AI-enabled medical devices on the FDA's list, by specialty and year

How large a part of FDA clearances are AI devices?

In 2025, 329 of the FDA's 3,198 510(k) clearances, 10.3%, were for devices on its list of AI-enabled medical devices. In 2016 the share was 0.6%, 17 of 2,930 clearances. From January to June 2026 it was 10.6%, 176 of 1,665.

Most of the AI-listed clearances come under one review committee. The FDA's 510(k) file tags each clearance with a review committee, a code for the medical specialty of the review, such as RA for radiology. 252 of the 329 AI-listed clearances in 2025 were tagged radiology.

Those 252 made up 54.0% of the 467 clearances tagged radiology in 2025, up from 2.6% in 2016. Across every other committee together, the share was 2.8%, 77 of 2,731.

How many AI devices has the FDA listed?

The FDA's list, in the version current as of September 4, 2026, holds 1,614 devices with final decisions from September 29, 1995, to June 29, 2026. The FDA says it found them "primarily based on the use of AI-related terms in the summary descriptions" of each device's authorization, and that the list "is not a comprehensive resource of AI-enabled medical devices."

335 of the devices had final decisions in 2025, 42.6% more than the 235 in 2024. In 2016 there were 18.

The first half of 2026 ran at about the pace of the first half of 2025: 181 devices from January to June, against 176 in the same months a year earlier. Three earlier versions of the list show that later updates changed the count for their most recent six months by 0 to 5 devices, so the 2026 count is unlikely to grow by much.

Which specialties do the devices belong to?

The FDA's list gives each device a lead panel, by medical specialty. This is a different field from the review committee in the first scene: the panel is the FDA's specialty grouping for the type of device, and the committee is the one the 510(k) file lists for each review. Radiology is the panel for 1,230 of the 1,614 devices (76.2%), cardiovascular for 154 (9.5%) and neurology for 73 (4.5%). 18 panels appear at least once.

From 2016 on, radiology's share of each year's devices was highest in 2020, at 86.8%, up from 61.1% in 2016. It has fallen every year since, to 75.8% in 2025 and 74.6% from January to June 2026.

From 2020, radiology's peak year, to 2025, neurology devices rose from 1 to 19 and cardiovascular devices from 8 to 30, so their shares rose from 0.9% to 5.7% and from 7.0% to 9.0%. Cardiovascular's share had been higher before 2020, between 15.4% and 25.9% from 2016 to 2019. The panel is a grouping by type of device, which is not always the specialty that uses it: several devices that analyze heart ultrasound scans are listed under radiology.

In which specialties is AI now common?

In 2025, more than half of the clearances tagged radiology were for devices on the list. For clearances tagged cardiovascular it was 27 of 301 (9.0%), for neurology 11 of 209 (5.3%) and for orthopedic 11 of 482 (2.3%). The FDA issued more than 100 clearances under each of the dental, general hospital and physical medicine committees in 2025, none of them for a device on the list.

Committees with fewer clearances, left off the chart, can show high shares on a few devices: pathology cleared 3 AI-listed devices of 13 in 2025 (23.1%) and hematology 3 of 22 (13.6%).

What kinds of devices are they?

The devices fall under 181 FDA product codes, the three-character codes the FDA's classification file uses for each kind of device. The largest is automated radiological image processing software, with 305 devices, then radiological image processing systems, with 161.

408 devices, a quarter of the list, are filed under the product code of a scanner rather than a software code: 137 under ultrasound systems, 103 under CT scanners, 100 under MRI systems, 30 under x-ray and fluoroscopy systems, 28 under PET and SPECT scanners and 10 under others. Some of them are software sold for a scanner, such as image reconstruction.

Outside the scanner group, the third-largest product code is for software that flags time-sensitive scans for review. 86 devices are radiology triage and notification software, which the FDA's rule for the type describes as a device that "notifies a designated list of clinicians of the availability of time sensitive radiological medical images for review." Another 40, under the same rule, prioritize scans with suspected lesions.

How many listed devices started a new type of device?

1,553 of the 1,614 devices were authorized through a 510(k) clearance. 21 came through premarket approval (PMA), the FDA's review of "the safety and effectiveness of Class III medical devices," the class with the greatest risk, which the FDA calls "the most stringent type of device marketing application required by FDA." 4 of the 21 PMA approvals were supplements to an earlier approval.

The other 40 came through a De Novo grant, which the FDA describes as "a marketing pathway to classify novel medical devices" for which "there is no legally marketed predicate device." A predicate is the device already on the market that a 510(k) is compared with. A De Novo grant sets up a new device type, and the new device "may now serve as a predicate device for 510(k) submissions of future devices of the same type."

The first De Novo on the list came in 2013, for a mass spectrometry system that identifies microorganisms. Seven came in 2018, among them a diabetic retinopathy detection device and the radiology triage and notification software type, which 86 devices on the list now carry.

Nine came in 2024, the most in one year, among them software to help predict or diagnose sepsis and an over-the-counter device to assess the risk of sleep apnea. Four came from January to June 2026.

Do AI devices take longer to clear?

For the AI-listed 510(k)s decided in 2025, the median time from the FDA receiving the submission to its decision was 141 days, against 126 days for all 510(k)s cleared that year. The count runs from receipt to decision, so it includes any time a submission was on hold while the company answered the FDA's requests for more information, as the FDA's 510(k) process page describes.

AI-listed 510(k)s took longer in five of the eight years from 2018 to 2025, including each of the last three. In 2024 it was 147 days against 129.