Grounded radiology extraction

RadExtract

Turn a free-text radiology report into structured findings, with every extraction linked to its exact source words.

Research demonstration — not for clinical use. RadExtract does not diagnose, recommend treatment, or provide medical advice.

Start with a study

Explore a cached report

Choose any modality. Results appear instantly from the demo cache.

Select a finding to trace it to the source. Edit the report and choose Process to run a live extraction.

Tap a finding to reveal its grounded source excerpt. Custom report input remains available on desktop and laptop computers.

Source → structure

Transformation Studio

Focus a structured item to illuminate its exact source span.

Choose a sample or enter a report

—Extracted items

—Source-grounded

—Significant

—Minor

Source report

LX Prompt

Structured extraction

Select a sample to begin exploring grounded findings.

Model:

Use Cache LX Prompt LX Data

Findings: General Significant Grounding

Go deeper

How RadExtract works

Grounded by LangExtract

LangExtract converts free text into schema-controlled data learned from a task-specific prompt and few-shot examples. Each extracted item retains a direct link to its source words.

Explore every layer

Switch between scan-friendly findings, the complete formatted report, raw LangExtract data, and the generated prompt. General and significant findings use distinct visual markers without implying a diagnostic confidence score.

Clinical background

Structured reporting can improve completeness, reduce ambiguity, and facilitate data sharing. Read the European Society of Radiology background paper.

Under the hood

LX Generated Prompt • LX Structured Output • Cache controls expand_more

Safety, license, and citation

This demonstration is for illustrative purposes only to show the baseline capabilities of LangExtract. It does not represent a finished or approved product, is not intended to diagnose or suggest treatment for any disease or condition, and should not be used for medical advice.

If you use RadExtract or LangExtract in production or publications, please cite accordingly and acknowledge usage. Use is subject to the Apache 2.0 License. See the README for details.