AI in laboratory diagnostics: how to assess specialist vendors such as medicalvalues

What does AI do in laboratory diagnostics?

A laboratory produces a great deal of structured data: measured values, reference ranges, trends over time. AI comes in at several points:

• Validation: flag conspicuous values, trigger repeat measurements.
• Diagnostic hints: derive suspected diagnoses from combinations of values and recommend further tests.
• Requests: help referring physicians choose sensible parameters.
• Report texts: draft comments on findings, including versions that patients can understand.

The first three intervene in diagnostics; the report text sits in between, depending on whether it only phrases or also assesses.

What does medicalvalues offer?

medicalvalues, based in Karlsruhe, was built from the start as an AI vendor for laboratory medicine. According to its website, its products include:

Product	What it does according to the vendor
Diagnostic Intelligence	analyses patient data, detects conspicuous values and possible suspected diagnoses, recommends further investigation; Class IIa medical device under the MDR
Lab Intelligence	modular laboratory information system with AI-assisted validation, report generation with language models and billing
Order Intelligence	support with requests, with chat
AI Reports	report texts from diagnostic results and patient context, editable before release

According to its own information, medicalvalues is certified to ISO 13485 (certification body DQS) and offers operation in the cloud or on the customer’s premises. The website does not give customer numbers or reference hospitals. Whether the individual products other than Diagnostic Intelligence are medical devices is not stated on the product pages; ask the vendor.

Details in the comparison with medicalvalues.

In addition, vendors of laboratory information systems are building AI into their own systems. This is the “specialist platform with AI” pattern: specialist platform or AI platform?

When is laboratory AI a medical device?

What matters is the intended purpose set by the manufacturer. Software that provides information for the diagnosis or treatment of individual patients is generally a medical device under Art. 2(1) MDR, and under Rule 11 in Annex VIII MDR at least Class IIa. If software analyses data from in vitro examinations, it may instead fall under the In Vitro Diagnostic Medical Devices Regulation (IVDR). The guidance MDCG 2019-11 (revision of June 2025) explains how to draw the line.

For the laboratory, this means:

• Suspected diagnoses, recommendations for further investigation, automatic assessment of values: medical device or in vitro diagnostic medical device, with CE marking.
• Organisation, documentation, information on guidelines and standards: generally not a medical device.
• If a hospital builds an application with a medical intended purpose itself, it becomes its manufacturer. More on this: Own AI apps and the MDR.

How do you assess a specialist vendor?

An AI-native specialist vendor is no substitute for an AI platform, and vice versa. It solves one specialist task in depth, often with regulatory approval; the platform covers the breadth of the hospital. Questions for the selection:

1. What is the intended purpose of each product, and which is a medical device or in vitro diagnostic medical device, of which class?
2. How does data get from the laboratory information system and the hospital information system (HIS) into the product, and how does it get back?
3. Where does the AI run: in the vendor’s cloud or on your premises?
4. Who checks and releases results, and how is this documented?
5. What evidence of benefit is there, such as studies or reference hospitals?

The overall picture: Horizontal AI platform or AI in clinical systems.

How does this work with basebox?

basebox is not a medical device; it is a platform for general tasks, without a medical intended purpose. Diagnostic analysis remains with approved products. In the laboratory and in diagnostics, basebox comes with use cases that need no diagnosis, for example:

• Capturing request forms with text recognition
• Commenting on proficiency test results in writing
• Maintaining the documentation of validation rules
• Information on the RiliBÄK
• Maintaining the catalogue of text modules for reports in microbiology and pathology
• Reading equipment inspection cards and compiling deadlines

For tasks involving patient data, basebox delivers drafts and summaries with sources, no diagnosis and no treatment suggestion; the doctor checks every result. We connect a laboratory system as soon as its vendor provides access. All 146 use cases for hospitals.