Machine Learning in the Medical Devices Sector

Clearly, a system whose behaviour is impossible to guarantee seems unsatisfactory from a safety or regulatory perspective. Why should we opt for an approach that uses machine learning (ML) — an important form of artificial intelligence (AI) — in which an artificial neural network learns its own rules for diagnosis or control, rather than one that is carefully engineered to specification? 

There are two primary reasons. The first is that it is virtually impossible to hand-engineer the rule-set or system design that some types of device software or functionality require. In the medical device or healthcare setting, this arises not from inadequate functional specification, but rather from the wide diversity and complexity found in human anatomy and physiology. If we are to provide the best care for individual patients, we must tailor devices, diagnostics, interventions and therapies on a patient-by-patient basis, and adapt as the patient’s state changes. This requires continuous analysis and decision-making, a potentially clear role for automation.

But, we can also identify a hidden reason; it takes the form of an increasing trend amongst the community of engineers and data scientists: machine learning is quickly becoming the de-facto approach to designing complex systems for analyzing data from sensors. In short, it is often easier to train a learning system through examples than it is to use rule-based programming. Indeed, even the use of explicit mathematical models within the design of measurement and control systems is arguably being challenged by recent developments in machine learning. This gives further impetus for us to seek ways of accommodating systems based on machine learning into regulatory frameworks: a phenomenon that might be considered a paradigm shift in the practice of engineering sensor-driven systems. 

To appreciate why systems that incorporate artificial intelligence (AI), and specifically, machine learning, might even be considered, it is helpful to gain a bit of insight into how modern AI systems are being built, and why they are changing the way in which complex software systems are being engineered. To find out more, look out for the BSI medical devices white paper ‘Recent Advancements in AI – Implications for Medical Device Technology and Certification’, which is due to be published soon.