RadVector API
Send a de-identified finding and its study. Get back the category a validated criteria pack reached, the pack and version that reached it, the exact text the engine read, and a trace id. Deterministic by default.
Built for humans, applications and agents.
Decision
Input: 7 mm solid nodule, CT thorax, risk high
RecommendationCT at 6-12 months, then CT at 18-24 months.
From a real RadVector API response. Clinical decision support for qualified professionals.
See the response
This is a real request and the response the RadVector engine returned for it: a 7 mm solid nodule in a high-risk 64-year-old, evaluated against Fleischner 2017.
curl https://api.radvector.ai/v1/inference/evaluate \
-H "Authorization: Bearer $RADVECTOR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"study": {
"modality": "ct",
"region": "thorax",
"age_years": 64,
"sex": "M",
"attributes": {
"risk": "high"
}
},
"findings": [
{
"label": "nodule",
"region": "thorax",
"laterality": "right",
"composition": "solid",
"measurement": {
"value_mm": 7
},
"attributes": {
"multiplicity": "single"
}
}
]
}'import os, requests
resp = requests.post(
"https://api.radvector.ai/v1/inference/evaluate",
headers={"Authorization": f"Bearer {os.environ['RADVECTOR_API_KEY']}"},
json={
"study": {
"modality": "ct",
"region": "thorax",
"age_years": 64,
"sex": "M",
"attributes": {
"risk": "high"
}
},
"findings": [
{
"label": "nodule",
"region": "thorax",
"laterality": "right",
"composition": "solid",
"measurement": {
"value_mm": 7
},
"attributes": {
"multiplicity": "single"
}
}
]
},
)
best = resp.json()["outcomes"][0]["best"]
print(best["impression_category"], best["provenance"]["pack_version"])const resp = await fetch("https://api.radvector.ai/v1/inference/evaluate", {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.RADVECTOR_API_KEY}`,
"Content-Type": "application/json",
},
body: JSON.stringify({
"study": {
"modality": "ct",
"region": "thorax",
"age_years": 64,
"sex": "M",
"attributes": {
"risk": "high"
}
},
"findings": [
{
"label": "nodule",
"region": "thorax",
"laterality": "right",
"composition": "solid",
"measurement": {
"value_mm": 7
},
"attributes": {
"multiplicity": "single"
}
}
]
}),
});
const { outcomes } = await resp.json();
console.log(outcomes[0].best.recommendation);{
"trace_id": "site-example-1",
"api_version": "v1",
"outcomes": [
{
"best": {
"schema_version": "1.0",
"system": "Fleischner",
"category_id": "FLE-SOLID-SINGLE-6TO8-HIGH",
"impression_category": "Solid single nodule 6-8 mm, high risk",
"canonical_language": "Single solid pulmonary nodule measuring 6 to 8 mm in a high-risk patient.",
"recommendation": "CT at 6-12 months, then CT at 18-24 months.",
"confidence": 0.85,
"authoritative": true,
"usable_for_output": true,
"is_deferral": false,
"provenance": {
"system": "Fleischner",
"pack_version": "fleischner-2017-r2",
"validation_status": "validated",
"validated_by": "Owner-directed (founding radiologist); size/risk table audited vs Fleischner 2017 (Radiology 2017;284:228-243)",
"validation_date": "2026-08-18",
"knowledge_release_id": "",
"engine": "rc_algorithmic",
"trace_id": "site-example-1",
"citations": [],
"is_clinically_validated": true
},
"evidence": {
"type": "solid",
"multiplicity": "single",
"size_mm": 7,
"risk": "high"
},
"notes": ""
},
"systems_considered": [
"Fleischner"
],
"engine_input": "CT of the chest. 64-year-old male. Patient is high-risk. Single solid pulmonary nodule on the right measuring 7 mm.",
"engine": "rc_algorithmic"
}
]
}Captured from the RadVector API engine. verdicts[] and authoritative repeat this verdict and are omitted here. The contract has no field for a name, MRN or date of birth.
Design rules
The decision routes never call a language model. The same input returns the same verdict.
Pack system, pack version, validation status, validator and date, engine and trace id.
A deferral says why it deferred. A draft pack answers advisory. Silence is never dressed up as a decision.
No field for a name, MRN or date of birth. Free text is masked before the engine reads it, or refused.
The /v1 contract only grows by optional fields. A breaking change would be /v2.
Metered per clinical decision, with a free monthly allowance and no card to start. A call that reaches no decision costs nothing.
The /v1 surface
The same engine powers RadVector Reporting. Full request and response schemas are in the public OpenAPI reference, readable before you sign up.
| Route | What it does | Engine |
|---|---|---|
POST /v1/inference/evaluate | The central call. Structured findings (or text) plus study in; every applicable standard's verdict with provenance, the best and the authoritative verdict out. | Deterministic |
POST /v1/inference/evaluate/batch | One to 25 evaluate requests answered in order, each with its own result or error and each metered as if sent alone. | Deterministic |
POST /v1/results/normalize | Another imaging AI system's output as FHIR R4 Observations or a DICOM SR TID 1500 measurement report in; findings graded by the same engine, each verdict naming its AI source. | Deterministic |
POST /v1/context/prepare | The pre-read: current order plus prior report texts in; comparable priors ranked, prior measurements, applicable standards, open recommendations and missing information out. | Deterministic |
POST /v1/findings/normalize | De-identified report text in; canonical findings with honest confidence out. | Deterministic |
POST /v1/compare | The same finding on two studies; direction (grown, shrunk, stable, indeterminate), deltas and interval. | Deterministic |
POST /v1/qa/report | Deterministic cross-check advisories on report text, such as laterality conflicts and recommendation concordance. | Deterministic |
POST /v1/references | Published guidance and emerging evidence as two separate citation streams, with source, version and dates. | Deterministic |
POST /v1/best-practices | Published best-practice entries for an imaging question, each citing the guideline version it rests on. | Deterministic |
POST /v1/impression | The one language-model route: an IMPRESSION and RECOMMENDATIONS drafted around the verdicts, mechanically checked to stay inside them. | Deterministic + constrained LLM |
GET /v1/guidelines | The criteria-pack catalogue: edition, source, validation status, rule version. | Deterministic |
GET /v1/guidelines/{system} | One criteria pack in full: categories, canonical language, recommendations, applicability. | Deterministic |
Built to be integrated
What a hospital integration team or an AI company needs before the first production call.
Any evaluate answer as a FHIR R4 Bundle: a preliminary DiagnosticReport, an Observation per finding and per verdict, deferrals kept visible.
Another AI system's FHIR Observations or DICOM SR TID 1500 measurements, graded by the same engine; each verdict names its AI source.
RadElement codes copied from the published Open Imaging Finding Models, re-verified at build; no code is sent rather than a wrong one.
Priors ranked same-type first, prior measurements, applicable standards, open recommendations and missing information, before the read.
Same engine answers, never billed, with a published file of synthetic requests for every route.
Pin a pack version; a stale pin answers 409 instead of silently changing recommendations.
Every verdict carries a code from a closed set, so integrations branch on a value, never on prose.
Up to 25 evaluations in one call; an Idempotency-Key replays the stored answer without re-running or re-billing.
Criteria packs
Only a validated pack with a named validator and date can produce a verdict marked usable for output. Everything else answers as advisory or defers, and says why.
Validated · may drive output
Every verdict names who validated its pack and when.
Draft · answers as advisory
Draft packs are marked not usable for output. A deferral always says why.
Plain-language guides: Lung-RADS, BI-RADS, TI-RADS, LI-RADS, O-RADS, PI-RADS, Bosniak, Fleischner, DEXA.
Who calls it
That is the conclusion RadVector API is designed to earn from platform teams, AI companies and imaging vendors.
Model and agent developers
Let a general-purpose model hand radiology criteria to a maintained, versioned engine. Explicit deferrals tell the agent when not to answer.
Learn more: Model and agent developersPACS, RIS and enterprise imaging
Keep your viewer, worklist and report. Call RadVector for the criteria step, with no patient identifiers crossing your boundary.
Learn more: PACS, RIS and enterprise imagingEHR and order entry
Versioned verdicts with a trace id can sit next to the order or the result, and be audited later against the pack version that produced them.
Imaging-AI vendors
A detection model tells a clinician that something is there. It rarely tells them what to do about it. RadVector supplies the category and recommendation that follow.
Get started
The example above is a real engine output.
Interactive reference, no account needed.
Developer portal: passkeys, Google, GitHub or email.
Free monthly allowance, then pay per decision. Pricing.
RadVector API is a clinical inference API for radiology from iTech Clinical Intelligence. You send a de-identified finding and its study; it returns the category a validated criteria pack reached, the pack and version that reached it, the exact text the engine read, a trace id, and optional references. Its deterministic routes never call a language model. Documentation is public at https://api.radvector.ai/v1/docs.
A language model recalls a guideline; RadVector applies one. RadVector API evaluates structured findings against a versioned criteria pack and returns the same answer for the same input, with the pack version, validation status and trace id attached. When the criteria do not apply, for example because a nodule has no stated composition or the patient is below the pack's minimum age, it returns a deferral that says why instead of guessing.
Three criteria packs are validated and may drive output today: Fleischner 2017 for incidental pulmonary nodules, ACR BI-RADS 5th edition, and DEXA bone-density classification (WHO 1994 with ISCD 2019). Draft packs for Lung-RADS v2022, TI-RADS, PI-RADS v2.1, O-RADS US, LI-RADS, Bosniak 2019, C-RADS and NI-RADS answer as advisory and are marked not usable for output while they are validated. Each standard has a plain-language guide at itecclinical.com/standards.
Yes. Agents calling RadVector API as a tool is a first-class use. The API takes structured input, returns structured verdicts with provenance, and says explicitly when it defers, which lets a general-purpose model hand radiology criteria to a maintained, versioned engine instead of reciting them from memory. The one language-model route, /v1/impression, drafts an impression constrained to the verdicts and mechanically checks that the prose stayed inside them.
Yes. Every POST /v1/inference/evaluate answer is available as a FHIR R4 Bundle (send Accept: application/fhir+json or add ?format=fhir): one preliminary DiagnosticReport, one Observation per finding and one per verdict, with the criteria pack, version and validation status on each verdict. A deferral is an Observation with dataAbsentReason, so an undecided finding stays visible. Findings carry verified RadElement codes where a verified code exists.
Yes. POST /v1/results/normalize accepts another AI system's output as FHIR R4 Observations or as a DICOM SR TID 1500 measurement report in the DICOM JSON model. Each finding keeps its source algorithm and version, is graded by the same engine against Lung-RADS, BI-RADS, TI-RADS, Fleischner and the other packs, and every verdict names both the AI source and the criteria pack. Patient fields are never read. RadVector grades findings another system reported; it detects nothing itself.
With a sandbox key and synthetic fixtures. A sandbox key returns the same engine answers as a live key, is never billed, and marks every answer X-RadVector-Sandbox: true; a published file of synthetic request bodies covers each route. Integrations can also pin criteria-pack versions (a stale pin answers 409 instead of silently changing recommendations), branch on closed-set deferral codes, send up to 25 evaluations in one batch, and retry safely with an Idempotency-Key. The developer guide is at https://radvector.ai/portal/developers.
POST /v1/context/prepare assembles what a radiologist needs before reading: from the current order and prior report texts it returns the comparable priors ranked same-type first with their intervals, prior measurements bound to their lesions, the reporting standards that apply with their required elements, recommendations from earlier reports that are still open, and missing-information warnings. It calls no language model.
No. RadVector Reporting and RadVector API are clinical decision support for qualified professionals. They have not received FDA clearance or approval as a medical device, are not a diagnostic device, and must not be used for autonomous diagnosis, triage or unsupervised reporting. The radiologist remains the final clinical authority.
RadVector API is built to need as little protected health information as possible. The contract has no field for a name, MRN, accession number or date of birth; age and sex are accepted because they change the clinical answer. Free text is passed through an identifier mask before the engine reads it, the call is refused if masking is unavailable, request text is never logged or stored, and customer data is never used for training. Self-serve keys are for de-identified input. For clinical deployments, including AI and clinical platforms that call RadVector on behalf of hospitals, iTech signs a Business Associate Agreement before PHI-bearing calls begin.
Read the public reference at https://api.radvector.ai/v1/docs without an account, then create an account at https://radvector.ai/portal to mint a key. Every account includes a free monthly allowance of decisions with no card required. Pricing is metered per clinical decision; a call that reaches no decision is not billed.
Built to need as little PHI as possible, with a BAA for clinical deployments. RadVector API is clinical decision support; it is not a diagnostic device, triage engine or unsupervised reporting workflow.