For patients and care teams
RadioAct helps patients and care teams understand what may need attention in chest X-rays, limb X-rays, and CT slices — grounded in HOLON's clinical knowledge, pinned to the patient's Ontomorph twin, and ready for professional review in seconds.
How it works
You'll see the same three moments in the demo video and inside the reader.
Any scan your machine already outputs — JPEG, PNG, or a DICOM slice. Nothing leaves your browser until you press Analyse.
Qwen3.6 returns structured findings with confidence and reasoning. Each is resolved against HOLON across 19 open vocabularies.
The finding lands on the correct anatomical structure of the patient's Ontomorph twin, where the patient and care team can follow it through clinician confirmation.
Built on Ontomorph
Every finding travels through the platform's clinical spine, not around it. You get real anatomy, real vocabularies, and real streams — not a UI in front of a chat completion.
Each finding is resolved to a HOLON concept, walked up its ancestry, and cross-mapped from SNOMED to ICD-10 for downstream systems.
Every pin resolves to a Foundational Model of Anatomy code, so "right upper lobe" is FMA:7311 and the platform can pull the real inspector snapshot for that structure.
Findings write back to the twin as scoped clinical events with a 3D body coordinate, not tag strings. The event carries source, region, confidence and HOLON URI.
The Clinic dashboard subscribes to twin.events.stream() across respiratory, cardiovascular, skeletal, and neurological — sorted, filterable, one-click review.
For patients and care teams
A general-purpose vision model is not a diagnostic device, and we don't pretend it is. RadioAct offers a plain-language first look for patients and care teams, always framed as possible with confidence and reasoning, then routes it for clinician confirmation.
This is an AI-assisted triage read, not a diagnosis. All findings require clinician confirmation.
Questions
No. RadioAct is a triage-assist reader. Every finding is presented as a possibility with confidence and reasoning, routed for clinician confirmation. The workflow enforces the framing — not just the marketing.
The image is sent to the vision model you configure (Groq by default) for the analysis call. It is not stored on any RadioAct server; a local database reference is kept only if you set DATABASE_URL. The patient's Ontomorph grant token controls what, if anything, gets written back to their twin.
Every finding is resolved to a HOLON concept, anchored to an FMA anatomical structure, flagged as a scoped clinical event with a 3D body coordinate, and — for a real twin — becomes the provider's inspector note for that structure. The clinical spine is the platform, not the prompt.
No. It compresses the latency on the first look, so nothing sits waiting for the second.
twin.flag with bodyCoord, twin.events.stream per system, twin.simulate, dtp.holon.concepts.search + getAncestors, dtp.holon.mappings.translate, plus the grant-authed /inspector/:fmaCode/snapshot and /notes endpoints that the SDK notes but does not wrap.
No card, no install. Every account starts in demo mode until you add your own keys.
