Machine Learning in MedTech

When Signals Become Movement

In advanced prosthetic solutions, control is not based on traditional buttons or interfaces. Instead, biosignals are used. Weak electrical signals from muscles and the nervous system.

These signals are neither clean nor easy to interpret. They vary between individuals, change over time and are affected by everything from muscle fatigue to electrode placement.

This is where Machine Learning comes in. By analysing large amounts of signal data, algorithms can be trained to recognise patterns associated with different intentions: bending an elbow, opening a hand or gripping an object. The result is a system that does more than simply respond to signals. It begins to interpret them.

But this requires more than model training. It requires an understanding of how signals are generated in the body, how they are affected by sensors and how they behave in real-world use outside the laboratory.

Technology That Has to Work Outside the Lab

Jan Zbinden, PhD in Bionics and Technical Consultant at Together Tech, has a background spanning mechanics, electronics, software and research in bionics. For several years, he worked in multidisciplinary teams where engineers collaborated closely with surgeons, physiotherapists and patients to develop the next generation of prosthetics.

“What quickly becomes clear in these types of projects is that it’s not enough for something to work in a controlled environment. It has to work when a patient goes home, when the hand is used in everyday situations and when the signals are no longer ‘perfect’,” says Jan Zbinden.

Machine Learning in MedTech: not more AI, but the right AI

In the broader technology discussion, AI is often described as a universal solution. In medical technology, reality looks different. This is not about large generative models. Instead, it is about controlled, verifiable and robust Machine Learning models that can be tested, understood and quality-assured.

Methods such as supervised learning are often used to train systems on annotated signals, while reinforcement learning can be used in more adaptive control systems.

What they have in common is that they need to be:

• traceable
• validatable
• stable over time
• safe for clinical use

“When patients’ health and lives are at stake, there is no room for ‘black box’ solutions that cannot be explained or verified,” Jan continues.

Where multiple engineering disciplines meet

What makes medical technology unique is that it is rarely about a single technical discipline. A modern medical device can combine mechanics, electronics, sensor technology, embedded software, signal processing and Machine Learning in one and the same product.

That is also why multidisciplinary expertise is essential. Jan Zbinden’s experience from both research and industry reflects this reality. Being able to move between hardware and software, between signals and systems, and between theory and practical implementation is often what determines whether a product makes it all the way to real-world use.

From research to products people can actually use

Developing a working prototype is one thing. Creating a product that patients can use in their everyday lives is something completely different. Too many innovations get stuck in the gap between research and productisation.

“At Together Tech, we work to close that gap. We help our customers take technologies from ideas and research environments to robust, verified and usable products,” says Håkan Rolin, MedTech Business Manager at Together Tech.

This can include:

• signal processing and biosensors
• embedded systems and real-time control
• Machine Learning for classification and prediction
• hardware and software integration
• testability and validation in regulated environments

The goal is always the same: technology that works in the real world, not just in theory.

When technology becomes truly meaningful

There is a recurring insight in projects where people and technology meet in this way: the complexity increases dramatically, but so does the sense of purpose.

“When a technical solution makes it possible for a person to regain function, independence or mobility, every line of code and every sensor placement becomes meaningful. It is incredibly motivating and what drives me in my work,” Jan Zbinden concludes.

FAQ 

What does Machine Learning in medical technology mean?

Machine Learning is used to analyse medical and physiological data, such as biosignals, and create models that can interpret, classify or predict user behaviour or physiological conditions.

Why not just use “AI”?

Medical technology often requires controlled and verifiable models. Machine Learning models are generally easier to validate and certify than generative AI models.

What are biosignals in this context?

Biosignals are electrical or physiological signals from the body, such as muscle activity or nerve signals, that can be measured and used as input to medical technology systems.

What are the main challenges?

One of the biggest challenges in Machine Learning for MedTech is access to relevant and reliable data. Algorithms can only learn from information that can be measured.

In prosthetics development, the challenge can even be that the available signals do not contain the information needed. Machine Learning only becomes an effective tool once the right signals can be captured and measured.

How does Together Tech work in this area?

We combine expertise in Machine Learning, embedded systems, sensor technology and product development to help customers develop medical technology solutions that work in real-world applications.

Would you like to explore how this could be used in your products?

Together Tech helps companies move from research-driven ideas to verified medical technology products where Machine Learning is used in a safe, robust and meaningful way.