Lab reports in. Coded, structured results out.
PDFs, scans and phone photos of lab reports and health records become structured data: each value with its test, a LOINC® code and a normalised unit. Confidence flags and human-in-the-loop review keep the output reliable.
Everything between a scanned page and a coded result
Built for the variety of real lab reports: different layouts, test names and units.
Layout-independent extraction
Text recognition with Google Document AI and interpretation by a large language model. No template per lab.
LOINC® coding
Each result is matched to a test in our catalogue, which carries LOINC-style codes, so the same test from two labs means the same thing.
Unit normalisation
Units are converted to a common form, so results from different labs and countries line up over time.
Confidence flags
Every value carries a confidence signal. Uncertain values are flagged instead of passed on silently.
Human-in-the-loop review
Flagged values go to a reviewer, who confirms or corrects them before they are used.
Batch and real-time
An asynchronous pipeline handles one upload from an app or a full archive of records.
A clear pipeline, with a person where it matters
Ingest
A lab report or record arrives as a PDF, a scan or a phone photo.
Recognise
OCR reads the page. A language model works out which lines are tests, values and units.
Code and normalise
Values are matched to catalogue tests with LOINC® codes and converted to common units.
Review
Low-confidence values are flagged and routed to a reviewer.
Deliver
Structured results go to your API, files or HL7® FHIR® server.
Confidence flags and review, built into the workflow
Automation handles the clear cases. When a value is hard to read, a unit is missing or a test name does not match the catalogue, the result is flagged and waits for a reviewer.
- Clear flagsUncertain values are marked, so reviewers know exactly what to check.
- Review that fits your processIn the Healsens app people check their own results. For projects, we set up the review step with your team.
- Nothing passed on silentlyFlagged values are only released after review.
value unclear on scanNeeds reviewunit not printedNeeds reviewcorrected by reviewerApprovedname differs from catalogueApprovedStructured data in the format your systems use
Results are delivered as JSON through an API, as CSV, or as HL7® FHIR® Observation resources. We agree the exact format per project.
- Test, value, unit and code for every result
- Confidence and review status included
- Deployments hosted and processed in the EU for projects
{
"document": "lab-report.pdf",
"results": [
{
"test": "Haemoglobin",
"loinc": "718-7",
"value": 14.2, "unit": "g/dL",
"confidence": "high",
"review": "not needed"
},
{
"test": "Vitamin D (25-OH)",
"loinc": "62292-8",
"value": null,
"confidence": "low",
"review": "pending" // human-in-the-loop
}
]
}Where teams use it
Incoming reports and archives
Structure lab reports from other labs, or digitise an archive of paper and PDF records so results can be followed over time.
Study data collection
Turn participants’ existing lab reports into coded data for a study, alongside the data you collect yourself.
Lab report upload in your app
Let users upload a lab report and receive structured results, without building the recognition yourself.
Frequently asked questions
Which documents can you process?
Lab reports are our core: PDFs from labs, scans and phone photos. Other health records can be processed as well; we start with a sample of yours.
How do you measure quality?
On a sample of your own documents. We show where extraction is reliable and where review is needed, and set the review step accordingly.
What is LOINC®?
An international coding system for lab tests and observations. A LOINC code tells software which test a result belongs to, whatever the lab called it.
Can you process an archive of old records?
Yes. The same pipeline runs in batches. We begin with a small sample to agree on the output and the review process.
Where is the data processed?
For projects, we host and process the data in the EU. We agree the setup with you before any data is shared.
Is this a medical device?
No. It structures and organises data. It does not interpret results medically or make clinical decisions.
See how it handles your documents
Share a few anonymised lab reports and we will show you the structured output, including what would go to review.