Radiology artificial intelligence firm asks FDA to exempt certain devices from premarket review

Harrison.ai is asking for optional premarket flexibility for radiology computer-aided detection devices, in exchange for post-marketing monitoring, ACR noted in a previous news update about the petition. This would be limited to eligible manufacturers of computer-aided detection, diagnosis and triage devices. Under the proposal, the manufacturer would need to continue satisfying the FDA’s labeling and design requirements. It also would have to adhere to new special controls tied to monitoring performance in the real world, including post-market monitoring and limiting distribution to facilities with appropriate training programs. 

Although filed as a citizen petition, the agency is processing the request under the Food, Drug and Cosmetic Act, requiring the FDA to publish a notice and rapidly issue a final order. If the agency grants the petition, the final order will directly amend the classification regulation without conventional notice-and-comment rulemaking, ACR noted. 

With the proposal, any manufacturer who already has at least one FDA-cleared, CAD-type device could launch a similar computer-aided detection offering without submitting a 510(k) premarket notification. This could offer companies like Harrison.ai more leeway when choosing testing methodologies to satisfy FDA requirements. 

“By moving some evaluation to post-market, the exemption proposal described in this petition will allow manufacturers more flexibility in picking appropriate, least burdensome testing methodologies that still satisfy the special controls,” the vendor wrote. “For example, manufacturers may choose to validate their technologies using multi-finding studies that are more reflective of clinical practice.” 

More responses

As of Friday, the FDA had received nearly 40 comments on the petition, from patients, provider groups and radiology AI vendors. 

RSNA said Friday that it opposes the exemption, seeing it as an “inadequate and potentially harmful solution.” 

“The regulatory checkpoints currently in place serve a critical function by requiring rigorous review, validation and oversight—processes that are pivotal to maintaining high standards of patient safety and clinical care,” Carolyn C. Meltzer, MD, chair of the RSNA board, wrote Feb. 27. “Removing these checkpoints could result in devices entering the market without sufficient evidence of their reliability or clinical benefit, potentially exposing patients to harm and undermining physician confidence in AI-assisted diagnostic tools.”

The Patient Alliance for Safe Imaging also wrote Friday, emphasizing it “forcefully opposes the petition.” Patients and their families are the ones who usually suffer when a diagnosis is delayed or missed. 

“To treat complex artificial intelligence software as if it were a simple Class 1 device exempt from premarket scrutiny is a dangerous gamble with patient lives,” the group wrote Feb. 27. 

The Coalition for Health AI—which represents nearly 3,000 organizations including 200-plus health systems—similarly said it strongly opposes the petition. 

“While we share the petitioner’s desire to foster innovation and reduce regulatory burdens, we believe that removing the 510(k) guardrails for these specific technologies presents a significant risk to patient safety and the long-term viability of the medical AI industry,” the group wrote Feb. 27. “The FDA’s current framework for Software as a Medical Device (SaMD) is working. It successfully filters out under-performing algorithms that could harm patients,” it added later. 

Meanwhile, multiple radiology AI vendors also wrote in to voice their support for the request, including Lunit and Quibim. 

“For experienced manufacturers, repeating the 510(k) process for every minor software update or new lesion type provides diminishing returns for safety while drastically increasing costs and delaying patient access,” Lunit said in its comments. “And this proposal would effectively prevent the indiscriminate market entry of unproven manufacturers while facilitating the rapid deployment of state-of-the-art AI models from companies with established track records.” 

Quibim added that it also endorses this initiative, provided it is accompanied by enhanced safeguards including mandatory transparency measures and qualified end-user training. 

“The current regulatory pathway has led to fragmentation in the U.S. AI radiology market, with manufacturers compelled to pursue narrow, single‑finding clearances that do not reflect the clinical reality of multi‑finding, cross‑organ radiology interpretation,” the company said. “As a result, American clinicians frequently receive tools that address isolated tasks while more comprehensive, integrated systems—already available in Europe, the U.K., Australia, and other regulated markets—remain inaccessible due to regulatory friction rather than scientific limitations.”

RevealDx, which recently earned FDA clearance for one of its products in February, said it opposes the petition. It noted that diagnosis-based products should not receive such leeway and contends certain such exemptions could potentially be “dangerous to the public.” 

“We recommend that the agency reject this petition in its entirety,” the vendor wrote. “AI for medical imaging is rapidly evolving, post market surveillance effectiveness for these products is not well established, and an exception for any of these products is not in the best interest of patient safety.”

ACR echoed some of these concerns when sharing findings from its recent survey of radiologists. Physicians are worried this partial exemption would redirect responsibility for assuring device safety to users. It would do so “without any corresponding change in payment policy or reimbursement to support the additional work required.” 

“Any regulatory framework that reduces premarket FDA review should therefore consider the practical burdens placed on clinical sites and the uneven capacity across the delivery system to manage them,” the ACR wrote. 

In a Feb. 26 news update, ACR said the FDA will follow a fast statutory review timeline, with expectations it will issue a decision “in the next few months.”