Radiology reading desk with a CT slice and an AI result on the second monitor

Services · Clinical AI selection and integration

Clinical AI that works on your own patients.

Hospitals, clinic groups and laboratories are offered AI tools for imaging, documentation and prediction, and the vendor's figures come from other patients and other scanners. Silurus Software compares the tools against your criteria, tests them on your own data before you rely on them, connects them to PACS (the image archive), the record and the LIS (laboratory information system), and sets up the oversight and monitoring the EU AI Act asks of the institution. In the US, patient data is handled under HIPAA.

Book a 30 minute call Write to us

Systems: PACS · RIS · HIS · LIS
Interfaces: DICOM · HL7 · FHIR
Regulation: EU AI Act · EU MDR · FDA
Data protection: GDPR · HIPAA

What we cover

From the first idea to safe daily use

Use case selection

Use cases ranked by benefit in your own numbers, the data they need and the strength of their evidence.

Sample use case ranking →

Vendor evaluation

Intended use, evidence and CE or FDA status (510(k) clearance, De Novo, PMA or not a device, which for decision support functions depends on the FDA clinical decision support criteria) of each product read independently, with a scoring matrix for the tender.

Sample vendor scoring matrix →

Local validation

The tool measured on your own data against a reference that fits its function before you rely on it: a retrospective set by patient group for a diagnostic tool, time and error measures for a documentation tool, and a forecast compared with the actual figures for planning.

Sample validation report summary →

Integration

DICOM (the standard for medical images and their exchange) services (C-STORE, C-FIND, C-MOVE), DICOMweb, HL7 (a standard for exchanging clinical messages) and FHIR (a web standard for exchanging health records) interfaces, single sign-on and launch from the record.

Interoperability →

Pilot

One tool in one department, with measures agreed before it starts and a written result. Where workflow effects matter, the tool runs silently on live cases before clinicians see its results.

Oversight and the EU AI Act

An internal AI register, named oversight, log retention and input to the DPIA (data protection impact assessment) where one is required.

Monitoring

Performance against confirmed outcomes, override rate and alerts per ward, at an interval set by case volume.

Staff training

Training on the tool, its limits and the oversight process, for the people who use it.

Use cases

Where AI helps in a hospital or clinic

Use cases our AI research team follows and tests in its own research and with clients. The EU MDR, IVDR (the EU regulation for in vitro diagnostic devices) or FDA status of each tool is checked, and each is measured on your own data before it is used.

Imaging worklist triage

Studies with a suspected bleed, embolism or fracture move up the radiology worklist, and the time to report is measured before and after.

A second reader in screening

An AI reader next to the radiologist, measured against double reading on your own screening data before any change to the reading protocol.

Reading external reports

Values and findings read into the record as drafts, each with its source line, and confirmed by staff.

Deterioration and sepsis alerts

Early warning scores from vital signs and laboratory values, with sensitivity and the number of alerts per ward measured on your patients before go live.

Medication safety checks

Prescriptions checked against interactions, allergies and laboratory values, with flags sent to the pharmacist for review.

Consultation notes

An AI scribe drafts the note from the consultation, the doctor edits and signs it, and minutes per note and the share of edited notes are measured.

Discharge letters

Letters drafted from the record, the medication list and the results, then checked and signed by the doctor.

Coding support

ICD-10 and DRG code suggestions from the discharge documentation, each linked to its source text and confirmed by the coder.

Follow up of incidental findings

Recommendations in radiology reports detected and tracked until the follow up study is booked or closed.

Digital pathology support

Case prioritisation and cell counting on scanned slides, with the intended purpose (what the manufacturer states the software is for) of each tool and its status under the MDR or the IVDR checked first.

The latest evidence for the case

An assistant collects the recent studies that match the patient's condition and the clinical question, with citations and the level of evidence, and the doctor reads and decides.

Patients for clinical studies

Eligibility criteria matched against the record, with a list for the study team to confirm.

Capacity planning

Forecasts of bed occupancy, emergency arrivals and missed appointments, compared with the actual figures each week.

AI research

A team that tests the tools first

Silurus Software has a dedicated internal AI research team. It follows the AI tools that reach the market, uses them in its own work and measures them on defined tasks before we propose them to a hospital or clinic. New use cases come from client projects and from our own research, and each is judged by whether it gives the patient better care, takes administrative work off doctors and nurses and lowers the cost per patient.

Silurus Software is also an industrial partner in LOGOS, an EU Horizon Europe Marie Skłodowska-Curie staff exchange led by the University of Bologna, starting in January 2027. The project connects Lie theory, geometry and quantum computing with graph neural networks and generative AI. Through exchanges of researchers and engineers, Silurus Software works on the engineering side of the project. The project is research, and none of its results is part of the services on this page.

In our own research, a laboratory report reader built on a language model running on local hardware read all 104 values from 24 synthetic laboratory reports correctly, with the source line quoted for each value and no data sent to an external service. The values go into the record as drafts, and a member of staff confirms them before they count. This was a first test on synthetic data. Before clinical use the reader is measured on the institution's own reports.

In practice

Situations we are asked to solve

Situation

A mammography AI to buy, tested on our own patients first

A hospital plans to buy a CE marked AI tool for mammography screening and wants to know how it performs on its own patients and scanners.

How we approach it

Acceptance criteria, the reference standard, the sample size and the analysis plan are agreed with your clinical lead in advance. The tool is measured on a retrospective set of your studies, and the validation report gives results with confidence intervals by scanner and patient group, written for your quality department.

Situation

AI already in use, and the EU AI Act duties

A hospital uses several AI systems and has to meet its obligations as a deployer under the EU AI Act.

How we approach it

We build the inventory of AI in use and map each system to the deployer duties that apply to it, among them human oversight, log retention and AI literacy of the staff. Which duties apply, and from which date, depends on the risk class of each system. A written AI policy and a monitoring routine with thresholds are internal controls we add on top. If the institution changes a tool's intended purpose or modifies it substantially, it can become its provider under Article 25, and the inventory flags these cases.

Situation

A prediction switched on by the record supplier

The record supplier added a sepsis and deterioration prediction without a procurement process.

How we approach it

We check its intended use and classification with the supplier in writing and measure it on your patients against a reference definition of sepsis and deterioration taken from your own records.

How it runs

From the first idea to monitoring

Step 1

Assessment

Use cases ranked by benefit, data and evidence, including the ones worth leaving alone.

Step 2

Evaluation

Products compared, a scoring matrix for the tender and written questions to each vendor.

Step 3

Local validation

The tool measured on your own data, by patient group, before go live.

Step 4

Integration and oversight

Interfaces, a pilot in one department, and the monitoring that watches performance afterwards.

What you receive

Results you can file

D-01

Evaluation and scoring

The products compared against your criteria, with the questions to the vendors.

D-02

Validation report

Data set, method and results with confidence intervals by scanner and patient group, written for your quality department.

D-03

Interfaces

DICOM, HL7 and FHIR connections, sign-in and launch from the record.

D-04

AI register and monitoring

Every AI tool in use, its owner, its logs and the thresholds that trigger a review.

EU and US

Duties in both markets

European Union

Europe

The obligations of the deploying institution under the EU AI Act, the device status under the EU MDR, and patient data under GDPR.

United States

The FDA status of each tool, and patient data under HIPAA with a business associate agreement.

Technologies and standards

What we work with

Technologies and tools

PythonPyTorchONNX RuntimeOrthancOHIFDICOMDICOMwebDICOM SRDICOM SEGHL7 v2FHIR R4SMART on FHIROpenID ConnectSAML 2.0TypeScriptReactPostgreSQLDocker

Standards, regulations and practices

AI literacyAI registerIVDREU AI ActEU MDRFDAGDPRHIPAADPIA

Silurus Software

Silurus Software, founded in 2017, offices in Budapest, Hungary, and Singapore
More than 40 engineers, EU MDR regulatory specialists and FDA consultants
Multi-year engineering partner of a European MedTech manufacturer
Founder and CTO Péter Török, in medical technology software since 1999, formerly software development director at Siemens Healthcare (syngo.via) and head of R&D at Mediso Medical Imaging Systems
ISO 9001:2015 (NQA, certificate 171041, valid until March 2027) and ISO/IEC 27001:2022 (NQA, certificate 728872, valid until August 2029)
Awarded AAA, Highest Creditworthiness, by Dun & Bradstreet in 2025
Research partner of ELTE Eötvös Loránd University, Budapest
Industrial partner of the University of Bologna, the University of Pisa, the University of Amsterdam and Masaryk University, Brno, in the EU Horizon Europe project LOGOS from 2027
Female Business Leader of the Year 2025 and The Pioneer Award 2026, awarded to CEO Dr. Linda Szántó

How we work →Quality and standards →Sample deliverables →Case studies →

Related work and pages.

Sample deliverables Interoperability Digitalisation and workflow automation Hospitals Diagnostic imaging centres Radiotherapy and oncology centres

FAQ

Frequently asked questions

Do you sell AI products?

We evaluate independently of the vendors. If we have worked for a vendor under evaluation, we tell you in writing before the evaluation starts.

Can you connect an AI tool we already bought?

Yes. We build the DICOM, HL7 and FHIR interfaces, the sign-in and the launch from the record, with the effort stated before the contract.

How do you test a tool on our own patients?

We agree acceptance criteria with your clinical lead, measure the tool on a retrospective data set and report the results with confidence intervals by scanner and patient group.

Where does our patient data go?

Where the tool runs in your systems, it stays there. Anything that has to leave is documented first: which data, where it goes and with which safeguards. The legal basis and any transfer outside the EU are decided with your data protection officer. Under the GDPR we pseudonymise where the task allows, and pseudonymised data is still personal data. Under HIPAA, data de-identified by the HIPAA standard is no longer protected health information, and where we handle protected health information on your behalf a business associate agreement applies.

What does a first assessment cost?

The first consultation is free. The assessment after it is quoted individually and agreed in writing.

How do you measure an AI tool on our data?

Acceptance criteria, the reference standard, the sample size and the analysis plan are agreed in advance. The tool is measured on a retrospective set of your own studies, and the results are reported with confidence intervals by scanner and patient group.

Does AI in the laboratory fall under the IVDR?

Software that analyses results for a diagnostic purpose can be an in vitro diagnostic device under the IVDR. We clarify the status of each tool with its supplier before it is used.

Do you train the staff?

Yes. AI literacy training for the people who use the tools, covering how the tool works, its limits and the oversight process.

Do you advise on the GDPR legal basis, secondary use of data or the EHDS?

The legal basis and the decisions on secondary use stay with your data protection officer and legal counsel. Silurus Software documents the data flows of each tool for them, with the place of storage and processing, access, logging and pseudonymisation where data leaves the clinical systems.

Talk to us

Tell us about the AI tool.

The department and the decision the tool should support. Write a few lines, or book a free call.

Book a 30 minute technical call