Final Determination on Premarket Notification Exemption
On September 17, 2026, the US Food and Drug Administration (FDA) published a final order in the Federal Register (91 FR 58817, Document 2026-19074) officially denying a citizen petition requesting partial exemption from premarket notification requirements (510(k)) for radiology artificial intelligence software. The order confirms that radiology computer-aided detection (CADe), computer-assisted diagnosis (CADx), and computer-aided triage and notification (CADt) devices remain subject to mandatory 510(k) premarket clearance before commercial distribution in the United States.
The petition (Docket No. FDA-2025-P-5560), submitted on behalf of Harrison.ai pursuant to section 510(m)(2) of the Federal Food, Drug, and Cosmetic Act (FD&C Act), requested that manufacturers who had previously obtained at least one 510(k) clearance in these classification categories be partially exempt from submitting subsequent 510(k) applications, provided they maintained robust post-market surveillance, transparency, and adherence to special controls.
Regulated Device Categories Retaining 510(k) Oversight
The denial applies comprehensively across four generic device classifications under 21 CFR Part 892:
- CADx software: Radiological computer-assisted diagnostic software for lesions suspicious of cancer (21 CFR § 892.2060, product code POK).
- CADe devices: Medical image analyzers (21 CFR § 892.2070, product code MYN).
- CADt software: Radiological computer-aided triage and notification software (21 CFR § 892.2080, product codes QAS and QFM).
- CADe/x software: Radiological computer-assisted detection and diagnosis software (21 CFR § 892.2090, product codes QBS and QDQ).
All four categories must continue to submit 510(k) notifications demonstrating substantial equivalence and compliance with applicable special controls.
FDA Evaluation Rationale
Under established CDRH criteria for Class II exemptions (63 FR 3142), an exemption is only justified when general controls alone provide reasonable assurance of safety and effectiveness, device performance characteristics are well established, and device modifications are readily detectable before causing patient harm.
FDA determined that the petition failed to meet these statutory thresholds for radiology AI/ML algorithms:
- High sensitivity to acquisition variation: Machine learning algorithms depend heavily on training data distribution and can degrade unpredictably across varying scanner hardware, acquisition protocols, contrast phases, and patient demographics.
- Inherent latency of clinical detection: Erroneous outputs—such as false-negative omissions or triage re-prioritization errors—cannot be reliably detected by clinicians prior to potential diagnostic delay or misdiagnosis.
- Indispensability of premarket validation: Independent multi-center clinical validation datasets, reader performance studies (MRMC), and Section 524B cybersecurity controls remain critical regulatory checkpoints that cannot be adequately substituted by general post-market controls alone.
Practical Guidance for Radiology AI Developers
Medical device software developers and regulatory affairs teams should consider the following baseline practices:
- Eliminate deregulation assumptions: Project planning and budget allocations for radiology AI and triage notification algorithms must account for a standard 510(k) premarket submission cycle.
- Standardize multi-center clinical validation: Clinical study designs must rigorously follow 21 CFR Part 892 special controls, incorporating independent test datasets across multiple clinical imaging sites with verified reference standards.
- Structure algorithm modifications via PCCP: For planned post-clearance model refinements, utilize Predetermined Change Control Plans (PCCP) within the original 510(k) to establish pre-cleared boundaries for incremental algorithm updates.
Sources
- Primary official sourceMedical Devices; Exemption From Premarket Notification: Radiology Computer-Aided Detection and/or Diagnosis Devices and Computer-Aided Triage and Notification Devices (opens in a new tab)Food and Drug Administration