My question for the MedX Open-Source Brainstorming Event: Can someone explain the new FDA guidance for ML and AI software medical devices?
Computer vision and image processing are being widely used in the medical device industry. Hospitals and insurance companies are using big data to gain insight on their patients and help them stay healthy. Which other technologies are likely to dominate the medical industry?
Hello Ed, the FDA regulates ML in medical device application in two ways. 1) SaMD and 2)Software contained in Medical Device.
SaMD means software as a medical device. This device does not contain any hardware, only software. In the past two years FDA approved these types of software that utilize ML/AL via a process known as De Novo Request. Furthermore, FDA has two paths for SaMD a) Locked System and b) Adaptive system. Upon the approval of the Locked System, the device can only be marketed for its approved indications. However, as we all know, the system self-learns new indications and at this point becomes an adaptive system for which an additional FDA approval may be required. This is where the EU and the FDA parted ways a bit. For more info on SaMD for FDA requirements please see this link. https://www.fda.gov/media/122535/download
On the other hand, Software contained in medical device is more straight forward. These types of devices contain unique hardware where the output is interpreted by the software. These too are gaining traction in healthcare. The difference form the regulatory perspective is that the entire system has to be validated (including the hardware) and more regulatory documents in support of the FDA application must be generated. Please take a look at the FDA requirements for these types of devices. https://www.fda.gov/media/73065/download
Please let me know if you have any other questions.
Victoria
Hi Ed,
In addition to Victoria's good summary and links, I've shared a post that has visuals and over simplification of the proposed regulatory framework for modifications of AI and ML based SaMD (as shared by Victoria). See https://www.linkedin.com/feed/update/urn:li:activity:6524385035989106689