On January 7, 2025, the U.S. Food and Drug Administration (FDA) released a draft guidance titled Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations. This document consolidates and refines the agency’s existing recommendations for the content required in marketing applications for medical devices that incorporate AI-driven software functions. The goal is to facilitate the FDA’s determination of the safety and effectiveness of these devices while ensuring clear expectations for manufacturers throughout the product lifecycle.
The FDA is actively seeking public comments on the draft guidance until April 7, 2025.
Defining AI-enabled device software functions
The FDA defines an AI-enabled device Software Function (AI-DSF) as a software component of a medical device that incorporates AI models to generate predictions or inferences based on new input data. Recognizing that AI terminology often differs between regulatory language and the broader AI community, the FDA recommends that sponsors consult its Digital Health and Artificial Intelligence Glossary when preparing marketing applications to ensure consistent and precise communication with the agency.
Key components of a premarket application
The guidance outlines the FDA’s recommendations on the content and structure of a marketing submission, providing a detailed framework for sponsors. A summary table in Appendix A of the draft guidance organizes these recommendations, which address not only premarket expectations but also certain post-market considerations.
1. Device description
Manufacturers must clearly state the integration of AI within the device and describe its function in achieving the intended use. Submissions should include details about intended users, workflow, and operational environments, as well as any calibration or configuration processes necessary to maintain device performance.
2. User interface and labeling
Sponsors should provide a thorough description of the user interface (UI), particularly if it serves as a risk control mechanism. Labeling recommendations emphasize transparency, ensuring users fully understand the AI-driven aspects of the device.
3. Risk assessment
A risk management file must be included, in line with FDA’s Premarket Software Guidance (2023). AI-specific risks, including performance monitoring, should be considered, especially for 510(k), de novo, or PMA submissions. The FDA encourages manufacturers to reference recognized industry standards, such as AAMI CR34971:2022 – Guidance on the Application of ISO 14971 to Artificial Intelligence and Machine Learning.
4. Data management
Manufacturers must provide a detailed explanation of their data management processes, including dataset characteristics for AI model development and validation. To enhance generalizability and mitigate bias, sponsors should ensure clear separation between training and test datasets and confirm that validation data represents the device’s intended use population.
5. Model description and development
Beyond the general device description, submissions should include technical specifications of the AI model and its development process. This encompasses details on training methodologies, threshold determination, and algorithmic behavior.
6. Validation
FDA requires manufacturers to demonstrate not only that the AI model meets performance criteria but also that users can effectively interact with and understand the device. Validation strategies should include both standalone performance testing and assessments of human-device interaction. The agency highlights the importance of pre-specified protocols and statistical analysis plans to ensure validation robustness. Additionally, subgroup analysis should be used to confirm the model’s reliability across different patient demographics and highlight any potential limitations.
7. Device performance monitoring
AI-enabled medical devices are uniquely susceptible to performance shifts due to changes in input data. To address this, FDA encourages manufacturers to implement proactive monitoring systems to track and manage performance changes post-commercialization. While performance monitoring is generally a post-market requirement, the agency notes that, in some cases, providing monitoring details in a 510(k) submission may be beneficial, and such details may be mandatory for de novo and PMA applications.
8. Cybersecurity considerations
The guidance reinforces existing cybersecurity recommendations, emphasizing the need for robust security measures to address AI-specific risks. Sponsors are advised to follow FDA’s 2023 guidance on Cybersecurity in Medical Devices to ensure compliance.
9. Public submission summary
To promote transparency, the FDA recommends specific content for publicly available submission summaries, such as 510(k) summaries. Appendix F of the guidance provides an example of a Model Card that can be included in a 510(k) summary to clarify model functionality for end users.
Implications for AI-enabled medical devices
This draft guidance is expected to enhance consistency in AI-DSF marketing applications and foster more effective interactions between sponsors and the FDA. By compiling and streamlining prior recommendations, the FDA aims to provide greater clarity without imposing new regulatory burdens. The agency also acknowledges the rapid pace of AI advancement and remains committed to collaborating with industry stakeholders to adapt regulatory frameworks accordingly.
Next steps
Stakeholders are encouraged to submit comments on the draft guidance by April 7, 2025.
At Aura Health, we specialize in ensuring compliance with evolving regulatory requirements for AI-enabled medical devices. If you have questions regarding the FDA’s AI-DSF guidance or need support in preparing a marketing submission, our team is ready to assist.
References
U.S. Food and Drug Administration. (2025, January 7). Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations (Draft Guidance). Retrieved from FDA.gov
Association for the Advancement of Medical Instrumentation. (2022). AAMI CR34971:2022 – Guidance on the Application of ISO 14971 to Artificial Intelligence and Machine Learning. Retrieved from AAMI.org
U.S. Food and Drug Administration. (2024, December). Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions (Guidance for Industry and FDA Staff). Retrieved from FDA.gov
U.S. Food and Drug Administration. (2025, January 7). FDA Issues Comprehensive Draft Guidance for Developers of Artificial Intelligence-Enabled Medical Devices (Press Release). Retrieved from FDA.gov
Association for the Advancement of Medical Instrumentation. (2023, January 31). FDA Recognizes First AI-Focused Document, AAMI CR34971:2022, in List of Consensus Standards. Retrieved from AAMI.org