What It Takes to Bring an AI-Enabled Medical Device to Market: The Testing Journey
The first wave of connected devices changed what devices could do. The second wave — the one being built now — adds an AI model to the hardware. And the third requirement, the one that separates a prototype from a regulated medical product, is evidence: the full testing journey that proves the device is safe, effective, and traceable.
Putting an AI model inside a medical device does not just add features. It adds a second compliance burden on top of the one every medical device already carries: hardware safety, electromagnetic compatibility, materials, usability, clinical evidence — and now, software and AI validation that regulators will examine line by line.
The device is the sum of its tests
Before any AI is discussed, the hardware itself must pass the standard battery that applies to active medical devices:
- Electrical safety & EMC (IEC 60601-1 / 60601-1-2) — leakage, insulation, applied parts, and electromagnetic immunity and emissions for a battery-powered device used close to the body.
- Biocompatibility (ISO 10993) — every surface that contacts the user, and gas-path testing (ISO 18562) where airflow reaches the user.
- Usability engineering (IEC 62366-1) — the device must be hard to use incorrectly, because use error is a safety event.
- Risk management (ISO 14971) — a documented, traceable analysis of every failure mode, including single-fault scenarios.
None of this is new. What is new is the AI layer sitting on top.
The AI layer is a compliance exercise of its own
Regulators treat the AI software as part of the device — not as an add-on. That means:
- Software lifecycle (IEC 62304) — full documentation of design, verification, and change control, graded by the software's safety class.
- SaMD risk framework (IMDRF) — the intended use and risk of the software determine how much evidence is expected.
- Model validation — dataset governance, training/validation/test separation, and performance metrics documented before submission.
- Locked vs. adaptive models — a model that updates itself after launch triggers a re-assessment. Most first-generation devices ship a locked model for exactly this reason.
Air movement is safety-critical
Where a device moves air — a pump, a fan, a flow path — the pump becomes one of the highest-energy components. Performance validation, endurance, leak and overpressure testing all apply, and fail-safe behavior must be verified under single-fault conditions: the safety chain has to work even if firmware fails.
Clinical evidence and cybersecurity
Two more gates are easy to underestimate. Clinical evaluation (ISO 14155 / TGA Essential Principle 14) requires evidence that the benefits outweigh the risks for the intended use — literature, and where appropriate, study data. And cybersecurity (IEC 81001-5-1) is a documented requirement for any device that connects wirelessly or touches the cloud, alongside privacy obligations such as the Australian Privacy Principles for data collected through connected features.
A plan, not a promise
This is why a certified manufacturer is the right partner for an AI-enabled device: the quality system that produces the paperwork is the same one that produces the hardware. At DMT, our planned evaluation program covers air-pump safety, AI software compliance, electrical safety and EMC, clinical evaluation, cybersecurity and data privacy, and materials and usability — each with named standards, documented under our ISO 13485 QMS, and executed with accredited laboratory partners.
Testing is not the end of development. It is the evidence that development was done properly — and for an AI-enabled medical device, that evidence is the product's license to reach its users.
Explore the TCM-1 Intelligence Stack → · Confidential discussions under NDA.
Sources
Editorial content based on DMT engineering practice. Standards referenced: IEC 60601-1 / -1-2, IEC 62304, ISO 14971, ISO 10993, ISO 18562, ISO 14155, IEC 62366-1, IEC 81001-5-1, IMDRF SaMD framework. Verify current editions with your regulatory advisor.
Planning a regulated-market device with an AI layer? Talk to an ISO 13485-certified partner.
Request OEM / ODM Quote →