AI-Powered Medical Devices Surge in U.S. – Europe Lags

10 Min. Reading Time

5 min de lecture

The US Food and Drug Administration (FDA) approved exactly 1,451 medical devices based on artificial intelligence by the end of 2025 – 295 of which were approved in 2025 alone. In Europe, there is no comparable public database. While American clinics deploy diagnostic tools based on AI daily, German hospitals use isolated solutions and are waiting for the EU’s Artificial Intelligence Act (AI Act) to come into effect, the requirements for high-risk systems not taking effect until 2027. A status update.

295 Authorizations in One Year: What the FDA Numbers Reveal

Since 2017, the FDA has maintained a public database listing all medical devices based on AI that have received market authorization. These numbers reveal a clear trend: from 29 authorizations in 2017 to 295 in 2025 – a record that significantly surpasses the previous one, set the previous year at 223.

76% of all authorized AI-based medical devices concern radiology – a field where pattern recognition from large amounts of imaging data works particularly well. GE HealthCare ranks first in the list with 120 authorizations, followed by Siemens Healthineers with 89. The fact that two of the top three manufacturers have European roots makes the regulatory asymmetry between the US and the European Union even more striking.

However, each authorization does not automatically equate to proven clinical utility. An analysis published in JAMA Network Open (April 2025), covering 903 AI-based medical devices authorized by the FDA, arrived at a disappointing conclusion: 24.1% of them had no validation clinical studies. Only 2.4% had been tested in a randomized controlled trial, and 8.1% in a prospective trial. Approximately 90% of the devices relied on retrospective data or no clinical evidence at all.

This means that the figure of 1,451 authorizations granted by the FDA is impressive as an indicator of market dynamics. However, it constitutes only a limited quality criterion. For hospital buyers evaluating an AI-based system, FDA authorization alone is not a sufficient criterion. What matters most are prospective studies conducted on comparable patient cohorts, as well as transparent and reproducible clinical validation in their own care context.

Europe’s Blind Spot: No Database, No Transparency

In Europe, no comparable public list exists. Neither the European Medicines Agency (EMA) nor the German Federal Institute for Drugs and Medical Devices (BfArM) maintains a publicly accessible database of medical devices based on AI that bear the CE mark. The European database EUDAMED, intended to ensure transparency, has been under development for several years. Certain modules will become mandatory from May 2026, but a comprehensive overview of all AI-based products on the European market will only be available several years later.

Independent observers estimate that at least 219 medical devices based on AI bear the CE mark (data as of January 2025). The actual number is likely higher, but no one knows for certain. For hospital buyers faced with the question of which AI-based system to evaluate, this constitutes a concrete problem.

DACH Clinics Deploy AI – But in Isolated Solutions

Despite regulatory gaps at the European level, some DACH (Germany, Austria, Switzerland) clinics are already actively using AI-based diagnostic tools. The results are promising, but remain fragmented.

Asklepios, one of the largest private clinic operators in Germany, has installed the Israeli AI system Aidoc in over 25 facilities. This software analyzes CT scans in real-time to detect strokes, intracranial hemorrhages, and vertebral body fractures – funded by grants from the KHZG program. Its advantage? Aidoc prioritizes the list of imaging reports, allowing critical cases to be treated immediately by the radiologist, rather than remaining on a waiting list.

The largest prospective study to date on the use of AI in diagnostics also comes from Germany. The Vara-PRAIM study, published in Nature Medicine, analyzed over 463,000 screening mammographies. Result: a 17.6% increase in breast cancer detection rate, accompanied by a reduction in radiologists’ workload. This is the largest prospective global study of its kind.

„The question is no longer whether AI works in diagnostics. It’s now about how we can successfully transfer research into everyday practice – on regulatory, organizational, and economic levels.“
– Prof. Alexander Berens, University of Tübingen (press release, 2025)

The problem? Each deployment is an isolated project. There is no national standard for integrating AI into hospital information systems, no standardized interface, and no central registry indicating which clinic uses which system. Although the G-BA’s Innovation Fund supports the systematic integration of AI in radiology through the xR.AI project, Germany is far from widespread deployment.

Consequence: German clinics wanting to introduce AI-based diagnostic tools must themselves pilot every step – from product selection to integration into their hospital information system (HIS), including radiologist training. Additionally, there are unresolved questions of legal liability, both nationally and at the European level. If an AI-based system misses a diagnosis and the attending physician has entirely relied on this system, who assumes responsibility? The Medical Device Regulation (MDR) governs the manufacturer’s liability, but the physician’s professional diligence remains intact. In practice, this uncertainty leads many radiologists to not use AI-provided conclusions as a decision-making basis, but at most as a simple information supplement.

The AI Act will come into effect in 2027: What MedTech companies can expect

The EU’s general AI regulation (AI Act) came into effect in August 2024, but a transitional period applies to medical devices: the full requirements for high-risk systems will only take effect from August 2, 2027. AI-based systems classified as medical devices under the MDR or the In Vitro Diagnostic Medical Devices Regulation (IVDR) are automatically considered high-risk systems.

What this means in practice: manufacturers will have to implement an AI-specific risk management system, comply with strict data quality and data governance requirements, produce technical documentation that goes beyond the MDR requirements, and integrate mechanisms that enable effective human oversight. The workload is considerable, especially for small companies.

A glimmer of hope: In November 2025, the European Commission presented an omnibus proposal aimed at simplifying procedures, intended to reduce overlaps between the AI Act and the MDR/IVDR. Instead of having to produce duplicate documentation and undergo a double conformity assessment, an integrated process is to be established. The adoption of this text is planned for summer 2026.

For German SMEs in the MedTech sector – 93% of companies in the sector have fewer than 250 employees – this issue is far from academic. In recent years, the MDR has already led some manufacturers to withdraw their products from the European market, as the costs associated with certification exceeded their revenues. If the AI Act adds a similar layer of bureaucracy, sector consolidation will continue.

At the same time, the regulation creates a qualitative advantage: medical devices based on AI that meet both the MDR and AI Act requirements will benefit from a trust advantage compared to devices approved only by the FDA. European manufacturers that overcome these two hurdles can exploit this double label as a differentiation factor on the global market – provided they survive financially and organizationally through the transitional period until 2027.

Accelerate or fall behind

The global AI market in healthcare was worth around $21.66 billion in 2025. It is expected to exceed $110 billion by 2030 – a compound annual growth rate of 38.6% (MarketsandMarkets). PwC Strategy estimates the European potential at $203 billion.

Germany has strong players specializing in AI applied to MedTech, such as Siemens Healthineers, Brainomix, Vara, and Merantix. Clinical research is world-class – the PRAIM study provides a striking illustration. What is lacking is the bridge between research and everyday practice: a European AI database guaranteeing transparency; a harmonized authorization process that does not crush SMEs; and a hospital landscape that views AI not as a subsidized project, but as a standard tool.

Over eight years, the FDA has authorized 1,451 AI-based medical devices, while learning to adapt its procedures. During the same period, Europe has drafted a regulation whose effects will only begin to be felt in 2027. Anyone who does not perceive this delay as a strategic risk has not understood the market dynamics.

Frequently Asked Questions