Healthcare Document Processing

Extract data from radiology reports

Unstract runs radiology report data extraction on X-ray, CT, MRI, ultrasound and mammography reports, typed or faxed in from outside imaging centers, and sends the exam, impression and follow-up recommendation, each tied to its line, to your EHR or clinical data warehouse.

ct-chest-report-okafor-2026-10-21.pdf · faxed from an outside imaging centerstructured data
{  "patient": "Okafor, Daniel",  "date_of_birth": "07/14/1958",  "exam": "CT chest without contrast",  "exam_date": "10/21/2026",  "ordering_physician": "Laura Bekker, MD",  "impression_1": "6 mm solid pulmonary nodule, right lower lobe",  "recommendation": "Follow-up CT chest in 6-12 months per guideline",  "radiologist": "T. Ishikawa MD",  "confidence": 0.94}

Trusted by healthcare and revenue cycle teams around the globe

G2 Leader, Fall 2026G2 Best Usability, Summer 2026
Moody'sEndpointMedaxionPinpointConvrSmart ClerkCyberSoftThoughtFocusZluriIOA
What arrives

Your own reports export cleanly. Outside reports come by fax.

Reports read in your own department sit in your systems already. Reports from outside imaging centers, referring hospitals and older studies reach the HIM office as faxes, scans and printouts, each laid out its own way.

An outside CT report with a fax streak across the impression

HOSPITALIMAGING CTRCLINIC

A hospital MRI, an imaging center CT and a clinic ultrasound

EXAMIMPRESSION12 MORECOMMENDATIONROUTINESTAT

A printed mammography report with a callback date written in pen

XRCTUSMRI23 PAGES · 1 PATIENT

Five years of imaging reports scanned into one file

ADDENDUM · 2 DAYS LATER

An addendum that changes the impression, filed behind the report

HIM and care management staff read each outside report to find the one line that says "follow-up CT in 6 months", and a recommendation nobody writes down becomes a nodule nobody rechecks.

Unstract reads each radiology report, pulls the exam, impression and recommendation you name, links every value to its line, and sends the record to your EHR or clinical data warehouse.

Book a consultationFree consultation worth $1,000. Bring a few de-identified outside reports.
How it works

Read the report, structure the impression, act on the recommendation

LLMWhisperer reads the report text, Document Studio returns the structured record, and Flow carries it into prior authorization or a follow-up queue. An imaging team can start with any one of the three.

Radiology report OCR

Read radiology reports with findings and impression kept apart

A radiology report has its own order: indication, comparison, technique, findings, impression. Radiology report OCR only helps if a finding in the lungs stays under Findings and the numbered impression keeps its numbers. LLMWhisperer keeps the layout of the page, so each organ heading stays with its sentence and an addendum reads as an addendum. It reads typed reports, dictated transcripts and outside reports that came through a fax machine.

LLMWhisperer reads the report text only. It does not open or read the images. Pass the text to your own LLM, or make it the first step of a full run.

Learn more about LLMWhisperer
PATIENTOkafor, Daniel E.EXAMCT chest w/o contrastEXAM DATE10/21/2026INDICATIONCough, former smokerIMPRESSION 16 mm solid nodule, RLLIMPRESSION 2No acute processFOLLOW-UPCT chest, 6-12 months
AI data extraction

Turn each radiology report into one record of the exam and its impression

Describe the record once: patient, exam and modality, exam date, ordering physician, indication, comparison, technique, findings, each impression item, the recommendation and the signing radiologist. Document Studio then returns the same JSON from a hospital report, an imaging center report or a faxed printout. Checks you set travel with the record: an impression that mentions a nodule with no recommendation, an exam date after the signature date, an unsigned preliminary read.

Teams looking at radiology report NLP usually want the impression and the follow-up as separate fields. Unstract returns both as the radiologist wrote them, with the line each came from. It does not read the scan, make a diagnosis or assign a code.

Learn more about Document Studio
examCT chest w/o contrast✓exam_date10/21/2026✓impression_16 mm solid nodule, RLL✓comparisonNone available⚑recommendationCT chest, 6-12 months⚑⚑ no prior study on file⚑ follow-up to track
Agentic process automation

Run prior authorization from the imaging order to the payer's answer

Describe how your prior authorization team uses radiology reports, the imaging orders they support and each payer's criteria, in plain language, and Flow's agents build the workflow from that description. You approve it, and every request then runs the same way: the order and its prior imaging reports picked up, the findings and history the payer asks about pulled from the reports, the request assembled and tracked until the payer answers, with approvals posted to your EHR.

Criteria that are not met, a missing prior report and urgent cases stop for a prior authorization coordinator or a nurse reviewer, with the reason attached. A follow-up recommendation is passed on as written for your team to track. Unstract does not decide whether the follow-up is needed.

Learn more about Flow
1
Order intakeThe imaging order arrives with prior reports and insurance details
2
Read and extractParseEach report as text, sections kept in orderExtractExam, impression, ordering provider and history
3
Check the payer's criteriaIs authorization needed, and do the prior reports show what the payer asks
4
Assemble and submitPayer form filled, prior reports attached, submitted once
5
Track the decisionApprovals post to the EHR; denials go to the coordinator

Extract a stack of outside reports today. Plan the full run with us.

Start for free with a few de-identified radiology reports. Connecting it to your EHR or clinical data warehouse begins with a consultation.

Supported documents, processes & fields

X-ray, CT, MRI, ultrasound and mammography reports

Documents

  • X-ray reports
  • CT reports
  • MRI reports
  • Ultrasound reports
  • Mammography reports
  • Nuclear medicine and PET reports
  • Fluoroscopy reports
  • Interventional radiology procedure notes
  • Outside imaging reports received by fax
  • Report addenda
  • Imaging histories scanned as one PDF
  • Any other imaging report you receive

Fields captured

Patient nameDate of birthMedical record numberAccession numberFacilityExam and modalityBody partExam date and timeOrdering physicianPriorityIndicationComparison studiesSigning radiologistSignature date and time
Who it's for

Built for the teams who act on the report after it is signed

Health information management (HIM) teams

Outside imaging reports indexed by exam, modality and date, with the key fields ready for your EHR or clinical data warehouse.

Clinical documentation teams

Each impression item returned with the sentence behind it, so a query to the radiologist or the ordering provider starts from the exact line.

Care management teams

Follow-up recommendations pulled from every report into one list, so outreach for a repeat CT or a callback mammogram is not left to memory.

Utilization review teams

Prior imaging and its findings gathered from the reports before a reviewer checks a new imaging request against the criteria.

Outcomes

Fewer reports read line by line, fewer follow-ups missed

<3 minfrom a faxed radiology report to a structured record with source lines
91%of imaging reports extracted without anyone keying a field
54%fewer hours spent abstracting imaging results by hand
2×faster from an outside report arriving to a follow-up on someone's list
LLM Whisperer turns messy clinical documentation into machine-readable structure — and that has been a tremendous productivity boost.
G2Christian ZillmanCEO, Buildnext · Reviewed on G2
1 / 3
Security

Diagnostic reports carry the findings a patient may not know yet

A radiology report names what the scan found: a mass, a fracture, a nodule, sometimes before the patient has heard the result. That diagnostic PHI can affect a patient's coverage, employment and family decisions if it reaches the wrong person.

AICPA SOC 2ISO 27001GDPRHIPAA
  • Encrypted in transit and at rest.
  • Radiology reports are never used to train models.
  • Access by role, modality and department, so reports reach only the reviewers who work with them.
  • Processed reports are deleted on the retention schedule your organization sets.
  • Self-hosted deployment for health systems that keep imaging records inside their own environment.
See how Unstract safeguards PHI in the Trust Center
Questions

What HIM, care management and review teams ask about radiology reports

Have Document Studio return the recommendation and its interval as their own fields, alongside the impression item that prompted them. Each report then adds a row to a follow-up list your care management team works from, with the line the radiologist wrote, so a repeat CT or a callback has a name and a date against it.

Yes. LLMWhisperer reads faxed and scanned outside reports, including pages with streaks, skew and handwritten margin notes. A value it cannot read with certainty goes to a reviewer instead of into the record.

Findings describe what the radiologist saw, region by region. The impression is the radiologist's summary and conclusion, usually numbered. Unstract returns them as separate fields, and each numbered impression item stays its own entry.

No. Unstract reads the text of the report, not the X-ray, CT or MRI images. It returns what the radiologist wrote. It does not read scans, diagnose or assign codes. Clinical decisions stay with your clinicians.

Yes. You describe the fields once, and the same record comes back from each modality, whichever hospital or imaging center wrote it. A BI-RADS category on a mammography report comes back when the report prints one.

Faxed and scanned reports need text before any language model can work on them. LLMWhisperer supplies that text with the report's layout kept, and Document Studio extracts the fields from it. Reports that already arrive as text skip straight to extraction.

An addendum comes back as its own field with its date, next to the original impression. Your reviewers see both and which one came later, so a changed conclusion is not lost behind the first read.

Document Studio deploys as an API your developers call, or as an ETL pipeline that reads a folder of outside reports and writes records to a data warehouse you already run. Flow runs prior authorization end to end and posts results to your EHR.

Unstract is SOC 2, ISO 27001, GDPR and HIPAA compliant. Reports are encrypted in transit and at rest, never used to train models, and kept only as long as your retention schedule allows. Imaging groups that need reports on their own servers can self-host. The Trust Center publishes the controls.

A few de-identified radiology reports, ideally outside ones from different imaging centers. An HIM lead can extract fields from them for free the same day. A full prior authorization run is built with your team in a consultation, around your modalities and payers.

Next step

Bring a de-identified stack of outside imaging reports

A faxed CT, a mammography report with a callback and an MRI with an addendum. We extract them in the session and show which values would have gone to a reviewer, and why.

Start for freeOr upload one radiology report and see its record now.