A new analysis of the U.S. Food and Drug Administration’s artificial intelligence-enabled medical device database details how the technology’s regulatory standing has evolved in recent years.
There are now over 1,000 AI-powered devices designed specifically for radiology that have been cleared by the FDA, according to the database. Radiology applications have dominated the list for years; between 2023 and 2025, the number of radiology tools cleared by the agency increased by more than 40%.
Authors of the new analysis, published in Academic Radiology, noted that their updated review could be a valuable resource for organizations that are considering integrating the technology into their practices.
“Given the pace and complexity of recent growth, a comprehensive characterization of the contemporary FDA-authorized radiology AI cohort is timely,” Stella K. Kang, MD, professor and vice chair of the department of radiology at Columbia University Irving Medical Center, New York, and colleagues noted. “For academic radiologists, an accurate characterization of this landscape informs procurement and governance decisions, identifies underserved clinical areas where research and tool development are most needed, and establishes a baseline for evaluating the foundation-model and generative tools now entering the regulatory pipeline.”
The team characterized 1,094 radiology AI devices authorized by the FDA through December 2025, examining their functions, clinical applications, manufacturers and regulatory characteristics. Although annual authorizations increased by 42% during the timeframe studied, the overall functional makeup of the devices remained relatively stable.
Medical image management and processing systems represented the largest category, accounting for 38% of authorized devices, followed by imaging systems at 33%. Interpretive AI tools accounted for 23% of devices; among non-dental interpretive devices, cardiothoracic imaging and neuroradiology accounted for 70% of authorizations, up from 42% before 2020. The proportion of multi-finding devices also increased, rising from 4% to 18% over the same period.
The analysis also identified differences between devices developed by original equipment manufacturers (OEMs) and those developed by non-OEM companies. Median FDA review time was shorter for OEM devices than non-OEM ones, at 118 days versus 138 days.
OEM and non-OEM devices also differed in their use of predetermined change control plans, which establish how future modifications to an AI-enabled device can be made without requiring a new device submission. Predetermined change control plan authorization was more common among non-OEM devices, occurring in 4% compared with 2% of original equipment manufacturer-made devices. Overall, predetermined change control plan authorization increased substantially in recent years, from just under 1% between 2020 and 2023 to nearly 9% in 2025.
This finding may signal a shift in how regulators and manufacturers approach AI-enabled devices, as it opens the door for certain software modifications to be managed within an already established regulatory framework, the research team noted. This also is another example of why it is important to quantify the regulatory pathway of AI, they added.
“The structural and regulatory features described here are likely to remain salient as the FDA-authorized radiology AI cohort continues to grow, the authors wrote. “As the field moves into an era characterized by foundation models, generalist tools, and generative AI, characterizing the existing landscape provides a baseline against which the next phase of AI integration into clinical radiology can be evaluated.”
Read more here.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.
