The U.S. Food and Drug Administration (FDA) took a hard line against treating CADe, CADx, and CADe/x devices interchangeably when it recently denied Harrison.ai’s petition to exempt from FDA review certain 510(k)-cleared AI software as a medical device (SaMD) used in radiology.
Instead, the FDA suggested manufacturers use predetermined change control plans (PCCP) to achieve some of the aims proposed in that petition and encouraged sponsors to pursue additional discussion with the FDA through its Q-Submission Program.
“Holding a 510(k) clearance may not reflect that a manufacturer is proficient in, or even has experience with, the processes used in the development of the cleared device, let alone in processes that would necessarily be appropriate for all future devices of the subject types,” the FDA noted in its April 1 decision letter to Rubrum Advising.
In 2025 alone, the FDA cleared at least 255 radiology AI devices through the 510(k) program, according to its AI-Enabled Medical Devices List, as identified by FDA primarily based on the use of AI-related terms in the summary descriptions of their marketing authorization document and/or the device’s classification.
4 device types
Meanwhile, the petition aimed to exempt four generic device types from premarket notification 510(k) requirements when certain conditions are met:
- Radiological computer-assisted diagnostic (CADx) software for lesions suspicious of cancer, classified under § 892.2060 (21 CFR 892.2060), product code POK.
- Medical image analyzers, classified under § 892.2070 (21 CFR 892.2070), product code MYN.
- Radiological computer-aided triage (CADt) and notification software, classified under § 892.2080 (21 CFR 892.2080), product codes QAS and QFM.
- Radiological computer-assisted detection and diagnosis (CADe/x) software, classified under § 892.2090 (21 CFR 892.2090), product codes QBS and QDQ.
The petition also called for allowing manufacturers to assess the risk of a given device and for determining that a post-market plan is proportionate to that risk, essentially flipping the standard, according to the FDA. This would have meant manufacturers could determine what action is necessary to mitigate the risks of the device in lieu of FDA review and clearance of a 510(k). However, the FDA said the petition included no mechanism for FDA to evaluate individualized plans.
Opposition
A majority of the more than 45 comments to the proposal opposed it. In its 16-page response, FDA countered several arguments, according to the FDA. One focused on determination of substantial equivalence.
“A determination that this standard has been met in the evaluation of a specific subject device compared to a specific predicate device does not automatically translate to a conclusion that the processes followed during development of the subject device will be appropriate for any future devices developed under the same or similar regulation; that the manufacturer would be proficient in those processes as applied to any such future devices; or that the manufacturer would necessarily follow the same processes for any such future devices, and the petitioner has not provided evidence to support such a conclusion,” the FDA determined.
The agency explained that among all 510(k)s cleared for devices under a particular CAD product code over a period of several years, all were placed on hold, and in each case for reasons that included deficiencies relating to performance testing methodology or performance results, even though, in many cases, the manufacturer had a prior CAD authorization.
Many aspects of AI development for one CAD or CADt indication may not readily translate to another CAD or CADt indication, such as the interdisciplinary expertise needed to understand the clinical workflow, desired benefits, and associated patient risks; the data collection protocols used to curate a dataset that is representative and appropriately sized for the indication; and the methods used to determine an appropriate reference standard for the indication, the FDA said.
In addition, CADe, CADx, and CADe/x devices cannot be treated interchangeably because the processes and methods necessary to evaluate a CADe device that is intended for a detection task are not the same as the processes and methods necessary to evaluate a CADx device that is intended to provide a diagnostic score for a suspicious finding that is defined by the user, the FDA said.
Vulnerabilities
Furthermore, the FDA rejected the petition’s broad statement that “[c]hanges in CAD devices that could affect safety and effectiveness are readily detected by users, licensed radiologists, through visual examination of the images that accompany the use of CAD, or routine testing.”
For an individual patient, agreement between a reader’s interpretation, CAD output, and the “true” disease state may not be routinely available or known before the CAD output leads to a misdiagnosis, potentially resulting in increased risk of injury or ineffective treatment, the FDA said.
In screening mammograms where the diagnosis of cancer is not routinely made based on the imaging data alone and a biopsy is often necessary according to standard of care, a physician will not be able to definitively distinguish between a device’s true-positive or false-positive output without subjecting that patient to biopsy, FDA noted.
Read the full response here.
Leave a Reply