FDA keeps premarket review for radiology AI detection and triage software
A final order published 17 September denies a petition that sought to exempt four classes of computer-aided detection, diagnosis and triage devices from 510(k) clearance. Makers must keep filing before they market.
The U.S. Food and Drug Administration has declined to loosen the review requirements for a fast-growing category of medical software: the artificial-intelligence tools that flag suspicious findings on medical images. In a final order published in the Federal Register on 17 September 2026, the agency denied a petition that asked it to exempt four classes of radiology computer-aided detection, diagnosis and triage devices from premarket notification — the clearance step known as 510(k) [s1].
What was requested
Under the Federal Food, Drug, and Cosmetic Act, most moderate-risk ("class II") devices must clear a 510(k) review, in which a manufacturer shows the FDA that a new device is "substantially equivalent" to one already on the market, before it can be sold [s1]. The same law lets the agency exempt a class II device type from that step if it determines that a 510(k) "is not necessary to assure the safety and effectiveness of the device" [s1].
On 22 October 2025, a petition filed on behalf of the AI imaging company Harrison.ai asked the FDA to grant a partial exemption for four generic device types: radiological computer-assisted diagnostic software for lesions suspicious of cancer (product code POK); medical image analyzers (MYN); radiological computer-aided triage and notification software (QAS and QFM); and radiological computer-assisted detection and diagnosis software (QBS and QDQ) [s1]. All four currently require a 510(k) [s1].
The petition did not ask for blanket relief. It proposed that the exemption apply only where a manufacturer had already obtained a prior clearance under the same classification — a "partial limitation" meant to confine the relief to companies with an existing track record in that device type [s1]. The FDA published notice of the petition on 29 December 2025 and accepted public comment through 27 February 2026 [s1][s2].
What the FDA decided
After reviewing the petition and the public docket, the agency "determined not to exempt" the four device types and formally denied the petition by letter dated 1 April 2026; the 17 September order sets out that final determination [s1]. Because section 510(m) of the statute requires the FDA to respond to such a petition within 180 days or have it "deemed granted," the timing mattered: the 1 April denial fell inside that window [s1].
The decision rests on the four factors the agency weighs for any exemption: whether the device type has a history of false or misleading claims or of risks tied to its inherent characteristics; whether the characteristics needed for safe and effective performance are well established; whether changes that could affect safety and effectiveness would be readily detectable, or would not materially raise the risk of injury, incorrect diagnosis or ineffective treatment; and whether a change would be likely to alter the device's classification [s1]. Applied to the imaging software, the FDA concluded that "the information presented in the petition does not demonstrate that premarket notification is not necessary to assure the safety and effectiveness" of the devices [s1].
The practical consequence is that nothing changes. Manufacturers of these detection, diagnosis and triage tools "must continue to submit and receive FDA clearance of a 510(k) submission before marketing their devices," along with all other applicable requirements under the act [s1].
Why it matters
Computer-aided detection and triage tools are among the most widely marketed forms of clinical AI, used to highlight possible abnormalities on scans or to move urgent cases up a radiologist's worklist. The petition framed a 510(k) exemption as a way to speed experienced developers to market; the FDA's answer was that, for now, the evidence did not support removing that gate for these specific device types [s1].
The order is narrow. It decides one petition under one statutory factor test, and the agency was careful to say what it is not: a verdict against lighter-touch regulation in principle. The FDA noted a "longstanding commitment" to "innovative and least burdensome approaches" for device software and digital health, and said it "supports the continued consideration" of such approaches [s1]. In other words, the premarket door the petition tried to open stays shut for these products, but the agency signalled it remains open to other mechanisms.
What to watch
Whether Harrison.ai or other developers refile with additional data addressing the four-factor test; whether the FDA revisits exemptions for these product codes on its own initiative under the periodic review that section 510(m) requires; and whether any broader digital-health framework the agency develops changes how image-analysis software reaches the market. For readers, the bottom line is unchanged: these AI imaging aids still pass through FDA clearance before a clinic can use them [s1].
Sources
- Medical Devices; Exemption From Premarket Notification: Radiology Computer-Aided Detection and/or Diagnosis Devices and Computer-Aided Triage and Notification Devices — Food and Drug Administration (Federal Register) , September 17, 2026
- Medical Devices; Exemption From Premarket Notification: Radiology Computer-Aided Detection and/or Diagnosis Devices and Computer-Aided Triage and Notification Devices (petition notice) — Food and Drug Administration (Federal Register) , December 29, 2025
FDA sets up a class II category for AI software that flags heart-disease risk
A final order files AI tools that flag possible cardiovascular disease as class II with special controls, letting similar triage software clear via 510(k). The controls require real-world, subgroup-tested validation.
The FDA's blueprint for AI medical devices: what a maker must show, cradle to grave
A January 2025 draft guidance lays out the documentation the agency expects across an AI device's whole life, including bias checks across demographic groups and postmarket monitoring. It is not yet binding.
New US rules make certified health records reveal how their AI was built
A federal rule now requires certified electronic health records to publish a standard set of facts about each predictive algorithm they ship, including how it was validated and whether it was tested for fairness.
How the FDA lets an AI device update itself without a new clearance
A predetermined change control plan is the FDA's answer to a model that keeps learning: pre-approve a bounded set of future changes, and the manufacturer can ship them without returning for a fresh review.