Artificial Intelligence and Machine Learning technologies are an ever-improving field with great strides seen in the last decade. This growth extends into many fields, polarizing some healthcare areas, and enables the AI and ML-enabled development of medical devices that are turning points in how healthcare technologies are being designed, approved, and used.

Below is an in-depth analysis, by Joshi et al., published in Electronic, of the landscape of FDA-approved AI/ML-enabled medical devices for the trends identified, regulatory pathways taken, specialties involved, and areas of opportunity in this growing sector.

Historical Context – Trends

The many uses of AI in health started to take serious steam in the 1990s, but it wasn’t until well into the current millennium, in 1995, that the first FDA approval was given for an AI/ML-enabled medical device: the PAPNET Testing System. It wasn’t until 2018 that takeup noticeably accelerated. At the end of October 2023, there were 691 unique FDA-approved AI/ML-enabled medical devices, representing 171 new approvals since 2022. Radiology has—to date—become the dominant specialty for AI/ML-enabled devices, with about 77% of approvals. Such a high volume of devices related to radiology include rich datasets from routine imaging procedures and a discipline reliant on pattern recognition and diagnostics—things that AI/ML algorithms do quite well. The next highest specialty is cardiovascular, representing about 10%, while other specialties such as neurology and hematology make up the remainder of smaller percentages.

Overview of Regulatory Pathways

The FDA regulatory framework provides three major approval pathways for medical devices: the 510(k) clearance process; the De Novo pathway; and Premarket Approval, commonly called PMA. The 510(k) process—the most represented route, taken by 96.7% of the AI/ML-enabled devices—is when manufacturers are allowed to prove substantial equivalence to the already marketed devices. The reliance on the 510(k) pathway is significant since the requirement for new clinical trials is obviated, provided a device can be shown to perform comparably to an already approved predicate device.

By comparison, the De Novo pathway, which covers novel devices for which there are no legally marketed predicates, was used in 2.9% of approvals, while the more stringent PMA process, reserved for devices classified as high-risk, had been used by 0.4% of devices.

While these pathways ensure that standards on safety and efficacy are maintained, the heavy reliance on 510(k) in the case of AI/ML devices raises many concerns regarding the adequate generation of clinical data for the long-term validation of those technologies.

Clinical Trials and Demographic Gaps

One finding of this study is that AI/ML-enabled medical device clinical trials are not well-represented.

Only 22 devices, or about 3.2%, of the approved devices reported conducting clinical trials. This is mostly interventional or observational, with a majority being conducted on an adult population. This is a striking observation—the pediatric subjects of these trials are almost totally missing. This suggests that most AI/ML-enabled devices target the adult demographic, leaving huge gaps in their possible applications for children and young adults.

Other areas of concern include the geographic limitation of these trials. Most clinical trials involving AI/ML devices are conducted within the United States, hence limiting the diversity of the patient populations that are involved. Because of this fact, there is an urgent need to develop clinical trials with more representative samples globally, in order to ensure the extendability of usefulness across different populations with different healthcare needs and genetic backgrounds.

Radiology-Mastery and the Role of AI in Imaging

The dominance of radiology among AI/ML-enabled medical devices reflects well on the fact that this is a field deeply related to image-based diagnostics, where applications of AI can work out very well. Machine learning algorithms are really great at recognizing images. These enable the device to give considerable assistance in detecting, diagnosis, and treatment of diseases, especially in oncology, where the role of imaging is very important in finding out illnesses at an early stage and planning for treatment.

Another major factor that makes radiology the leading domain for AI/ML innovation is probably its enormous data availability. While many healthcare systems worldwide go digital by converting their imaging records into electronic ones, huge datasets are at hand for researchers and developers to really train AI models. It is this very abundance of data that has led to fast development in AI-driven diagnostics, whereby most of the devices focus on improving the accuracy and speed of radiological evaluations.

Opportunities and Challenges

While the rise in FDA-cleared AI/ML-enabled devices is encouraging, there are a number of challenges that remain. The first and foremost issue to be mentioned is a certain underrepresentation of medical specialties. Whereas the number of AI/ML applications has been growing conspicuously in such fields as radiology and cardiology, other medical fields related to pediatrics, immunology, and ophthalmology remain backward in this respect. This situation would suggest that ample room for growth exists in those areas, particularly in developing devices which could address the unique needs of pediatric patients.

The other challenge is the regulatory landscape itself. It is not clear that substantial equivalence, as needed to be demonstrated via the 510(k) pathway, could ever be fully adequate for determining the safety and efficacy of an AI/ML-enabled device—one that evolves over time through machine learning. Many AI models will keep learning and adapting after they have been deployed; how the regulators ensure that such devices remain safe and effective is a challenge. This suggests that, going forward, the FDA will have to consider new frameworks reflecting the dynamic nature of AI/ML technologies.

Final thoughts

The FDA-approved AI/ML-enabled medical device landscape is rapidly changing. It has seen great development in the last couple of years. Radiology is at the forefront, but cardiology and neurology also join in the adoption of AI/ML. However, there are still gaps in pediatric devices and global diversity in the clinical trials themselves. The FDA should continuously update its regulatory frameworks to ensure these devices continue to be safe and effective during the period of their growth. This will require new approaches to post-market surveillance, clinical trials, and the validation of AI/ML models. While the future of AI/ML-enabled medical devices looks bright, complete realization will be brought head-on with challenges.

Dr. Prahlada N.B
MBBS (JJMMC), MS (PGIMER, Chandigarh). 
MBA (BITS, Pilani), MHA, 
Executive Programme in Strategic Management (IIM, Lucknow)
Senior Management Programme in Healthcare Management (IIM, Kozhikode)
Postgraduate Certificate in Technology Leadership and Innovation (MIT, USA)
Advanced Certificate in AI for Digital Health and Imaging Program (IISc, Bengaluru). 

Senior Professor and former Head, 
Department of ENT-Head & Neck Surgery, Skull Base Surgery, Cochlear Implant Surgery. 
Basaveshwara Medical College & Hospital, Chitradurga, Karnataka, India. 

My Vision: I don’t want to be a genius.  I want to be a person with a bundle of experience. 

My Mission: Help others achieve their life’s objectives in my presence or absence!

My Values:  Creating value for others. 

Reference:

Joshi G, Jain A, Araveeti SR, Adhikari S, Garg H, Bhandari M. FDA-Approved Artificial Intelligence and Machine Learning (AI/ML)-Enabled Medical Devices: An Updated Landscape. Electronics. 2024; 13(3):498. https://doi.org/10.3390/electronics13030498

Leave a reply