Medannot
Medannot is a browser-based, AI-driven platform that turns DICOM scans into patient-specific 3D anatomical models in minutes, lets clinicians annotate and segment volumetric imaging, and trains custom segmentation models without a single line of code.
What is Medannot?
Medannot is an AI-driven collaborative platform for radiology annotation and 3D surgical planning that runs entirely in a web browser. It is published by MEDANNOT s.r.o., a Slovak company, and presents itself as built for healthcare professionals by healthcare professionals.
The platform does three connected things. It lets clinicians annotate and segment volumetric medical scans; it converts DICOM studies from CT, MRI or CBCT into patient-specific 3D anatomical models; and it lets teams train their own segmentation models without writing code. Four use cases are advertised around those capabilities: radiologic diagnosis, 3D modelling, 3D surgical planning and radiomics. The product page reports more than 2,300 annotated cases processed, 3D model generation twenty times faster than manual workflows, and under five minutes on average from scan upload to segmentation.
Five claimed advantages structure the offer. It is web-based, with nothing to install and no dependency on a particular machine. It is versatile, because custom AI models can be imported or trained in place. It is cloud-hosted yet, according to the homepage, easily hostable on-premise. It is collaborative, with role assignment and remote review. And it is fast, generating models in seconds. The annotation workspace bundles an AI tool called 3D Sam alongside brush, freehand and polygon lasso, ruler, fill and threshold tools, 3D cut, split, ROI and axis controls, windowing, mesh rendering, volume rendering, comments and STL import and export. A separate virtual-reality environment, MedannotVR, lets colleagues view, label, edit and inspect the same models together in real time.
Medannot is training models across the body, from oesophagus, spine and heart chambers to cranium with mandible, maxilla and vertebrae, aorta, left atrium, spleen, pulmonary lesions, pancreas, colon, kidneys, liver and coronary arteries, tumours included, under the stated goal of covering the entire human body. Two clinical deployments are named: automatic renal segmentation and perfusion management at Ghent University Hospital in Belgium, and semi-automatic segmentation of epicardial adipose tissue at Slovakia's National Institute of Cardiovascular Diseases. Six audiences are addressed directly: surgical teams, medical engineers, radiologists, AI and machine-learning professionals, hospitals and clinical research institutions, and medical educators.
What it does
- Convert DICOM studies from CT, MRI or CBCT into patient-specific 3D anatomical models
- Annotate and segment volumetric medical images with AI assistance
- Train and deploy custom segmentation models without writing any code
- Plan surgery on interactive 3D reconstructions of tumours, vessels and organs
- Extract structured features from annotated images for radiomics research
- Collaborate in real time with role-based access across teams and institutions
- Inspect, label and edit 3D models in virtual reality through MedannotVR
When to use Medannot / When not to
A quick filter to help you decide if Medannot is the right fit.
When to use Medannot
- Surgeons and surgical teams who need patient-specific 3D anatomy for pre-operative planning
- Radiologists and diagnostic specialists looking to speed up segmentation and image labelling
- AI and machine-learning teams building annotated 3D imaging datasets and training segmentation models
- Hospitals and clinical research institutions centralising imaging workflows across departments
- Medical educators and academic institutions teaching anatomy and radiology from real patient scans
When not to use Medannot
- Buyers who need a published price list, since Medannot discloses no rates and routes every enquiry through a contact form
- Individuals wanting instant self-service sign-up, because access is granted only after a sales conversation
- Teams working outside DICOM imaging: without CT, MRI or CBCT studies the platform has no purpose
- Organisations unable to anonymise DICOM files before upload, as the platform expects patient metadata already stripped
- Anyone needing a mobile app, a desktop app or documented public API endpoints, none of which Medannot publishes
How to use Medannot
A typical end-to-end flow, from setup to results.
- Request access through the contact form, since Medannot has no self-service sign-up and grants its free trial on request
- Sign in to the platform through the login page once your account has been opened
- Anonymise your CT, MRI or CBCT study, then drag and drop the DICOM files into the browser
- Work as a clinician or as a technician, the two profiles the interface distinguishes
- Segment and annotate with the 2D, 3D and AI-assisted tools, or run one of the existing models
- Train a custom segmentation model on your own labels, with no coding required
- Deploy the trained model in a few clicks so it segments the studies that follow
- Generate the patient-specific 3D model and review it in mesh or volume rendering
- Import STL files and labels to compare successive scans of the same patient over time
- Download the finished model as an STL file from the segmentations panel, or inspect it in MedannotVR
Pros & Cons
Pros
- Nothing to install: the whole workspace runs in a standard browser on any computer
- Fast results, with under five minutes on average from upload to segmentation according to the site
- Clinicians can train their own AI models, with no engineers and no coding involved
- Deployment choice between cloud hosting and an on-premise installation
- Strong compliance posture advertised: MDR and FDA ready, ISO and SOC 2 certified, GDPR compliant
- DICOM files are anonymised and patient metadata stripped at upload
- Named clinical deployments at Ghent University Hospital and Slovakia's National Institute of Cardiovascular Diseases
Cons
- No public pricing at all: no rate card, no named plan and no figure anywhere on the site
- No self-service access, since every route to the product runs through a contact form
- No terms and conditions and no legal notice, so the service contract cannot be read before contact
- No public API documentation, although the product page advertises APIs for scaled deployments
- The claimed ISO certification is never numbered, and the trust centre that would document it only renders with JavaScript
- The site publishes two different company addresses, one in Bratislava and one in Špania Dolina
- No customer data processing agreement is mentioned and no training opt-out is documented
Pricing & Plans
Medannot publishes no pricing whatsoever. There is no rate card, no named plan and no figure anywhere on the site, and the pricing path returns a page-not-found error. No permanent free plan is announced. A free trial is offered, but on request only: both the homepage and the product page send their calls to action to a contact form. Pricing is therefore established through a sales conversation, which is consistent with the enterprise and hospital positioning the site adopts.
Data, GDPR & hosting
A consolidated view of how Medannot handles your data.
GDPR overview
GDPR compliance is claimed explicitly and in some detail. The privacy policy, effective 10 October 2024, states that Medannot adheres to GDPR and relevant privacy laws, and the product page describes the platform as fully GDPR compliant. An internal data protection officer is named, Dávid Oros, reachable at doros@medannot.com or +421 915 510 100, alongside a dpo@medannot.com role address. All eight data-subject rights are listed, access is free of charge and refusals must be explained. Transfers to the United States rely on binding standard data protection clauses under Article 46, and the policy openly notes that the United States holds no adequacy decision under Article 45. Medannot reports zero government requests for information since its founding. As a Slovak company it needs no Article 27 representative, and a Vanta trust centre is published.
Who owns the data?
Medannot publishes no terms and conditions, so ownership of uploaded scans and generated 3D models is never contractually stated. The privacy policy, effective 10 October 2024, covers personal data only: it limits collection to names, job titles, employer names and work contact details, states that Medannot does not sell personal information, and says it is shared solely with third parties facilitating delivery of the service. Clinical imaging is addressed once, on the product page, where DICOM files are described as anonymised with patient metadata stripped at upload. No clause claims ownership of customer content, but none expressly grants it either, and that gap should be settled in writing before any deployment.
Reuse rights
With no terms of service published, Medannot never restricts what customers may do with their own material, and the platform is plainly built for reuse. Segmentations and generated meshes can be downloaded as STL files directly from the segmentations panel, then opened in virtual reality, on a mobile device or inside an engineering workflow. STL files and labels can equally be imported back in, so scans of the same patient can be compared over time. Custom segmentation models trained inside the platform stay with the team that trained them, and models built on other platforms can be brought in. Medannot's own use of personal data is confined to delivering the service, understanding website usage and occasional enrichment from third parties such as LinkedIn. The policy is silent on whether customer imaging ever feeds Medannot's own model training, and no opt-out mechanism is documented anywhere.
Data retention & training
Hosting summary
Medannot runs on Amazon Web Services, with infrastructure managed through Terraform and security handled by AWS Security Hub, CloudTrail and GuardDuty, plus daily backups. The privacy policy states that personal data sits in one or more third-party databases located in the European Union and the United States, and that information collected is processed in both. The retention section is narrower still, placing the cloud database servers in the United States. Three subprocessors are named: Google, Amazon Web Services and ClickUP. Transfers outside the European Union rely on binding standard data protection clauses under Article 46, enhanced along European Data Protection Board guidance, and the policy states plainly that the United States has received no adequacy finding under Article 45. The site's own address resolves to Dublin, Ireland, on Amazon's network, which suggests European infrastructure but is not something the documents confirm. For institutions that cannot accept a cloud arrangement, the homepage states the platform is easily hostable on-premise.
Things to keep in mind
Risks and trade-offs to weigh before adopting Medannot.
- Automation complacency: a segmentation produced in seconds still needs a clinician's eye, and speed makes it tempting to accept a model without checking it against the source slices
- Deskilling: teams that stop segmenting by hand may lose the anatomical judgement they need when the AI is wrong or the platform is unavailable
- Patient data exposure: DICOM files must be anonymised before upload, and that responsibility sits entirely with the uploading institution
- Regulatory over-reading: MDR and FDA ready is not the same as certified, and treating a planning aid as an approved diagnostic device would be a serious mistake
- Unclear data ownership: with no terms and conditions published, nothing contractually settles what happens to uploaded scans or to models trained on them
- Silence on model training: the privacy policy never says whether customer imaging feeds Medannot's own models, and no opt-out is documented
- Budget and lock-in risk: no price is public, so cost, renewal terms and exit conditions only become visible once a sales process has started
Setup & Integrations
Technical difficulty
Very low for the end user. Nothing is installed: the workspace runs in a browser on any computer, and the first action is dragging a DICOM file onto the page. Model training is presented as no-code, and the Ghent case study reports complete elimination of dependence on coding, specialised hardware and external technical support. Two things raise the effort. Access is not self-service, so a sales conversation precedes any hands-on trial. And the optional on-premise deployment, along with the advertised hospital-system and PACS integration, involves the institution's IT team and is not publicly documented.
Deployment
Supported languages
Behind Medannot
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What does Medannot actually do?
Do I need to install any software?
Which file formats does it accept?
Do I need to know how to code to train a model?
How long does it take to obtain a 3D model?
How much does Medannot cost?
What happens to patient data?
What certifications does Medannot claim?
Is there a mobile application?
Who is already using it?
Should you pick Medannot?
Medannot occupies a narrow but genuinely useful niche: taking a DICOM study and returning a patient-specific 3D anatomical model, in a browser, fast enough for the result to enter a clinical routine rather than a research backlog. Two audiences meet on the same platform, clinicians planning surgery and AI teams assembling annotated datasets, and the no-code training pitch is what joins them. The named deployments at Ghent University Hospital and at Slovakia's National Institute of Cardiovascular Diseases give the claims more weight than a testimonial page usually carries, and the anatomical coverage under construction is unusually ambitious.
The reservations are all about what the site does not say. There is no published price, no plan, no self-service route in: every path leads to a contact form, so the cost of ownership only appears once a sales process has begun. There are no terms and conditions and no legal notice, which means the service contract, the liability terms and the ownership of uploaded scans cannot be read in advance. The compliance banner is confident, MDR and FDA ready, ISO and SOC 2 certified, fully GDPR compliant, but the ISO standard is never numbered and the trust centre meant to document it does not render without JavaScript. The site also gives two different company addresses.
The privacy policy, by contrast, is detailed and unusually candid, naming a data protection officer, listing subprocessors and stating outright that the United States has no adequacy decision. Read together, Medannot looks like a credible clinical tool published by a company that has invested in privacy documentation and not yet in commercial transparency. Worth a demonstration if you work with DICOM data; worth a careful contract review before deployment.
- Choosing a selection results in a full page refresh.
- Opens in a new window.