Is AI used for image analysis in a medical device always classified as high-risk under the EU AI Act?

Question

I am evaluating the impact of the EU AI Act on a medical device that automates allergy skin prick tests. The process involves pricking a patient's arm with allergens, taking an image, and using AI to suggest the diameter of any resulting wheals, which is then reviewed by a doctor.
When reviewing Article 6, I do not see how this system would classify as high-risk:
• The AI is not a safety component or the product itself, but only a part of it.
• The term 'safety component' seems to imply direct endangerment, whereas here, a wrong readout could result in a wrong diagnosis or treatment, which I interpret as an indirect risk.

However, many sources claim that any Class IIa or higher medical device using AI is automatically considered high-risk. Is there explicit guidance on this, or am I missing something in the legislation?

Dr. Oliver Eidel · Accepted Answer

As of now, there is no official or exhaustive guidance that states every Class IIa or higher medical device using AI is automatically high-risk under the EU AI Act. However, the prevailing interpretation is that if your AI functionality contributes to the intended medical purpose of the device (such as supporting diagnosis or influencing clinical decisions), it is very likely to be considered high-risk.
The legislation is somewhat vague, but the key factor is whether the AI system could impact patient safety, even indirectly. Most regulatory experts and sources err on the side of caution and treat such devices as high-risk to ensure compliance, especially since forthcoming guidance from the European Commission may clarify and reinforce this approach.
It is advisable to monitor updates from the EC, as draft guidelines on definitions and prohibited AI have been published, but nothing definitive on medical devices yet.

Answer

There are a few things to consider here. The AI could be considered 'the product' if it directly supports the device's intended medical purpose. For example, if the AI is essential for measuring the wheal, then it is likely part of the product and could be high-risk.
However, there is currently no official guidance or precedent that clearly states this. Some sources suggest that if the AI system only does retrospective analysis not specific to a patient, it may not be considered 'the product.' But if it supports clinical decisions for individual patients, it probably falls under the high-risk category.
So, until there is more formal guidance, it's best to interpret the law based on how directly the AI contributes to the medical intended purpose of the device.

Is AI used for image analysis in a medical device always classified as high-risk under the EU AI Act?

2 Answers

Keep reading

How will the new EU AI Regulation affect conformity assessment for medical devices?

Questionnaire "Cybersecurity for Medical Devices - Technical Documentation"

Is Your Medical Device Safe? Understanding Biological Evaluation