Zeno AI
Zeno AI is a real-time AI platform that detects and classifies benign and malignant head and neck lesions during routine flexible endoscopy. Built for ENT clinicians and nurses, it is CE-MDR Class IIb certified decision support, not diagnosis.
What is Zeno AI?
Zeno AI is a real-time artificial intelligence platform for head and neck endoscopy, presented by its publisher as the first on the market to carry CE-MDR Class IIb certification for detecting and classifying benign and malignant lesions, starting with the throat. The analysis happens during the examination itself, a flexible transnasal laryngoscopy, rather than afterwards on recorded material.
The clinical problem is one of expertise distribution. Spotting and classifying a lesion live is hard for an ENT who is not a head and neck specialist: in Scholman et al., 111 observers reached an overall sensitivity of 66.3% in high definition, meaning close to a third of laryngoscopic diagnoses were incorrect. Head and neck surgeons work in academic hospitals while ENTs practise everywhere, so the expert opinion is not available where most patients are first seen. Delay follows: in the Netherlands patients wait an average of 37 days before treatment, and the literature links waiting time to overall survival for several head and neck cancers. The publisher estimates that generalised use across Dutch ENT practice would save between 53.6 and 107.2 million euros a year.
The architecture is deliberately austere: a closed system with no external connections, plug-and-play on existing endoscopy systems, vendor agnostic, storing no sensitive patient data. On screen it marks where the lesion is and whether the model reads it as benign or malignant.
The evidence is public. A multicentre study published in Cancers (2026) with Radboudumc and UMCG, with additional cases from Martini Hospital in Groningen, used 1,336 lesion videos and 123 healthy control videos collected between 2012 and 2024, and more than 142,000 annotated frames on histopathologically or clinically confirmed diagnoses. On an independent test set the unified model reached 92.1% detection precision, 73.3% recall, at least one correct detection in 98.2% of lesion videos and 94.3% sensitivity for malignancy among correctly detected lesions, at 76 frames per second. Earlier clinical validation was published with Wiley in 2023.
The publisher is explicit about the boundary: Zeno AI is decision support only, not direct diagnosis, and false positives, false negatives and misclassification remain possible. It comes from WSK Medical B.V., an Amsterdam company founded in 2018, is distributed worldwide by Olympus and is live in Europe.
What it does
- Detects head and neck lesions in real time during flexible endoscopy
- Classifies each lesion as benign or malignant
- Locates the lesion on the live endoscopic image
- Covers the glottic, supraglottic and hypopharyngeal regions with a single unified model
- Runs at 76 frames per second, fast enough to follow the examination live
- Operates as a closed system, with no internet connection and no patient data storage
- Plugs into existing endoscopy systems, vendor agnostic
When to use Zeno AI / When not to
A quick filter to help you decide if Zeno AI is the right fit.
When to use Zeno AI
- ENT clinicians in general outpatient clinics who have no dedicated head and neck expertise on site
- Hospitals trying to shorten the delay before treatment, which averages 37 days in the Netherlands
- Care centres that want on-device AI running with no internet connection and no patient data storage
- Residents and clinical staff still learning laryngoscopic diagnosis, as reported by clinicians at HUS in Finland
- Clinical research teams running multicentre studies on AI-assisted endoscopy
When not to use Zeno AI
- Anyone looking for a diagnostic tool: the publisher itself warns of false positives, false negatives and incorrect benign/malignant classification
- Pathology teams working on digital slides, which are covered by the publisher's separate product, Zeno Pathology
- Departments whose priority is catching early T1 tumours, named as a limitation by the authors of the 2026 study
- Consumers and private individuals: the terms set a minimum age of 18 and the product is sold through a commercial demonstration
- Services looking for oral cavity coverage, still a grant-funded research project with UMC Groningen rather than a catalogue product
How to use Zeno AI
A typical end-to-end flow, from setup to results.
- Request a demonstration through the site's contact form, choosing "Product Demo" as the enquiry type and identifying yourself as a business
- Discuss purchase with WSK Medical or with Olympus, the worldwide reseller
- Have the unit connected to the endoscopy system already installed in the department: it is plug-and-play and vendor agnostic
- Plan no network work: the system is closed, needs no internet connection and requires no account
- Request the eIFU password through the contact form and read the electronic instructions for use before clinical use
- Run a routine flexible transnasal endoscopy exactly as usual
- Read the live overlay showing where the lesion is and whether the model classifies it as benign or malignant
- Make the clinical decision yourself: the tool supports the decision and never issues a diagnosis
- Use the same contact form for support questions, which are handled as a dedicated enquiry type
Pros & Cons
Pros
- Presented by the publisher as the first solution on the market with CE-MDR Class IIb certification for this use
- Performance published in a peer-reviewed journal (Cancers, 2026), with earlier clinical validation published with Wiley (2023)
- Clinical testimony with figures: Dr Boudewijn Plaat (UMCG) reports laryngeal carcinoma recognised in more than 90% of cases with Zeno AI, against 60-70% correct specific diagnosis without it
- No data leaves the institution: closed system, no patient data stored
- Light installation on endoscopy hardware already in place, with no dependency on one manufacturer
- Adoption reported as easy by residents and clinical staff (Prof. Ahmed Geneid, HUS)
- Documented quality compliance (ISO 13485, IEC 62304, ISO 14971, IEC 82304-1, ISO 14155, IEC 62366-1, ISO 20417, ISO 15223-1, IEC 81001-5-1), a downloadable ISO 13485 certificate, a designated PRRC, and worldwide distribution backed by Olympus
Cons
- No public price: the figure is only obtained through a sales conversation
- 73.3% recall in the 2026 study, with early T1 tumours named as an explicit limitation by the authors: some lesions go undetected
- No API and no documented interoperability; DICOM, HL7, FHIR and PACS appear nowhere on the site
- The eIFU is password-protected, so conditions of use cannot be assessed before contacting the vendor
- Legal documents are old (terms 28 February 2022, privacy policy 16 March 2022), written for the former wskmedical.org domain, and the controller address they give (Science Park 608 unit K09) differs from the site's own (Science Park 301 unit 4.65-4.67)
- No subprocessor list, no published DPA, no explicit claim of GDPR compliance, and nothing published on whether customer data might be used to train the models
- No interface language is documented, and the site announces deployment in "5 different countries" before listing six
Pricing & Plans
No price is published. The site has no pricing page at all, as its complete page sitemap confirms, and no sales amount appears anywhere. Zeno AI is sold through a commercial contact, using the "Book a demo" form, with worldwide distribution by Olympus. Neither a free plan nor a free trial is announced, and the publisher neither confirms nor rules out either. The only monetary figures on the site are estimated healthcare savings (53.6 to 107.2 million euros a year in the Netherlands if every ENT used the tool) and public research grant amounts; neither of these is a price.
Data, GDPR & hosting
A consolidated view of how Zeno AI handles your data.
GDPR overview
The letters GDPR appear nowhere on the site or in the two legal PDFs, so there is no explicit compliance claim to report. The privacy policy nevertheless follows GDPR grammar: information, access, rectification, erasure, restriction, objection, portability and withdrawal of consent are all described, complaints may go to the Dutch Autoriteit Persoonsgegevens (Postbus 93374, 2509 AJ Den Haag, 088-1805250), and privacy questions go to privacy@wskmedical.org. Being established in the Netherlands, WSK Medical needs no Article 27 representative and names none. The gaps are notable: no subprocessor list and no published data processing agreement, the policy stating only that it is "required to conclude a data processing agreement in situations where we process personal data on your behalf". The documents are old, dated 16 March 2022 (privacy), 28 February 2022 (terms) and 29 September 2022 (cookies), and written for the former wskmedical.org domain.
Who owns the data?
Two very different regimes sit side by side. On the clinical side, the publisher states that Zeno AI is "developed as a closed system with no external connections" and that it "does not store any sensitive patient data", so patient material never leaves the hospital and no ownership question arises for the vendor. On the website side, WSK Medical is the data controller, named in the privacy policy at Science Park 608 unit K09, 1098 XH Amsterdam, an address that differs from the one shown in the site footer. Visitors and customers may request access, rectification, erasure, restriction, objection and portability. That policy covers site visitors and customers only, never the clinical data handled by the device.
Reuse rights
The published terms and privacy policy say nothing about an end user reusing data produced by the device, and there is no vendor-held clinical dataset to reuse: Zeno AI runs as a closed system, stores no sensitive patient data and sends nothing outside the hospital. For the website, WSK Medical states that personal data may be shared with suppliers, partners and processors under contractual arrangements, and disclosed to authorities where the law requires it. Transfers to recipients in third countries, including the United States, are said to be made only on the basis of EU standard contractual clauses. First-party cookies measure visits and time spent; the Complianz cookie declaration of 29 September 2022 names Wistia, Divi, WordPress, Google Fonts and YouTube, and beyond geolocation, IP address, device type, browser and version, no identifiable personal data is said to be stored in cookies. Nothing on the site states whether customer data is ever used to train the models.
Data retention & training
Hosting summary
Two layers must be kept apart. The device hosts nothing externally: Zeno AI is described as a closed system with no external connections that stores no sensitive patient data, so clinical material stays inside the hospital and no vendor jurisdiction attaches to it. For the website, WSK Medical declares no hosting country and no hosting region in its privacy policy; it states only that transfers to recipients in third countries, including the United States, are made on the basis of EU standard contractual clauses. No subprocessor list is published. As a technical observation outside any statement by the publisher, the website resolves to 185.37.71.28, hosted in the Netherlands by Your Hosting B.V.; that describes where the marketing site runs, not where clinical or customer data lives, and it should not be read as a hosting commitment.
Things to keep in mind
Risks and trade-offs to weigh before adopting Zeno AI.
- Decision support, never diagnosis: the publisher itself warns of false positives, false negatives and incorrect classification, and the clinical call stays entirely with the clinician
- A 73.3% recall and an explicit limitation on early T1 tumours mean a lesion can be missed, so an absence of alert must never be read as reassurance
- Over-reliance is the real human risk: residents learning with a live overlay may lean on it, and the publisher's own comparison (over 90% recognition with the tool against 60-70% without) can quietly encourage that dependence
- CE-MDR marking targets the EU and markets that accept CE marking; no clearance from any other authority, the FDA in particular, is claimed
- The site announces deployment in five European countries then lists six (Netherlands, Finland, Germany, Poland, Portugal, United Kingdom): figure and list do not match
- Documentary inconsistencies: the privacy policy gives a controller address (Science Park 608 unit K09) different from the site's (Science Park 301 unit 4.65-4.67), the legal and privacy mailboxes sit on wskmedical.org, the old domain that now redirects, and the regulatory compliance address is hosted at a third-party provider, MedQAIR
- Nothing is published about any training of the models on customer data, and the publisher gives only a founding year, never a full incorporation date
Setup & Integrations
Technical difficulty
Technically light, organisationally heavy. The unit is plug-and-play on the endoscopy systems already installed and is vendor agnostic, no internet connection has to be opened, no account has to be created, and nothing is installed on the end user's side: there is no web, mobile or desktop application. In practice the effort lies elsewhere, since installation goes through the supplier, a sales contact and then the Olympus reseller channel, and the eIFU needed for use requires a password on request. Expect a hospital equipment project with regulatory review and procurement, not an online sign-up.
Behind Zeno AI
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What does Zeno AI do?
Is Zeno AI a diagnostic tool?
What is its regulatory status?
What performance figures are published?
Does patient data leave the hospital?
How is it installed?
How much does it cost?
Where is it deployed?
Is there an API?
Who publishes Zeno AI and where are the instructions for use?
Should you pick Zeno AI?
Zeno AI is a narrow product, and that is its strength. It does one thing, reading a flexible head and neck endoscopy in real time and marking a lesion as benign or malignant, with a genuine CE-MDR Class IIb marking behind it and results published in peer-reviewed journals rather than in marketing copy alone. For an ENT clinic with no head and neck surgeon down the corridor, that answers a real gap.
The closed architecture cuts both ways. Nothing leaves the hospital, no patient data is stored and no internet connection is required, which is about as reassuring as clinical AI gets for a hospital privacy officer. The same design closes off interoperability entirely: no API, no documented DICOM, HL7, FHIR or PACS integration. Buyers who expect imaging AI to sit inside a data pipeline should read this as a deliberate trade-off rather than an oversight.
Commercially, the product is opaque: no price, no tier, no trial, only a demonstration request and a worldwide reseller. The legal documentation reinforces that impression, since the terms and the privacy policy date from 2022, were written for the publisher's former domain and give a controller address that differs from the one in the site footer.
European deployment is real but still narrow, and the study authors themselves flag early T1 tumours as a limit alongside a 73.3% recall. Zeno AI deserves evaluation by ENT departments that want live decision support inside an examination they already perform, provided they run their own clinical validation and keep treating the tool as what its publisher says it is: support for a decision, never the decision itself.
- Choosing a selection results in a full page refresh.
- Opens in a new window.