FDA digital cervical cytology AI classification order (2026)
AI-generated text. This page was generated using artificial intelligence.
FDA digital cervical cytology AI classification order
United States — federal
Final medical-device classification order
21 CFR 864.3900; 91 FR 69687; FR Doc. 2026-22593[1]
Final order published November 5, 2026[1]
Order effective November 5, 2026; classification applicable January 31, 2024[1]
November 5, 2026 — complete final order and special controls
The FDA digital cervical cytology AI classification order codifies a class II medical-device category and special controls for prescription systems that assist readers interpreting cervical cytology images. The final order was published and became effective on November 5, 2026; FDA states that the classification was applicable on January 31, 2024, when it issued the original De Novo classification order to Hologic. The November publication adds 21 CFR 864.3900.[1]
Device category and regulatory pathway
The category covers an automated digital slide creation, viewing and management system for images prepared from Pap test specimens and conventional Pap smears. Its machine-learning algorithm selects and presents areas of interest, supplying information about their presence, location or characteristics to assist the user in reaching a cytology diagnosis. The device is a prescription in vitro diagnostic device.[1]
FDA received Hologic's De Novo request for the Genius Digital Diagnostics System with the Genius Cervical AI algorithm on August 27, 2021. The agency's January 31, 2024 order placed the device in class II, with general and special controls. The November 2026 order codifies that classification and identifies the controls for this device type. FDA explains that a De Novo classified device can serve as a predicate for future substantially equivalent devices. The order states that FDA has not exempted this device type from section 510(k) premarket notification.[1]
Labeling and human interpretation
Section 864.3900(b)(1) requires labeling to identify the specimens, inputs and outputs, algorithm outputs, intended users and patient population. Instructions must explain the image-review and interpretation protocol and how to address cybersecurity or malfunction issues. The device description must cover its components, infrastructure and configurations, compatible scanning hardware, interactions with other devices or software, and recommended user training.[1]
Labeling must summarize specified performance studies and relevant subgroup analyses, and provide screening-time information for daily workload limits. It must explain situations in which the device may fail, including limitations associated with the development datasets. Required limiting statements call for the user to review the entire gallery before diagnosis, use the device alongside the standard of care, patient history and professional guidelines, and employ safeguards such as quality control, glass-slide review and cybersecurity measures, including for remote image review.[1]
Validation and performance controls
Section 864.3900(b)(2) requires component and system performance assessments. These address imaging and display components and, as appropriate, three-dimensional cellular features, color, resolution, focus, specimen coverage, stitching errors and turnaround time.[1]
Software documentation must describe the algorithm and its development datasets, study protocols and outputs, representative training data, an independent validation dataset containing clinically relevant subsets, and system architecture with communication and security protections. Analytical studies must assess outputs and reproducibility across relevant images, operators, scanners and sites using representative and challenging specimens.[1]
Clinical validation must demonstrate that device-assisted screening and diagnosis are non-inferior to an FDA-appropriate comparator. The controls require an intended-user study with and without the algorithm, appropriate diagnostic-accuracy measures, representative diagnostic categories and processing instruments, and screening-time data supporting recommended daily maximum workloads. Human-factors studies must evaluate usability and user experience with representative users.[1]
Related coverage
- United States Federal Authorities
- AI law: healthcare
- PATIENT AI Act discussion draft (2026) — separate pending federal healthcare-AI proposal.
- AI Law Wiki News for November 2026 — original November 5 news coverage.
References
- ↑ 1.00 1.01 1.02 1.03 1.04 1.05 1.06 1.07 1.08 1.09 1.10 FDA, Medical Devices; Hematology and Pathology Devices; Classification of the Digital Cervical Cytology Slide Imaging System With Artificial Intelligence Algorithm, 91 FR 69687–69690, November 5, 2026, FR Doc. 2026-22593, docket FDA-2026-N-12384; complete final order, including codified special controls, reviewed November 5, 2026.