On the 1st of August 2024, the European Artificial Intelligence Act (Regulation 2024/1689) came into force. The world’s first comprehensive AI law, as per the European Parliament, this regulation is particularly noteworthy for healthcare stakeholders as other harmonised legislation such as EU MDR 2017/745, does not contain detailed sections specifically devoted to AI.

The advent of, and widespread public engagement with neural networks such as ChatGPT and Gemini has resulted in increased discourse around ethical considerations and practical limitations of AI, but such discussions around medical device AI have been underway for years, and some leading voices in European healthcare have already raised concerns as to the AI Act’s suitability as a horizontal piece of legislation that is meant to sit astride numerous different industries. Koen Cobbaert from Philips described the AI Act and MDR as “conjoined twins” – inseparable but distinct entities that manufacturers will “have to make work” and alluded to the same issue that Erik Vollebregt, a lawyer specialising in EU and national legal and regulatory issues relating to medical devices, raised, namely that terminologies in the AI Act and MDR do not align, a concern that will immediately prick up the ears of any stakeholder who has had to become literate in the language of EU MDR in recent years.

The requirement for AI-specific legislation in medical devices has been acknowledged almost ubiquitously, though some major figures have voiced their first preference would be to have seen an integration of relevant passages to EU MDR to introduce such regulation, rather than a new, and not entirely complementary measure, though others, such as Stephen Gilbert of the Else Kröner Fresenius Stiftung (a major German medical research non-profit foundation) have supported the AI Act, viewing a distinct legal framework as being essential due to his view of AI as a paradigm altering development that is not just informing but effectively making decisions about medical interventions, rather than acting as all other devices do: as assistive tools in healthcare.

In radiology above all other medical fields, this paradigm shift is already particularly evident, and results have been extensively publicised; mirroring, and in some cases, outstripping, human-level efficacy. Aidoc’s radiology algorithms, widely used across the United States and Europe, have already processed over 10 million scans to date, assisting radiologists by flagging acute abnormalities in CT scans, such as intracranial haemorrhages and pulmonary embolisms, enabling faster prioritisation and treatment. Similarly, Viz.ai, another prominent tool with extensive use in over 1,000 stroke centres globally, rapidly analyses CT scans to detect signs of large vessel occlusion, which is crucial for timely stroke intervention. This tool has been shown to reduce treatment decision time by up to 40%, a significant improvement in stroke care. Lunit INSIGHT, which is actively used for cancer screening in hospitals in the United States, South Korea, and Europe, leverages deep learning to identify lung nodules and breast cancer with a high degree of accuracy, and has been used to analyse more than 5 million medical images worldwide. These AI systems, far from being mere auxiliary tools, now play a critical role in interpreting complex medical images.

Radiology’s adoption of intelligent technology is being followed elsewhere, with several other medical fields making significant strides in integrating AI into both clinical practices and outside-hospital medical devices. In cardiology, wearable devices, such as those developed by Biofourmis, are being used to remotely monitor patients with chronic heart conditions. These devices analyse data from wearables to predict cardiac events, enabling earlier interventions. Biofourmis’ platform is now actively monitoring thousands of patients globally, particularly in the United States and Asia, in partnership with over 50 healthcare providers.

CureMetrix is advancing AI-driven tools that can be used in conjunction with imaging devices to monitor progression in diseases like Parkinson’s and Alzheimer’s. These tools allow for continuous monitoring and early detection, with initial studies suggesting a 20% improvement in the accuracy of early diagnoses.

In diabetes management, AI-driven continuous glucose monitors (CGMs), such as those developed by Dexcom and Abbott’s FreeStyle Libre, automatically analyse glucose levels in real-time and provide patients with actionable insights. These devices are integrated with AI to predict glucose trends and alert patients to potential issues, significantly improving diabetes management. AI-powered CGMs are already used by millions of patients worldwide, offering a level of personalised care that was previously only available in clinical settings.

The widespread adoption of AI in medical devices is continuing apace, but manufacturers will encounter a challenging and potentially confusing regulatory landscape as the AI Act begins to be applied. In Europe, navigation of the already complex and time-consuming EU MDR assessment process for CE marking of an AI-featuring medical device will now be compounded by the requirements of the European AI Act. How this new regulation will be interpreted and applied by Notified Bodies concurrently with MDR will be critical, with specific implications for risk categorisation, transparency and explainability of technologies and real-time demonstrable clinical efficacy being a requirement in a fashion that is not expected to be entirely consistent with the way in which clinical evidence of functionality and post-market data of medical devices are currently assessed.

Set to be enforced in stages, the AI Act’s full implementation is expected to be concluded by 2027-2028, making the coming years pivotal for both manufacturers and regulatory bodies as they adapt to these new and overlapping frameworks. In the UK, the regulatory strategy for AI in medical devices is still developing and remains less advanced than its European counterpart. The UK’s approach will likely evolve in response to both domestic needs and international pressures, but for now, it lags behind the comprehensive structure being rolled out in Europe. As these frameworks continue to evolve, the practicalities of sufficiently monitoring and assessing the safety and efficacy of new medical device technologies without a consequent lengthy grind to get to market will be ironed out, but regulators will undoubtedly face a busy second half of the 2020s as they develop pathways for manufacturers as they aim to balance rigorous assessment with timely market access.