AI Developments in the medical devices industry
The latest generation of AI is making a lot of noise in the press at the moment, but away from the hype, it promises to bring real-world benefits to users, medical professionals and the healthcare industry as a whole. The application of AI (Artificial Intelligence) and ML (Machine Learning) in the healthcare industry has the potential to bring great advances in the diagnosis and treatment of patients, accelerating the drug discovery process, whilst bringing expert-level knowledge and analysis to the wearable device and medical equipment market.
Of course, there are some downsides and concerns in using AI: the predictability and accuracy of the results it generates and the verification of the “decision-making” process in a medical context. This requires a careful and thoughtful approach if you are considering incorporating AI into your products or services.
Advantages of incorporating AI in medical devices
Embedded processors used in medical devices are becoming more and more powerful, this gives the opportunity of moving AI to the edge of the computing pathway. This means that processing and identification is performed close to where the data is gathered and can be beneficial in multiple ways. For example:
- generate early information of identified issues. An example of this would be the arterial fibrillation detection being performed in most modern smart watches.
- reduction in the amount of data that is passed from the edge device. As the data is processed on the device, only significant, identified events need to be passed back to a central point.
- If data is processed ‘on device’, then connectivity to a central processing module could be eliminated, thereby simplifying the device.
So, what are the risks of utilising AI?
In the press, we have already seen reports of generative AI creating plausible sounding but completely inaccurate responses to questions asked to them. At their essence, AI and ML solutions are very complex filters through which data is passed to identify specific patterns. AI systems can adapt their training so that they ‘learn’ and apply better “filters” to required data resulting in more accurate identification of the required pattern.
The training models used in AI products must utilise a dataset of input material that encompasses the expected variety of inputs that the system is likely to analyse. For example, an AI-based glaucoma diagnosis system must have training material that shows different coloured eyes, eyes with cataracts, macular degeneration, and other diseases/abnormalities. In addition, the system must be trained with enough images that have both positive and negative outcomes for the AI System to accurately establish the positive pattern that is required to be detected. This makes building the training models a data-intensive and sometimes lengthy process and there is always the risk that edge cases might be overlooked.
Using self-training AI in medical devices has its risks; one being that historical data-processing may affect the results that each device generates, in an uncontrolled manner. Therefore, devices would have divergent performances depending on what self-learning had been performed.
One possible solution to this for embedded medical devices would be to restrict the device to using a fixed training model.
Practical examples of AI in medical devices
One of the benefits of incorporating AI is the ability to ‘build in’ experience from experts and practitioners to the onboard data-processing capabilities of your medical device. This leads to the device being used by a wider group of users but retaining the skills normally associated with a highly trained user. For example, a Nurse practitioner may be able to use a scanning device normally used by a consultant.
Software libraries such as TinyML and Tensorflow Lite now exist to allow AI models to run in the embedded world. This combined with modern high-performance AI-focussed processors such as the Jetson Nano and the Coral AI processors now make embedded AI a very practical scenario.
Some scenarios where the application of AI could be beneficial are:
- Medical scanners which incorporate diagnosing algorithms to automatically detect medical conditions whilst a patient is being scanned.
- Drug discovery algorithms being enhanced by AI.
- Wearable heart monitoring systems to detect the early stages of heart conditions.
- Patient data analysis systems using best practices to flag early warning signs to a general practitioner. For example, combining elevated levels detected in a blood test with family history with to identify potential concerns.
Medical device approval considerations when using AI
The development of medical devices is a highly regulated process and there are numerous standards that must be adhered to. The use of AI in medical devices is a new and rapidly expanding inclusion to the development process and has rightly attracted the attention of the regulatory bodies to provide guidance.
The FDA has released an action plan on what it believes the regulation should include, with the highlights of the report categorised into five main areas:
- A tailored regulatory framework for AI-/ML-based Software as a medical device
- Good machine learning practice
- Patient-centric approach incorporating transparency to users.
- Regulatory science methods related to Algorithm Bias and robustness.
- Real world performance
An update to the regulatory framework is in progress by the FDA, but this is a rapidly evolving area and as such further updates to the regulations are to be expected, in order to keep pace with developments in AI technology.
eg technology and medical device development
eg technology are carefully monitoring the development of AI and the legislative frameworks that are being put in place for the incorporation of AI into the medical software and device development processes.
With our background in developing medical devices, we understand the rigor and necessary procedures required to bring a successful medical product to market. Going forward, the use of AI components ‘in device’ (or cloud software for healthcare systems), will undoubtedly need to comply with regulations and incorporate thorough risk analysis and real-world processing examples. This will ensure that the AI element is consistent and reproducible across a variety of expected scenarios.
eg technology are abreast of the regulatory developments within the AI field especially in the use of AI within a medical software environment. The use of AI looks to be an exciting technology that will undoubtedly help us create novel and powerful products for our customers