Understanding FDA regulations for AI in medical devices

In recent years the term artificial intelligence (AI) has become ubiquitous across multiple sectors including medicine. The first AI-enabled device received FDA approval in 1995, but it wasn’t until 2015 that it became more widely adopted. Between 2015 and March 2025 the FDA approved more than 1,000 AI-enabled devices. In this article we look at regulatory considerations for AI in software as a medical device.

What is AI?

AI refers to computer algorithms capable of complex reasoning, decision-making and even learning, which mimic human intelligence artificially. The intention of AI is to replicate human decision-making by rapidly evaluating complex and large quantities of data to improve the speed, consistency and accuracy of decision outcomes for the designed application. Perhaps AI’s greatest benefit: the ability to learn from real-world use and experience to improve its own performance.

How AI is being deployed in healthcare

AI-enabled medical devices are used most extensively in the field of imaging, specifically radiology. AI in medical imaging is integrated to quickly process large quantities of data and assist physicians to identify anomalies to better inform a patients’ diagnosis and/or treatment regime. Indeed, many of the recently approved AI-enabled medical devices with imaging applications are designed to detect and flag anomalies within a given type of imaging mode and within the specified indication (i.e. identifying an abnormal growth within a mammogram).

The second highest number of AI-enabled devices are in cardiovascular. The FDA recently gave De Novo authorisation for an at-home blood pressure monitor that simultaneously reads blood pressure and detects AFib. The device uses AI algorithms to read the pressure waves and detect irregularities. Robotics is an emerging field for AI-enabled devices, especially to aid preparing for, conducting or following up for surgery.

FDA regulatory considerations for embedding AI in medical devices

Despite the increasing number of AI-enabled medical devices the FDA has yet to establish a unique regulatory pathway for these devices. Most approvals follow a 510(k) pathway according to their risk level and similarity to a predicate device. The remaining approvals have been granted on a de novo basis. The FDA has recognised that traditional medical device regulations were not designed for AI-enabled devices and issued several guidance documents on regulatory submissions for the use of AI. The guidance emerged from joint discussions and research between multiple divisions of the FDA and feedback gathered outside the FDA.

The most recent FDA draft guidance from January 2025 proposes lifecycle management considerations and specific recommendations to support marketing submissions for AI-enabled medical devices.

As of August 2024 the FDA cleared 97% of AI-enabled devices via the 510(k) pathway. To follow this pathway applicants must prove their low to moderate risk device is substantially similar to a device already approved by the FDA (predicate device). During the same period, 22 AI-enabled low to moderate devices with no predicate went through de novo classification. Only four devices required the most rigorous premarket approval pathway as high risk devices.

For clinical decision support, if an AI-enabled device is used to make specific recommendations around a diagnosis or treatment then it falls under regulations. It is exempt from regulations if the AI-enabled device matches patient data with current treatment guidelines for common illnesses.

The FDA considers other factors when evaluating safety and effectiveness of algorithms in AI-enabled devices. These include data quality, robustness and clinical performance. AI-enabled devices must be validated and include an evaluation of appropriate study diversity based on the devices intended use and technological characteristics.

Conclusion

FDA approvals of AI in medical devices have increased rapidly in the past decade. AI offers medical device researchers new possibilities for innovation, improved accuracy and efficiency. Regulators have yet to catch up with the speed of adoption, and for now regulatory pathways depend on the level of associated risk and whether a predicate device exists. In early 2025 the FDA released draft guidance containing recommendations and considerations linked to the AI lifecycle. In the absence of a distinct regulatory pathway for AI-enabled devices, medical device researchers must plan their regulatory route to meet current requirements and anticipate future ones.

Download our whitepaper for a detailed examination of AI in SaMD and FDA regulations.

What is AI?

How AI is being deployed in healthcare

FDA regulatory considerations for embedding AI in medical devices

Conclusion

In this section

Connect with us