America's AI Doctor Boom: The FDA Has Now Cleared 1,451 AI Medical Devices, Most of Them in Radiology

Artificial intelligence is quietly reshaping American medicine, one regulatory clearance at a time. According to the latest data from the US Food and Drug Administration, the number of AI-enabled medical devices authorized for use has climbed to a striking total, reflecting how deeply the technology has embedded itself into the tools that doctors rely on every day.

1,451 devices and counting

By the end of 2025, the FDA had authorized a cumulative total of 1,451 AI-enabled medical devices, according to figures revised in March 2026. The pace of approvals has accelerated dramatically in recent years, with 295 devices cleared in 2025 alone, a record annual figure that easily surpassed the 253 cleared in 2024 and the 221 cleared in 2023.

The contrast with the past is stark. Back in 2015, only six such devices had been cleared in a year, and in 2022 the figure stood at 91. The steep climb since then illustrates how artificial intelligence has moved from the fringes of medical technology to the mainstream, becoming a routine component of the devices submitted for regulatory review.

Radiology dominates the field

Not all areas of medicine have embraced AI equally. Radiology stands far ahead of the rest, accounting for 1,104 of the authorized devices, or roughly 76 percent of the total. Its dominance reflects how well suited image based tasks are to machine learning, where algorithms can be trained to spot patterns in scans that might otherwise escape the human eye.

Beyond radiology, the distribution thins out considerably. Cardiovascular applications come second with around 130 devices, about 9 percent of the total, followed by neurology with roughly 68 devices, or 5 percent. Hematology and a range of other specialties make up the remainder, showing that while AI is spreading, its footprint remains heavily concentrated in imaging.

The first generative models arrive

Until recently, nearly all cleared devices relied on predictive models rather than the generative AI that powers popular chatbots. That began to change in 2025. In February, Aidoc's CARE1 became the first foundation model powered clinical AI device to receive FDA clearance, marking an important shift in the kind of technology entering the medical field.

The frontier pushed further later in the year. A tool called RecovryAI, an assistant powered by a large language model and designed to support surgical recovery, received an FDA Breakthrough Device Designation in late 2025. It was the first such designation granted to a generative AI product in the medical device space, hinting at where the field may be heading next.

A new approach to oversight

Regulating software that keeps learning poses a unique challenge, and the FDA has adapted its approach accordingly. In December 2024 it finalized guidance on Predetermined Change Control Plans, which allow developers to update their algorithms within agreed limits. By 2025, around 10 percent of clearances already included such plans for iterative algorithm updates.

The effort has also taken on an international dimension. In August 2025, the FDA joined forces with Health Canada and the United Kingdom's regulator to publish five shared guiding principles for change control in machine learning enabled devices. The move signalled a growing recognition that AI in medicine is a global issue requiring coordinated standards across borders.

Who is building these tools

The market is led by familiar names in medical imaging. GE HealthCare tops the list with 120 cleared devices, followed by Siemens Healthineers with 89 and Philips with 50. Canon, United Imaging, Aidoc and DeepHealth round out the leading manufacturers, though the field is far from closed, as 183 companies secured just a single clearance during 2025 alone.

Questions about the evidence

For all the momentum, important concerns remain about how rigorously these tools are tested. Fewer than 2 percent of cleared AI devices were supported by randomized clinical trials, and less than 1 percent reported actual patient health outcomes. Roughly 5 percent generated post market adverse event reports, and a similar share were eventually recalled, often due to software bugs.

The picture that emerges is one of extraordinary growth paired with lingering caution. Artificial intelligence has become a fixture of American healthcare, clearing regulatory hurdles at a record pace and reaching into ever more corners of medicine. Yet as the technology advances, the challenge will be ensuring that the speed of adoption is matched by solid proof that these tools genuinely help patients.