solution to consider for those patients.

Dive Brief:

Johnson & Johnson Medtech has partnered with Nvidia to use artificial intelligence to improve surgery, the companies said in March 2024.

The agreement positions J&J to use platforms from Nvidia, a maker of computer chips used for technologies like graphics and AI, to support the deployment of surgical software and devices that enable real-time data analysis in the operating room.

J&J and Nvidia see the technologies facilitating fast, secure deployment of AI algorithms for surgical decision-making, education and collaboration.

Dive Insight:

Researchers have identified AI as a way to help surgeons navigate through the body during procedures, provide surgical robots with some level of autonomy and otherwise improve operations. Companies such as Intuitive Surgical are already using AI in their products. Medtech firms are still working to realize many of the opportunities identified by researchers. 

J&J, one of Intuitive’s competitors for the robotic surgery market, has secured access to Nvidia’s IGX and Holoscan edge computing and AI platforms. Edge computing allows data to be processed close to the user, rather than being sent to remote servers.

Moving processing to the edge eliminates the lag and security risks that can happen as data is moved to and from servers. Those benefits may be relevant to surgical uses of AI, where processing delays could negatively affect patient outcomes and the sensitivity of health information makes it important to keep the data in a secure environment. 

“Bringing advanced edge computing hardware and software to the OR enables scalability of innovation and new AI-powered solutions for clinical decision-making, education and training, and collaboration — with the ultimate goal of advancing patient care,” Shan Jegatheeswaran, J&J Medtech’s global head of digital, said in a statement. 

David Niewolny, Nvidia's director of business development for healthcare/medical, discussed potential uses for AI in surgery in a blog post about the J&J deal. The technology could support the development of an AI model that analyzes endoscope footage and identifies organs, tissue and potential tumors in real-time, Niewolny wrote in the Monday post. Algorithms could also remove personally identifiable data to allow external experts to view videos.

The Nvidia collaboration is part of J&J’s broader push to use AI. CFO Joe Wolk said on an earnings call in January 2024 the company is “looking for opportunities, whether it be aided by artificial intelligence or just infrastructure overall, as to how we can further improve the medtech profitability profile.”

Dive Brief:

The Food and Drug Administration is seeking examples of artificial intelligence and machine learning models that can identify and predict freezing of gait events related to Parkinson’s disease. 

Freezing of gait is a temporary loss of forward movement while walking. These episodes affect people’s quality of life and daily activities, but they can be difficult to measure because they often happen when patients are outside of a clinic or hospital setting. 

By testing these models against its own data, the FDA hopes to better understand the ability of these technologies to provide digitally derived endpoints that could help with early disease detection and prevention or support treatment and care in the home. 

Dive Insight:

As digital health technologies proliferate, the FDA is taking a closer look at how they can capture information about a person’s health outside of a clinical setting. For Parkinson’s disease, a handful of wearable technologies have already been cleared to track symptoms. 

Apple published study results in 2021 evaluating its Motor fluctuations Monitor for Parkinson’s disease, which uses sensors to track fluctuations in resting tremor and dyskinesia. Later that year, a company called H2O Therapeutics received 510(k) clearance for its Parky app that sends Apple Watch data to a patient’s doctor. In April 2024, H2O received clearance for another app that provides haptic feedback when people experience freezing of gait symptoms.  

Other cleared apps that use Apple Watch data include Rune Labs’ StrivePD software and an app developed by digital health company NeuroRPM. 

Nearly 1 million people in the U.S. are estimated to have Parkinson’s disease, according to the FDA. 

“With the fast-paced innovation and proliferation of DHT capabilities, including the use of artificial intelligence, the healthcare and technology ecosystems have a responsibility to ensure and support the safety, effectiveness, and privacy of digitally-derived endpoints,” the agency said in a summary of the challenge.

To start, the agency will ask developers, innovators and researchers to submit AI models in August 2024. They can use public data sources or bring their own data. 

Then, the FDA will pick the top performers from that group, and test those models against its own freezing of gait data. Its goal is to determine best practices for using AI models to develop and validate measures from wearable devices and smartphones.

The challenge is being hosted by the FDA’s Digital Health Center of Excellence, Office of Science and Engineering Laboratories and Office of Digital Transformation’s precisionFDA.

Dive Brief:

A U.K. agency outlined its position on the regulation of artificial intelligence as a medical device in a policy paper published in April 2024.

The Medicines and Healthcare products Regulatory Agency (MHRA) said many AI products that can be put on the market now without conformity assessment will move to a higher risk category in upcoming reforms.

The paper explains how MHRA interprets a government AI strategy focused on principles such as safety, security and robustness and aligns it with international standards.

Dive Insight:

The U.K. government committed to five cross-sector principles for the regulation of AI in February 2024. A consultation found “strong support” for the principles, the government said, and established them as the basis for a regulatory framework designed to keep pace with rapidly advancing AI technology. Days later, the government wrote to the MHRA to request details of its approach to AI. 

In response, the MHRA published a policy paper that describes how it is interpreting and applying the five principles. The paper covers AI as a medical device, how the MHRA is using AI internally and how the use of AI by industry may affect the agency.

AI with a medical purpose is very likely to meet the definition of a medical device, the MHRA said. Today, the agency puts many AI products in the lowest risk classification, meaning they can come to market without being assessed by a notified body. The MHRA said many AI products will move up to higher risk classifications as part of the post-Brexit medical device regulatory framework.

The agency made multiple references to the need for international alignment in its discussion of AI as a medical device, citing its use of documents from the International Organization for Standardization and the International Medical Devices Regulators Forum to mitigate the risk of bias in AI. The MHRA is also working on its own machine learning and AI development practices. 

The agency has three people, or full-time equivalents (FTEs), working on AI as a medical device activities. It plans to grow the team to 7.5 FTEs through the 2024 to 2025 financial year. The government is providing funding for two FTEs. 

The other two sections of the paper could also have implications for the medical device industry. In one section, the MHRA discusses ways it may use AI internally, such as the application of machine learning to real world data to understand links between products and clinical outcomes. The final section covers the internal use of AI by companies, which the MHRA said is generally unlikely to affect how it regulates.