Artificial Intelligence and Machine Learning in Software as a Medical Device
Medicine has long depended on devices. From surgical tools used during antiquity to the sophisticated imaging devices used to diagnose a range of medical problems, medical devices are invaluable tools used to aid physicians and other medical practitioners. However, regulatory agencies around the world are grappling with a relatively new addition to the medical device landscape: Software. The Food and Drug Administration, in particular, is still working through new methods of ensuring devices used are both effective and safe, raising a host of questions. How should these devices be regulated? What is the FDA’s role in ensuring optimal patient outcomes without stifling innovation? A cornerstone of its effort involves Software as a Medical Device (SaMD).
What is SaMD?
The FDA adopts its definition of SaMD from the International Medical Device Regulators Forum: “Software intended to be used for one or more medical purposes that perform these purposes without being part of a hardware medical device.” While the software has long played a role in powering imaging technology and other devices, software that does more than power specific hardware is relatively new, and it was previously referred to simply as standalone software or health software, terms which can be ambiguous and lead to confusion. SaMD covers several use cases: Various types of virtual networks might qualify, as do medical device platforms that go beyond a single device or even type of device. With this new categorization, the FDA aims to provide clarity for all involved parties.
Why is SaMD Necessary?
One of the prime drivers of the FDA’s move to properly regulate SaMD is the rise of new types of technology in medicine. Artificial intelligence has long shown promise in aiding doctors in diagnosing and treating patients, and this promise is growing at an accelerating rate. Similarly, machine learning is providing new methods of diagnosing patients and finding appropriate treatments based on empirical data, letting technology separate weak but clinically important signals from what looks to be noise to people. The medical device industry is capable of providing potentially revolutionary new methods of improving patient outcomes, but it needs a proper regulatory framework.
Medicine and Software in Conflict?
Medicine moves slowly, focusing on ensuring safety and properly testing devices before they’re used to treat patients. Contemporary software development, on the other hand, prefers fast-paced, iterative development, and a high tolerance for small bugs and other setbacks along the way. The FDA acknowledges the potential for rapid improvements in SaMD, but it also needs to ensure software problems tolerable in other fields don’t lead to potentially harmful results for patients. Much of the conversation surrounding SaMD involves finding the right balance between timely software iteration and keeping the high standards of medical devices intact.
Hear From Industry Experts
Read the latest tips, research, best practices, and insights from our community of expert B2B service providers.
SaMD is Already Here
Although the SaMD regulatory framework is still undergoing discussion, the FDA has already approved certain software. In early 2020, the FDA approved marketing for software designed to help automate echocardiography images, which lets medical professionals use these devices without the extensive training usually required. Even though the process for regulating SaMD devices is still under discussion, the FDA approved use through a de novo request, which focuses heavily on the potential risks involved. In doing so, the administration created a new classification of devices. Even though more thorough regulations are pending, the FDA has shown a willingness to permit new SaMD’s to come to market if they can qualify under existing frameworks.
AI and ML have already been used to refine existing types of devices. However, much of the excitement involves using these technologies in the devices themselves. Through proposed regulations, the FDA plans to seek out two aspects when determining whether a device is effective and safe. The SaMD Pre-Specifications portion involves looking at what aspects of the device will use AI and ML, while the Algorithm Change Protocol deals with how the algorithm evolves. The second portion is where the difficulty truly lies, as evolving software might make decisions that aren’t quite as easy to predict as with previous technologies.
Crafting New Standards
Also of interest to the FDA is fostering an environment that encourages safe practices as medical device companies bring new devices to the market. In discussing the principle of Good Machine Learning Practice, the FDA looked toward existing standards considered to be good software design practices. For example, strong documentation is often hailed as a mark of high-quality software, and the FDA intends to encourage device manufacturers to ensure their code is appropriately described. Furthermore, effective data management is critical to protecting patient privacy while ensuring bugs don’t occur as a result of poor data management. These goals aim to make the software development element of SaMD devices meet the high standards the FDA imposes, which can be challenging given the seemingly ever-changing nature of software development.
The FDA is also aware that it needs to maintain the public’s trust even in the face of rapid development. While some people are excited to hear that their medical providers rely on innovative new technologies, others are more reluctant to entrust their health to new software. Part of these efforts involves encouraging companies to offer high levels of transparency, both to the users and to patients. Finding out how much disclosure is needed is ongoing; medical professionals and their patients deserve to know approved devices work, but it can be difficult to explain how and why a device’s algorithms change over time. How devices will need to be labeled is also an ongoing discussion.
There’s no question that SaMD will play a key role in medicine over the coming years, as the potential benefits are simply too substantial to ignore. These devices are agile, and the FDA acknowledges to it needs to ensure its regulations are agile as well. While discussions remain ongoing, device manufacturers can expect more clear guidelines going forward, and medical professionals and patients can expect to see a steady rise in the number of devices that hit the market in the coming years.