Kuriki VLM Report DrafterAI-drafted radiology reports from images
⚠ DEMO ONLY — not for clinical use
NOT FOR CLINICAL USE. Educational/demo only — not HIPAA compliant. Do not upload real patient data.
Stored locally in this browser only. Supported images: PNG, JPEG, WebP (max 20 MB). DICOM not supported — convert first.
About this demo
An educational tool that turns a radiology image into a structured draft report using a two-stage AI pipeline orchestrated in n8n. It illustrates how agentic, vision-language workflows can support — not replace — radiology reporting.
How it works
A vision-language model (VLM) reads the image and extracts findings.
A language model rewrites those findings into a structured report following a fixed template, guided by the optional clinical indication.
Why two models?
Separating perception (seeing) from reporting (writing) keeps each step simpler, independently swappable, and easier to validate — a core principle of agentic radiology workflows.
Limitations
VLMs can miss or invent findings. Every output is an unverified draft and must be reviewed by a qualified radiologist. Not for clinical use — never upload patient data (PHI).
Built by Paulo Kuriki, MD — AIR-Hub / Kuriki Lab, UT Southwestern · kuriki.ai
🩻
Click, drop, or paste (Ctrl+V) a radiology image here
REPORT
🩻
Image
You add a radiology image
→
👁️
Vision-language model
Reads it and extracts findings
→
📝
Language model
Drafts a structured report
Add an image and click Generate report. The result is an AI draft — always verify it against the image. New here? Tap ℹ️ About above.