AI Medical Devices Surge Tenfold in a Decade: FDA Cumulative Approvals Reach 950

The number of medical devices incorporating artificial intelligence technology has grown significantly over the past decade.

According to the U.S. Food and Drug Administration (FDA) database, as of August 7, 2024, the agency has authorized 950 AI or machine learning medical devices. Although the FDA approved its first AI medical device in 1995, the number of submissions has surged in recent years.

A review of the data by MedTech Dive shows that the FDA approved only 6 AI medical devices in 2015, while that number jumped to 221 in 2023.

Experts interviewed noted that this trend is driven by multiple factors: an increase in connected devices, growing investment in AI and machine learning, and the industry's increasing familiarity with the regulatory pathway for software as a medical device.

"We are definitely seeing a huge increase in investment, there is no doubt about that," said Jennifer Goldsack, CEO of the Digital Medicine Society, an industry organization in the digital health field.

Large medical technology companies, including GE Healthcare, Siemens Healthineers, and Medtronic, are integrating AI into their devices and developing standalone software tools. Startups such as Aidoc, RapidAI, and Butterfly Network are focusing on creating targeted solutions for identifying health conditions and improving ultrasound imaging.

Companies outside the medical device field are also actively entering the space. Apple has developed features that use watch data to detect heart arrhythmias. Chipmaker Nvidia has established partnerships with medical device companies such as Medtronic and Johnson & Johnson to help them expand their AI applications.

AI medical devices have a wide range of applications, including improving image quality, reducing scan times, and assisting clinicians with diagnosis or surgical preparation.

MedTech Dive analyzed the FDA's list in late September. The data shows that the number of AI devices cleared through regulatory approval has risen sharply over the past decade. The data also shows that imaging dominates the field, but companies are gradually expanding into other specialties.

The following charts will be updated as new data is released.

Between 2015 and 2023, the number of AI/ML device submissions to the FDA grew almost exponentially.

So far in 2024, the FDA has authorized 107 devices, and at this pace it is on track to match 2023 levels. David Niewolny, director of healthcare business development at Nvidia, said he has been working with connected medical devices since 2007. At that time, connected medical devices were still a "vision," and people were still figuring out how to extract data from devices and where that data should be stored.

"Today, these devices generally all have some form of connectivity," Niewolny noted. "Everyone feels the pressure to innovate faster." He added that this is "the first time in my career that all the technological elements are in place."

More than three-quarters of AI devices authorized to date fall within radiology. These devices include functions for improving image quality, assisting with patient positioning for scans, optimizing radiation dose, and identifying potential health conditions.

In the FDA's list, 55 devices are classified as radiological computer-aided triage and notification software, and 24 are radiation therapy treatment planning systems.

Companies are developing more AI devices in other areas. For example, cardiovascular is the second most common specialty, with 98 devices on the FDA's list. Examples of AI cardiac devices include electronic stethoscopes and software that uses ECG data to detect signs of arrhythmias or heart failure.

"These specialties are confident in these tools and have been using them for some time," Goldsack said. "Interestingly, when we start to see more products entering those specialties that may not have 15 years of experience... when will we see a new wave in new therapeutic areas?"

Nvidia's Niewolny believes robotics is an emerging area for AI applications. Before surgery, software can pull up a patient's medical history and other relevant data; during surgery, augmented reality applications can help surgeons visualize imaging data on the patient; after surgery, AI can also read video clips to provide analysis or generate reports.

"The technology that started in radiology is now extending to other departments in the hospital," he said.

Since MedTech Dive first analyzed FDA data in 2022, GE Healthcare and Siemens Healthineers have consistently topped the list in terms of AI device authorizations.

As of August 7, 2024, GE Healthcare had 81 authorized AI devices. 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 improves image quality and can reduce MRI scan times by up to 50%. As of October, the company had used the software to scan more than 34 million patients.

Beger categorizes GE Healthcare's AI strategy into three types: products that improve imaging efficiency (such as Air Recon DL), AI that integrates multi-source data to assist clinical decision-making, and enterprise-level systems for planning.

These features are sometimes built into the imaging devices the company sells, and sometimes sold as standalone subscriptions.

"Overall, the core idea behind AI is, how do we automate those redundant, repetitive, and monotonous tasks?" Beger said.

GE Healthcare has also recently acquired several AI product companies, including Caption Health, which develops ultrasound imaging guidance software, MIM Software, which develops cancer treatment and radiation dose software, and BK Medical, which produces ultrasound devices for surgical guidance.

Beger said his interest is not in "point solutions" such as triage for specific conditions, but rather in "foundation models" that can be adapted for specific uses.

Siemens Healthineers has a total of 70 AI devices on the FDA's list. Peter Shen, head of digital and automation for Siemens Healthineers in North America, said the company's AI products include features built into imaging devices as well as diagnostic algorithms. For example, the company has developed a feature that helps patients position themselves on an MRI scanner to obtain optimal images, and has also developed algorithms that assist clinicians in diagnosing conditions such as coronary artery calcification.

Shen noted that radiation therapy treatment planning is one of Siemens Healthineers' key focus areas. The goal is to precisely target cancerous tumors while avoiding radiation to surrounding healthy tissue. Shen said AI can help clinicians find the optimal target contours, allowing them to develop treatment plans more quickly.

Looking ahead, Shen is most interested in multimodal AI. Multimodal AI can integrate different types of data, such as imaging, lab results, and patient history. Shen said this can help support clinical decisions, such as whether to biopsy a tumor or how much radiation dose to give to a particular tumor.

"It won't replace any decisions, but rather respects the relationship between the doctor and the patient, providing more information or data so that the doctor can make the appropriate decision for the patient," Shen said.

The vast majority of AI devices cleared by the FDA go through the 510(k) pathway. Compared to other FDA marketing authorization options, the 510(k) pathway is relatively less burdensome, faster, and less costly. As of August 2024, approximately 97% of AI devices on the list were cleared through the 510(k) pathway.

The 510(k) pathway applies to moderate-risk devices, where the applicant must demonstrate that their device is substantially equivalent to a "predicate device" that has already been authorized by the FDA.

Twenty-two devices were cleared through the De Novo classification pathway, which applies to low-to-moderate-risk devices without a predicate. Only 4 AI devices have received Premarket Approval (PMA), the most stringent approval pathway for high-risk devices.

The FDA states that all devices on the list must undergo validation and that the diversity of the study population must be assessed based on the device's intended use and technological characteristics. Nevertheless, patient advocates are calling for stronger regulation.

Additionally, many AI tools are not subject to FDA regulations. According to the Pew Charitable Trusts, the FDA does not regulate software intended to assist with administrative tasks, such as scheduling, inventory management, and financial processing.

Software intended to assist clinicians in decision-making has been in a regulatory gray area. Guidance issued by the FDA in 2022 clarified that AI intended to make specific recommendations regarding diagnosis or treatment—such as using patient information to identify potential sepsis cases—should be considered a medical device; however, software that matches patient data with current treatment guidelines for common conditions may be exempt from regulation.