Medical devices using artificial intelligence (AI) may be employed to identify illnesses and persons at risk of developing medical conditions. They can do many time-consuming tasks on behalf of physicians and radiologists in order to quicken the diagnosis of conditions. Quicker diagnoses allows patients to get treatment quickly while it is most effective. They could also help to determine the most helpful treatments including personalized medicine.
At present, the U.S. Food & Drug Administration (FDA) reviews medical devices before granting market authorization. In general, the algorithms the medical devices use must be locked and should not be learning every time they are employed to pass the market authorization process.
Because of the locked algorithms, developers need to update them eventually at intervals using new information. Nonetheless the updated devices will still be manually reviewed to validate the updated algorithm.
In 2018, the FDA certified two medical devices using AI: one can identify diabetic retinopathy and the other can alert providers when patients will possibly have a stroke. The FDA foresees the development of more devices to be used in healthcare which demands the finalization of the review process.
In healthcare, the potential is tremendous if adaptive algorithms are continuously updated instead of being periodically updated. With adaptive algorithms, medical devices learn from new information as they are used in the real world and get better as time passes. For example, algorithms may be used to recognize cancerous lesions. Adaptive algorithms can learn to enhance the level of confidence in discovering cancerous lesions and can potentially recognize several sub-types of cancer depending on real-world reviews.
The FDA is trying to create a regulatory framework so that AI-based medical devices can be approved for use which integrate machine learning and is thinking about reducing prohibitions on adaptive algorithms. To begin that process, the FDA published a discussion paper about the brand new framework for the medical devices using AI on April 2, 2019.
The framework is influenced by the
- benefit-risk framework of FDA
- International Medical Device Regulators Forum risk classification
- risk management guidelines of the software
- device manufacturer’s complete product life cycle
In some instances, it would be essential for the device manufacturers to make a new submission to the FDA to get further approval, however generally speaking, the framework will not require further reviews for updates to be done via their adaptive algorithms.
The discussion paper only outlines the FDA’s plans and is not considered as guidance. It talks about medical devices using adaptive algorithms and shows the appreciation of the FDA on the present software regulatory framework that seeks to improve medical devices.
See the FDA’s PDF document entitled Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device for which the FDA is requesting feedback.