Navigating the regulatory landscape for artificial intelligence in healthcare can feel overwhelming, especially regarding its deployment as a medical device in Europe. But there’s good news! A newly released consensus paper from the AI Working Group of the European Society of Radiology (ESR) provides much-needed clarity on this complicated issue.
This comprehensive document sheds light on essential aspects of post-market clinical follow-up (PMCF) and outlines the regulatory framework surrounding the use of radiology AI. It offers twelve valuable recommendations aimed at helping physicians tackle these intricate challenges. According to Dr. Renato Cuocolo, an associate professor of radiology at the University of Salerno in Campania, Italy, who led the initiative alongside his colleagues, these insights are crucial for healthcare professionals involved in the implementation of AI technologies.
Shockingly, only 29% of those deploying medical AI—be it hospitals, radiology departments, or individual clinicians utilizing AI for diagnosis or treatment—report feeling well-versed in the EU’s medical device regulation (MDR) and post-market surveillance (PMS) requirements. This gap in knowledge poses significant risks, as radiologists not only share responsibilities related to PMS but are also expected to take part in ongoing evaluations to ensure safety and efficacy.
The group emphasized, "Radiologists must educate themselves about the AI as a Medical Device (AIaMD) regulations in the EU due to their shared PMS obligations and the current lack of clear guidance tailored specifically to imaging."
Cuocolo and his co-authors pointed out that the existing EU MDR (2017/745) was not specifically designed with AI functionalities in mind, leaving many important questions unanswered regarding the unique characteristics of AI in medical contexts. Moreover, the new EU AI Act (Regulation (EU) 2024/1689) does not significantly clarify these issues either, which could explain why many radiologists remain unaware of their responsibilities under these regulations.
To fill this knowledge gap, members from the ESR’s eHealth and Informatics Subcommittee, along with the AI Working Group, have crafted twelve consensus statements. These statements serve as a guide for healthcare professionals currently employing or planning to adopt AIaMD in their practices. Here are a few highlights:
- Transparency in Metrics: To enhance effective PMS, it's essential for providers to present baseline accuracy metrics of the AI medical devices, including uncertainty measures—like 95% confidence intervals. Making this information easily accessible on the PMS platform will help users spot underperformance or other complications swiftly.
- Facilitating User Feedback: For ease of compliance with PMCF obligations, it’s recommended that the PMS platforms allow user feedback to be documented. Furthermore, this input should be shared among other practitioners in the same institution via the same platform, ensuring local awareness of any critical issues associated with an AI medical device.
- Optimizing Software Access: Providers might consider using shared software platforms to streamline access to multiple AI medical devices. In such cases, adopting interoperable PMS standards is advisable, enabling the aggregation of data across all devices. Ideally, a unified user interface would allow monitoring of all devices accessed through a single platform, rather than having isolated PMS systems for each specific device.
The authors view this paper as an initial step toward establishing necessary guidelines and fostering awareness within the radiology community. Their goal is to improve the consistency of AI deployment and monitoring while providing physicians with a reliable reference point when engaging with these innovative technologies, which introduce distinct challenges compared to traditional practices.
Additionally, the paper aims to present informed perspectives to EU regulatory bodies, potentially influencing future regulations.
For those interested in exploring the complete paper, you can find it here.
So, what do you think? Are you familiar with the current EU regulations surrounding AI in medical practice? Or is there something you believe needs clearer guidance? Share your thoughts in the comments!
