Artificial Intelligence is the IT word in Medical Devices nowadays when it comes to software development. Every medical device company wants to fling it!
Medical device manufacturers who want to take advantage of the AI claim, however, need to have a good understanding of what lies ahead, especially how the regulations will affect their medical device. Despite a burst of AI standardization activity, practical guidance for medical devices is scarce.
Do you want to learn more about the do’s and don’ts of Artificial Intelligence (AI) in medical devices? And do you need a clear overview of what is currently going on regarding the definitions and regulatory framework for AI? In this blog post, we’ll share the latest AI insights and discuss:
- What exactly is AI in Medical Devices? How is it being defined today?
- Who started the discussion topic on Artificial Intelligence in Medical Devices?
- How is Europe handling it?
- And what are the obstacles on our regulatory path?
Curious about the answers? Be sure to read on!
What is Artificial Intelligence in Medical Devices?
Although Artificial Intelligence (AI) has existed for decades within the healthcare industry, under the umbrella of “improving patient outcomes”, it is only recently that we have felt the need to regulate its definition(s), and then primarily focused on algorithms related to software.
There are several definitions of AI, each with different aspects. But the focus we are on today (based on the AMA’s 2018 AI Board Report):
- Artificial intelligence represents a multitude of computational methods that produce systems that perform tasks that normally require human intelligence.
- These computational methods include, but are not limited to, automatic image recognition, natural language processing, and machine learning (ML).
- Augmented intelligence (AI), however, is a more appropriate term, reflecting the enhanced capabilities of human clinical decision-making when combined with these computational methods and systems.
To give an example, an AI diagnostic system that autonomously diagnoses diabetic retinopathy and macular edema in patients. But medical imaging, for example, is also one of the best-known AI applications.
We need to be careful about using the term AI. Currently, it’s overused and sometimes even misused, as there is no clear regulatory strategy in place so far since we are still learning a lot about AI.
Why is everyone talking about AI in Medical Devices?
In a world of continuous improvement on many levels, machine learning is currently a hot topic. It offers significant improvements for public health and clinical healthcare in the areas of detection, diagnosis, treatment decision support, etc.
The FDA (Food and Drug Administration) in the US were the first to reach out to create a regulatory definition, which was noted by many healthcare professionals.
On April 2nd, 2019, they published a discussion paper “Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) – Discussion Paper and Request for Feedback”, which kicked off discussions on the regulatory pathway for the AI claim. More information on FDA’s AI topics can be found on their website www.fda/gov.
From that point on, many people active in healthcare jumped on the topic to clarify and try to define a regulatory strategy:
- In 2019, the BSI-AAMI published a position paper on “the emergence of artificial intelligence and machine learning in healthcare: recommendations to support governance and regulation”, for example.
- After that, in 2020, a follow-up to this paper was published by the BSI-AAMI in collaboration with MHRA and FDA on “adapting regulatory frameworks and standards to ensure safety and performance“.
- This eventually resulted in the “TIR 34791 – Guidance on the Application of ISO14971 to Artificial Intelligence and Machine Learning”.
This shows that the US is making quite some progress on the topic.
What about AI-based medical devices in Europe?
The European Medical Device Regulation (MDR) does not specifically mention AI or ML, but it does, however, mention standard software requirements, including changes and validation of software algorithms, which are primarily used in AI and ML.
But next to that, Europe currently has no specifics on the regulatory use of AI/ ML.
Europe is taking a different path than the US and wishes to integrate AI/ML into a broader framework, broader than just healthcare.
Next to this broader framework, where Europe is working on, several additional requirements for medical devices will be integrated into the MDR and possibly these requirements will be drawn from the IMDRF approach.
IMDRF (the International Medical Device Regulators Forum) is well known for its regulatory harmonization work and would therefore be a logic source of information for drawing up these additional requirements. This Quality Management for AI/ ML based on the IMDRF approach is currently being established. For more information, please visit www.imdrf.org.
Europe has already started working on the aforementioned broader framework. On Wednesday 21st of April 2021, the European Commission published their proposal on “Coordinated Plan on Artificial Intelligence 2021 Review”. Their goal is to motivate innovation and enhance trust in AI/ ML.
The hurdles that still stand in the way of regulation
AI is quite unique because of its adaptability. This causes several hurdles that we do not encounter with other medical devices. What about adaptive technologies, for example? What does this mean in terms of “substantial changes” and all their implications for documentation with regards to the expectations of Notified Bodies?
The same goes for advanced capabilities that modify the software program based on live data and the additional risk of changes without appropriate change control.
And what about situations where medical device regulators have to deal with integrated practice of medicine? This has not been taken into account so far.
It comes down to what to do about validation of AI, since AI changes during use. Strategies will need to be developed to deal with these changes.
And finally, industries will need to determine an approach that is acceptable to Notified Bodies.
AI in Medical Devices: conclusion
With so much information at hand – papers, blogs, etc. – it is sometimes difficult to see the forest for the trees anymore. Not only an overload of information but also an abundance of definitions about AI leads to a blurry overview of what is actually required.
Basically, the goal of regulating AI is to get a safe and effective device. And with the FDA, IMDRF working groups, and Europe working intensively on this topic, we hope to quickly unravel the mysteries of artificial intelligence and capture them in regulatory language.
But it must be said: we are constantly learning on Continuous Learning!
Do you need help unraveling AI for your product? Or do you need help knowing exactly what this claim means for your regulatory journey, which documents are affected, etc.? QbD is happy to help! Don’t hesitate to get in touch.