Did the EIC Accelerator Pay Off for Radiobotics' Fracture AI?

More than 140 million X-rays of bones and joints were analysed in Europe in 2018, by Radiobotics' own count to the EU, and there are not enough radiologists to read them quickly. Radiobotics, founded in Copenhagen in 2017, set out to let software do the first read. In January 2020 the European Innovation Council selected it in the EIC Accelerator pilot, and the EU paid €1,333,816.05 towards AutoRay, a project to turn its algorithms into a market-ready suite for musculoskeletal X-rays.

Six years later the company has a fracture detector, RBfracture, that is CE marked as a Class IIa medical device and that the UK's National Institute for Health and Care Excellence lets the NHS use while more evidence is gathered. It also has DKK 79 million of accumulated losses, three consecutive annual reports with a going-concern warning, and a new owner: Medimaps Group SA of Geneva, registered as its shareholder on 16 April 2026, according to North Data, which reproduces the Danish company register (CVR).

Where Radiobotics Stands Now

Radiobotics still exists as a Danish company and still sells RBfracture under its own brand, but it is no longer independent. On 2 March 2026 it and Medimaps Group, a Geneva developer of bone microarchitecture software, announced a "strategic merger agreement", subject to Danish foreign direct investment approval. Financial terms were not disclosed, according to both the announcement and AuntMinnie. The word "merger" describes the deal more generously than the register does. On 16 April 2026 Medimaps Group SA was entered as a shareholder, the four Radiobotics board members and chief executive Peter Ulvskjold were removed from the register, and Medimaps' chief executive Didier Hans became chair and director, according to North Data, which lists Medimaps as the one known shareholder. The Radiobotics website now calls it "a Medimaps Group company", and Medimaps says the combined group has "80+ employees" in Geneva, Copenhagen and Wilmington.

The sale followed a cash squeeze. Radiobotics' last annual report, for the year to 30 September 2025 and signed on 13 March 2026, says its liquidity budget for 2025/26 "shows an unsecured liquidity need", that it was "in the final phase of a capital raise from a international company", and that existing investors had committed bridge financing until that deal closed, according to the filing with the Danish Business Authority. The auditor, Beierholm, flagged a material uncertainty about going concern, as it had in the 2022/23 and 2023/24 reports.

The Technology: A First Reader for Bone X-rays

Missed fractures on plain X-rays are the most common diagnostic error in emergency departments, Radiobotics' SBRI announcement says, citing NHS Resolution. Radiobotics trains neural networks on labelled X-rays so that the software marks suspected findings before a clinician looks. Its regulatory page describes RBfracture as "computer-assisted detection and diagnosis software" for "trauma skeletal radiographs", which flags fractures, dislocations, lipohaemarthrosis and joint effusions. RBknee, the company's first product, grades the X-ray signs of knee osteoarthritis (bone spurs, hardened bone under the cartilage and narrowing of the joint space) and measures joint space width, according to AuntMinnie.

The approach is not unique. When NICE assessed AI fracture detection in January 2025, it allowed four tools, Rayvolve, TechCare Alert, BoneView and RBfracture, for NHS use during an evidence-gathering period, and limited a fifth, qMSK, to research. What Radiobotics argued would set it apart was the breadth of its suite (several anatomical regions and conditions in one package) and the depth of its clinical validation.

What AutoRay Promised, and What CORDIS Shows

Radiobotics first took a €50,000 SME Instrument Phase 1 grant for a three-month AutoRay feasibility study from March to May 2019, according to CORDIS. The full project was selected at the 8 January 2020 cut-off of the EIC Accelerator pilot; the European Commission's list of selected companies names "RADIOBOTICS APS" of Copenhagen as the sole beneficiary of AutoRay, with grant-only funding. The grant agreement was signed on 30 June 2020 and the project ran from 1 July 2020 to 31 March 2023.

The objective on CORDIS promised "a market-ready software suite addressing multiple conditions in multiple anatomical regions", "seamless integration in IT systems" and "a large-scale demonstration of its performance in clinical practice". The final periodic report, covering July 2022 to March 2023, is unusually thin for a closing report. It says work had "centred around preparation of studies with clinical partners", that "the first part of the software products" had been developed and that several hospital integrations had been "prepared". It also says "it is expected to have completed the multicenter clinical trials by the end of the project", in a report that covers the end of the project. CORDIS lists no deliverables or publications.

1. Market-ready products for several conditions and body regionsDeliveredRBknee CE marked and FDA cleared in 2021; RBfracture CE marked under the MDR (reported June 2022).
2. Integration into hospital IT systemsDeliveredBoth products offered on TeraRecon's Eureka AI platform from April 2023.
3. Multicentre clinical trials completed by project endLateNot complete in the final report; a three-hospital RBfracture validation was published in Radiology in August 2026.
4. Large-scale demonstration in clinical practiceAfter the projectNHS use under NICE's January 2025 assessment; an Oxford randomised trial approved in October 2025.

The products that AutoRay was meant to mature exist, are certified and are on sale, so the grant bought what it was for. The clinical evidence came years after the project closed, and much of it was paid for by others. Two dates on the CE mark also disagree: Radiobotics' awards page says RBfracture has been CE marked "since 2021", while AuntMinnie Europe reported the certification in June 2022.

What Radiobotics Achieved After the Grant

RBfracture became the company's flagship and, by mid-2026, its only product on the solutions page. Its regulatory page, updated on 27 May 2026, lists a Class IIa CE mark under the EU Medical Device Regulation certified by TÜV SÜD, a Class IIb listing in Australia, registration in Brazil and authorisations in Great Britain, Israel, New Zealand, Saudi Arabia, Singapore, South Africa, Switzerland, Tunisia, Türkiye, Ukraine and the United Arab Emirates. It lists no US clearance for RBfracture. RBknee, the one product the FDA cleared, no longer appears among the company's solutions, although its studies remain on the publications page.

The strongest outside endorsement is British. In January 2025 NICE's early value assessment said RBfracture was one of four AI technologies that "can be used in the NHS during the evidence generation period" for people aged 2 and over, on condition that the company generates the evidence NICE asked for within two years, according to the guidance. In February 2025 Radiobotics won £500,000 from SBRI Healthcare, an NHS England innovation programme, to install RBfracture at three NHS trusts for a study led by Oxford University Hospitals, it announced. That study became SAMURAI Pro, which Radiobotics and the Oxford trust call the first randomised clinical trial of radiology AI in urgent care.

The independent evidence is solid but not spectacular. A study led by Erasmus MC in Rotterdam tested RBfracture on 500 consecutive cases at each of three academic hospitals in Denmark, Germany and the Netherlands and found sensitivity of 80% to 86% and specificity of 93% to 96%, with sensitivity for wrist and hand X-rays at one hospital down at 75%, AuntMinnie reported. With the AI's help, the 12 clinicians in the reader study raised their sensitivity by 11 percentage points without losing specificity. The company's own marketing quotes a higher figure, 94% accuracy, sensitivity and specificity, in its Oxford trial announcement.

The Money

Radiobotics files small-company accounts that do not disclose revenue. They do show the trajectory. The net loss rose from DKK 2.0 million in 2018/19 to DKK 19.6 million in 2024/25, and the gross result, which nets revenue against external costs, was negative every year from 2019/20, according to the annual reports filed with the Danish Business Authority. Over eight financial years the losses add up to DKK 79.0 million. The latest balance sheet holds the first clear sign of commercial traction: trade receivables of DKK 2.7 million, up from DKK 70,103 a year earlier, and DKK 3.7 million of prepayments from customers, against none in 2024. Equity was negative again, at minus DKK 1.7 million.

Shareholders kept the company going. The statements of changes in equity show DKK 14.5 million of new capital in 2020/21, most of it the conversion of DKK 10.4 million in convertible loans, then DKK 27.0 million in 2022/23, DKK 18.2 million in 2023/24 and DKK 16.8 million in 2024/25, each time in cash and converted loans. The register lists Danish holding companies and investment vehicles such as inQvation ApS, Innovations A/S and MM Medical ApS among those whose holdings changed when Medimaps came in, according to North Data. Radiobotics does not appear in the EIC Fund's list of invested companies, which is consistent with its grant-only award.

Why Europe Should Care

AI that reads medical images sits where two EU priorities meet. Under the AI Act, AI that is a safety component of a regulated product such as a medical device counts as high-risk, and the Commission now dates those obligations for product-integrated systems to 2 August 2028, according to its AI Act page. The European Health Data Space, in force since 26 March 2025, is meant to open pseudonymised health data for research and innovation, the raw material such tools are trained on. A European developer that already holds an MDR certificate is the kind of company those rules assume will exist.

Radiobotics' fate shows the other side. Its first randomised trial is funded and run by the British health system, and control passed to a Swiss group, outside the EU, in a deal that needed Danish foreign direct investment approval. The team remains in Copenhagen; the board is now chaired from Geneva.

The Verdict

Radiobotics and Medimaps have not been asked for comment for this article. It is based entirely on public records, company statements and EU project data, linked throughout.