Managed services
Justitia.Swiss: a secure digital justice system for Switzerland
Learn more
Government & public sector
Smarter medical device classification, powered by AI
A national health authority in Southeast Asia partnered with Zühlke to prototype an AI-driven solution for medical device risk classification – proving that AI can support faster, more consistent regulatory decisions without compromising the rigour that public health demands.
Collaboration overview
- A national regulator needed to modernise a manual, time-intensive process for classifying the risk of medical devices entering its market.
- Zühlke delivered a machine learning-powered prototype that classifies device risk with 92% accuracy, exceeding the initial 80% target.
- The prototype validated the viability of AI in a highly regulated public health environment, laying the groundwork for broader automation.
The challenge: classification that protects public health
Every medical device entering the market – from wheelchairs to pacemakers – must be assessed and assigned a risk class before it can be approved. Getting this right is a direct safeguard for public health.
The authority's classification process relied entirely on manual review. While thorough, it was time-intensive, difficult to scale as submission volumes grew, and vulnerable to inconsistency and human error. The question was clear: could AI help reviewers work faster and more consistently, while maintaining the standards their mandate requires?
The approach: rigorous AI for a regulated environment
The client engaged Zühlke through the AWS Marketplace, enabling a fast start while meeting public sector procurement requirements. Together, the team built a proof-of-concept solution trained on publicly available regulatory data and the client's own historical submissions.
The platform leverages machine learning (ML) and deep learning (DL) techniques, including large language models (LLMs), to assess and classify the risk of medical devices based on information provided by the distributor.
Traceability was built in from the start. The model highlights which input terms most strongly influenced each classification decision, so every result can be inspected, explained, and validated.
“Medical device classification is highly regulated for good reason, which is why these decisions have always been handled with great care and human judgment. Our goal wasn't to replace that rigour, but to support it. Introducing AI into this environment requires rigorous testing, strong governance, and deep domain expertise.”
The results: a proven foundation for responsible AI adoption
The prototype achieved the following outcomes during conservative rigorous testing:
- 92% classification accuracy, exceeding the 80% benchmark target.
- Full traceability of every classification decision.
- Secure-by-design infrastructure aligned with government cloud compliance requirements.
- Validated architecture, reviewed by the client's internal oversight committee and an AWS Solutions Architect.
The prototype didn't just deliver a technical proof point – it built the confidence and governance foundations the authority needs to expand AI-assisted decision-making across other regulatory workflows. Always with the same commitment to rigour, transparency, and public safety.
Explore how Zühlke can help you translate emerging AI capabilities into practical, trusted solutions.
“When a device is described as being 'intended for surgery', the term 'surgery' carries greater weight in determining its risk class. This level of transparency allows classification decisions to be inspected, explained, and validated.”
