Share this article on:
AI-based medical devices can be used to identify diseases and individuals at risk of developing medical conditions. They can perform a great deal of time-consuming work on behalf of doctors and radiologists and can help to speed up the diagnosis of diseases. Faster diagnoses mean patients can receive treatment more quickly at a time when it is most likely to be effective. They can also help to identify the most effective treatments to allow personalized medicine to be provided.
Currently, the U.S. Food & Drug Administration (FDA) performs reviews of medical devices as part of its market authorization processes. Generally, in order to be granted market authorization the algorithms used by the devices need to be locked and not have the ability to learn each time they are used.
These locked algorithms can be subsequently updated by developers at intervals using new data, but after those updates have been applied, the devices need to be subjected to a further manual review and the updated algorithm must be validated.
The FDA authorized two AI-based medical devices in 2018: An AI-based device which can detect diabetic retinopathy and another that can generate alerts for providers of potential strokes in patients. The FDA anticipates there will be many more such devices developed for use in healthcare and is looking to formalize the review process.
In healthcare, there is tremendous potential for adaptive algorithms that continuously update rather than those that require periodic developer updates. Adaptive algorithms learn from new data through real world use and get better over time.
These algorithms could, for example, be used to identify cancerous lesions. Adaptive algorithms could learn to improve the level of confidence in detections of cancerous lesions and could potentially identify different sub-types of cancer based on real-world feedback.
The FDA is looking to develop a regulatory framework that will allow AI-based medical devices that incorporate machine learning to be authorized for use and for restrictions on adaptive algorithms to be eased. To start that process, the FDA has released a discussion paper outlining a proposed framework for AI-based medical devices.
The framework is based on the FDA’s benefit-risk framework, the International Medical Device Regulators Forum risk categorization, the risk management principles of the software and the device manufacturer’s total product lifecycle.
In certain situations, it would be necessary for the device makers to provide the FDA with a new submission and obtain additional approval, but in general, the framework would not require additional reviews to be conducted for updates to the devices made through their adaptive algorithms.
The document is only a discussion paper that outlines the FDA’s thinking. It doesn’t count as guidance, but it does start a conversation about medical devices that use adaptive algorithms and shows the FDA appreciates that its current regulatory framework for software-as-a-medical device needs to change.
The FDA has detailed its proposal in the PDF document: Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device. The FDA has requested feedback on the proposed document, which can be downloaded here.
The FDA say the document is “the foundational first step to developing a total product lifecycle approach to regulating these algorithms that use real-world data to adapt and improve.”
“As algorithms evolve, the FDA must also modernize our approach to regulating these products. We must ensure that we can continue to provide a gold standard of safety and effectiveness. We believe that guidance from the agency will help advance the development of these innovative products,” said FDA Commissioner Scott Gottlieb, M.D.