Aidence: The EIC Accelerator Covid CT Scan AI That Never Launched

In March 2020, as hospitals across Europe filled with Covid-19 patients and test kits ran short, Aidence, an Amsterdam company whose software found lung nodules on chest CT scans, asked the EU to fund a new product: AI that reads a chest CT and reports within minutes whether the patient has Covid-19. The European Innovation Council selected the project, CoronAI, in the March 2020 round of the Horizon 2020 EIC Accelerator pilot, among the companies picked for Covid-19 work, and listed Aidence Holding BV as its only beneficiary, with a grant only, according to the Commission's list of selected companies. The EU contribution was €1,910,125 towards a total cost of €2,728,750, for a project running from 1 April 2020 to 30 June 2021, according to CORDIS.

The product was never launched. Aidence built the models, but a clinical validation study at the Netherlands Cancer Institute found that the AI "did not distinguish enough" between Covid-19 and other lung infections, and the company "decided not to launch the product", its final report on CORDIS says. The peer-reviewed results, published in European Radiology in January 2023, concluded that the model's limited generalisability restricted its potential for clinical use. Seven months after the project ended, the US imaging chain RadNet bought Aidence for its lung cancer software, on the same day as it bought Quantib, another Dutch winner of the March 2020 round.

Aidence Today: A Lung AI Maker Inside RadNet

RadNet completed its purchase of all equity interests of Aidence Holding B.V. on 20 January 2022, according to its quarterly report to the SEC and its current report of 24 January 2022. Aidence was folded into DeepHealth, RadNet's AI unit, in 2023, and its Veye Lung Nodules software is now called DeepHealth Lung, according to DeepHealth.

The legal entity that won the EIC grant is no longer in use. OpenCorporates lists Aidence Holding B.V. as inactive, and RadNet's lists of subsidiaries for 2022 and 2025 no longer name it; they name its operating company, Aidence B.V., and Aidence UK Ltd. When and how the holding company ended (merger or dissolution) could not be verified from free register sources. A trust office foundation that held its shares, Stichting Administratiekantoor Aidence Holding, is marked as terminated on North Data.

Aidence B.V. is active. North Data lists it at Westblaak 130 in Rotterdam, Quantib's address, with "DeepHealth" as a trade name, and shows balance sheets for 2022, 2023 and 2024, the last published on 4 August 2026, according to North Data. The figures sit behind a paywall and could not be checked for this article.

The Technology: Reading Covid-19 From a Chest Scan

The idea behind CoronAI was that a CT scanner already in every hospital could stand in for scarce PCR tests. The software was to take each chest CT straight from the scanner, analyse it in the background and return "an easy-to-read report on the most likely diagnosis" within minutes, according to the CORDIS objective, which cited studies suggesting CT is "as least as sensitive" as the RT-PCR test. Because it would run on every chest CT, Aidence said that once deployed in hospitals around Europe it would also serve as a case registry for tracking new outbreaks.

The model graded scans on CO-RADS, a five-point scale radiologists use to rate how likely Covid-19 is on CT, and measured how much of the lung was affected. It was developed on 1,286 CT scans from 1,266 patients, including 194 control scans from the US National Lung Screening Trial, gathered from institutions in the Netherlands, Belgium, France, Singapore and the UK under a consortium called ICOVAI, according to the European Radiology paper by Laurens Topff and colleagues at the Netherlands Cancer Institute, four of whose co-authors were Aidence employees. Aidence's established product, Veye Lung Nodules, does a narrower job: it finds, measures and tracks the growth of lung nodules, which can be early lung cancers, on chest CT.

What CoronAI Promised, and What CORDIS Shows

The EIC treated the March 2020 round as an emergency call. On 8 June 2020 the Commission said it had awarded close to €166 million to 36 companies working on the coronavirus, out of more than 1,400 coronavirus-related applications among almost 4,000 in March, as reported by the Commission's representation in Portugal and by ERRIN. The Commission's list puts CoronAI in its section of COVID-19-related projects from the 20 March 2020 cut-off.

The one periodic report on CORDIS, covering the whole project, is frank. Aidence built two models to detect Covid-19 on chest CT and a third to grade its severity, and rebuilt its single-product engine so that it could run several AI models. Collecting scans from hospitals took longer than planned and the team had to build its own annotation tool, but the model was ready for clinical validation by the end of 2020, nine months in. The Netherlands Cancer Institute (NKI/AVL) began the validation study in the first quarter of 2021. It was "extremely positive" about the severity model, the report says, but found that the detection model could not tell Covid-19 well enough from other lung diseases. By the end of June 2021 Aidence had more than 50 sites contracted to receive the Covid-19 model; on 30 June it told customers and the hospitals that had supplied scans that it would not launch.

The published study shows why. On an external test set of 374 analysed scans from five hospitals in the Netherlands, Belgium and Italy, the agreement between the AI's CO-RADS scores and the reference fell on every measure, and the segmentation of diseased lung tissue fell from a Dice score of 0.76 to 0.59. The authors wrote that the results "showed limitations in the generalizability of the ICOVAI model", and the paper states that the study was funded under EU grant agreement 961522.

1. Covid-19 detection on chest CTNot achievedFailed to separate Covid-19 from other lung infections in NKI/AVL validation.
2. Product deployed in European hospitalsNot deliveredLaunch cancelled on 30 June 2021 despite more than 50 contracted sites.
3. Outbreak case registryNot deliveredDepended on deployment, which never happened.
4. Severity modelBuilt, not launchedRated promising by NKI/AVL per CORDIS; no product followed.
5. Validation publishedDeliveredEuropean Radiology, January 2023, crediting grant 961522.
6. Project run to termOn timeClosed on its planned end date, 30 June 2021.

The project's lasting output was indirect. Aidence's report says the multi-model engine and the annotated dataset would help its lung cancer work, and that the project had helped it build "a customer base" and "commercial traction" on which it planned a B round. No B round was announced; the company was sold instead. CORDIS lists no other EU-funded project for Aidence, according to a CORDIS search of its name. The EIC Fund's list of investments, checked page by page for this article, does not include Aidence, DeepHealth or RadNet, consistent with a grant-only award.

What the Company Achieved Instead

Before the grant, Aidence had raised a $2.5 million seed round in 2017 and a €10 million Series A in February 2019, co-led by INKEF Capital and Rabo Ventures with Northzone, HenQ and Health Innovations, to move towards FDA clearance in the US and expand its R&D team, according to Global Venturing; MedNous reported the €10 million round with INKEF as lead.

The growth came from lung cancer, not Covid-19. In England's Targeted Lung Health Check programme, 15 of the 19 pilot sites used Aidence's nodule software, HTN reported in September 2021. In December 2021 Aidence, Hardian Health and the University of Edinburgh launched INPACT, an evaluation funded by the NHS AI in Health and Care Award, under which Veye Lung Nodules was to be installed in 20 more UK hospitals, six of them in the study, with results due in 2022, according to the University of Edinburgh. Published INPACT results were not found for this article. When RadNet bought it, Aidence had customers in seven European countries including France, the Netherlands and the UK, according to RadNet's press release, AuntMinnie and Silicon Canals.

The Money: A Price Tied to FDA Clearances That Never Came

RadNet's numbers for the deal differ by filing. Its January 2022 current report put the aggregate consideration at about $50 million, including future milestones valued at $10 million. Its quarterly report and its 2022 annual report put the purchase consideration at about $45.2 million: 1,117,872 RadNet shares worth $30.0 million, $1.8 million in cash, $11.9 million in contingent consideration ($7.4 million in milestones and a $4.5 million share holdback) and the settlement of a $1.5 million loan from RadNet. RadNet booked $27.7 million of intangible assets, including $21.1 million of developed technology, and $22.9 million of goodwill.

The milestones were tied to FDA authorisation of Aidence's software. RadNet's 2025 annual report says that in September 2023 it concluded the milestones could not be met "because the original submissions of artificial intelligence screening solutions did not receive regulatory clearance", booked a $7.2 million gain on the contingent liability and took a $3.9 million impairment on the research asset for Veye Lung Nodule and Veye Clinic. The sellers received the general holdback as 144,227 RadNet shares, valued at $4.0 million, on 13 April 2023.

The US gap has not closed. The FDA's 510(k) database returns no clearance in Aidence's name, and DeepHealth's product page states that neither Veye Lung Nodules nor Veye Reporting is FDA cleared and that "DeepHealth Lung AI is FDA 510(k) pending". Aidence's own revenue is not reported; RadNet reports it inside its Digital Health segment, whose revenue was $92.7 million in 2025, a figure that cannot be attributed to the lung software.

Why Europe Should Care

Lung cancer screening is where Aidence's technology now matters to EU policy. The Council recommendation on cancer screening adopted in December 2022, a flagship of Europe's Beating Cancer Plan, invites member states to "explore the feasibility and effectiveness of low-dose computed tomography to screen individuals at high risk for lung cancer", according to the Commission's press release and WONCA Europe. Screening on that scale would produce more CT scans than radiologists can read unaided, the problem Aidence's nodule software addresses.

CoronAI itself also carries a lesson the EU paid for. The Covid-19 imaging AI built at speed in 2020 worked on the data it was trained on and lost accuracy on other hospitals' scans, and Aidence published that result instead of shipping the product. As with Quantib, the European company the EU funded is now owned by a US imaging chain, which decides where the lung AI is developed.

The Verdict

Aidence, DeepHealth and RadNet have not been asked for comment for this article. It is based entirely on public records, regulatory filings, company statements and EU project data, linked throughout.