Document type

Extracting data from medical imaging reports

Radiologists dictate each imaging report into reporting software, and every practice or hospital sets its own letterhead and headings. Claims and utilization review teams at insurers and workers' compensation carriers request them by the thousand, as faxed pages, records-release PDFs or patient portal exports.

Claims review needs the examination, the body part, the impression and any recommended follow-up study. A utilization review or claims system matches the impression to the billed procedure and the injury claimed, and flags whether a recommended study needs authorization.

Start for Free Talk to Us

Page 2 of a sample Medical Imaging Report, with extracted values outlined

3,959Markdown characters

36parse blocks

47fields extracted

The problem

What makes medical imaging reports hard

Facts held in narrative prose

The diagnosis, the recommended follow-up and the date findings were communicated sit inside dictated sentences under a free-text heading.

Headings vary by practice

One radiologist writes Impression and another writes Conclusion, while section order and anatomic headings follow each practice's own template.

Headers repeat on every page

Each page repeats the letterhead and patient header, and a long findings section continues across the page break under its heading.

How it works

Parse, extract, verify

Parse

Parse turns the document into reading-order Markdown and a hierarchy of blocks (text, tables, figures), and grounds each block to its exact location on the page.

Extract

You define the fields in a JSON schema, and Extract returns their values in a consistent, structured format, each linked back to where it appears in the Markdown.

Verify

Each value links back to where it appears, down to the character range in the Markdown, and the page and bounding box in the source document, so every result can be verified.

Worked example

One medical imaging report, end to end

About this document: A sample MRI report that the New York State Department of Corrections and Community Supervision published in a procurement, to show bidders the format it expects. Three portrait pages carry the exam details, findings under anatomic headings, the impression and a sign-off page. Page 2 is featured. The patient header is blank, and every doctor and facility name is a placeholder.

Processing objective: The objective is to capture the radiologist's conclusion and when it reached the care team. The impression and the bone finding behind it sit in narrative paragraphs, and the discussion date sits mid-sentence. Each value grounds to the printed line that states it.

Demonstration fields: Impression, Joint and bone finding, Findings discussed

The complete example is on GitHubThe sample document, the full extraction schema, the complete parse output and every extracted field.

Open the folder

The document

3 of 3 pages

Page 1
Page 2shown below
Page 3

Where each value came from

Page 2 of the sample Medical Imaging Report, with every featured field outlined

Impressionprimary_diagnosis

Non-displaced fracture of the medial cortex of the distal tibial metadiaphysis.

The region of the sample Medical Imaging Report containing Impression, with the extracted value outlined

extraction_metadata["primary_diagnosis"]

{
  "value": "Non-displaced fracture of the medial cortex of the distal tibial metadiaphysis.",
  "page": 2,
  "range": {
    "end": 2929,
    "start": 2823
  },
  "box": {
    "xmin": 0.11441,
    "ymin": 0.58719,
    "xmax": 0.88539,
    "ymax": 0.60357
  }
}

Joint and bone findingfindings[3].key_finding

moderate marrow edema in the anterior/lateral aspect of the distal tibia

The region of the sample Medical Imaging Report containing Joint and bone finding, with the extracted value outlined

extraction_metadata["findings[3].key_finding"]

{
  "value": "moderate marrow edema in the anterior/lateral aspect of the distal tibia",
  "page": 2,
  "range": {
    "end": 2069,
    "start": 1974
  },
  "box": {
    "xmin": 0.11395,
    "ymin": 0.29956,
    "xmax": 0.80712,
    "ymax": 0.31559
  }
}

Findings discussedfindings_communicated

12/15/2020 2:20 PM

The region of the sample Medical Imaging Report containing Findings discussed, with the extracted value outlined

extraction_metadata["findings_communicated"]

{
  "value": "12/15/2020 2:20 PM",
  "page": 2,
  "range": {
    "end": 3304,
    "start": 3230
  },
  "box": {
    "xmin": 0.11457,
    "ymin": 0.69375,
    "xmax": 0.68449,
    "ymax": 0.71059
  }
}

How to try it yourself

Everything on this page is reproducible. The sample, its schema, and the Parse and Extract API outputs are published. You can repeat the same two calls from whichever surface you already build on.

Claude Skilldescribe the job; the agent writes it
CLIno code at all
MCPexpose ADE docs as an agent tool
REST APIany language
Pythonlandingai-ade
TypeScriptlandingai-ade

Run this sample yourself This document on GitHub

Why LandingAI?

Documents into data

Accurate

The best accuracy on the difficult documents other tools get wrong, such as messy scans, complicated tables, checkboxes and handwriting.

Auditable

Every answer cites its sources. Values come back with an audit trail (page, bounding box, character range) so a human can review them.

Agentic

Applies multi-step reasoning, and plans before it acts. Handles documents it has never seen before and adapts to variations.

API-first

Composable APIs that drop into any tech stack and can be called from anywhere. Easy to build on and embed in your pipelines and products.

Affordable

Cost-competitive rates with complete price transparency. Spend controls before you run, and every job reports its exact cost.

Enterprise-grade

Built for production and scale from day one. Fully observable and monitored, running securely in our cloud or yours.

Built for high-volume, high-throughput production

Zero data retention
Priority rate limits
Enhanced support
VPC and on-prem deployment

Available on Team and Enterprise plans. VPC and on-prem deployment require an Enterprise plan.

Compare plans

Capabilities

Medical Imaging Report processing highlights

Examination and modality

Body part and laterality

Contrast use

Clinical indication

Comparison and technique

Findings by anatomic section

Abnormal section flags

Impression and primary diagnosis

Recommended follow-up studies

Findings communication date

Patient and order header

Bounding box for every value

FAQs

Questions, answered

Yes. Describe the diagnosis in your schema as the impression's first sentence, quoted verbatim. Extraction reads the paragraph for meaning and returns that statement, grounded to the line that prints it.

Yes. Define the findings as an array with the section heading, the finding text and an abnormal flag. Each anatomic heading comes back as its own entry, for as many sections as the radiologist dictated.

No. A schema names the fields you want and describes them in plain language. Extraction reads the parsed report for meaning, so one schema covers each practice's own headings and section order.

Yes. Enable zero data retention for your organization and sign a Business Associate Agreement with LandingAI. Zero data retention provides the technical safeguards for protected health information, alongside SOC 2 Type II and GDPR compliance.

Every extracted value carries the character range it was quoted from in the parsed Markdown. The Ground API maps that range to a page and a bounding box, so a reviewer sees each statement on the line that printed it.

Start extracting medical imaging report data

Reliable, structured outputs with full traceability in minutes