Radiology API Specifications

Overview

Veltneon's radiology APIs enable secure, low-latency transmission of pixel matrices and metadata files to local NVIDIA H100 inference nodes.

In alignment with our seed-stage scope, sandbox tokens are distributed exclusively to participating beta partners. All integration targets run on isolated local networks prior to our public launch on September 16, 2026.

Authentication

All requests must include your client credentials inside the HTTP header:

Authorization: Bearer YOUR_BETA_ACCESS_TOKEN
X-Client-ID: YOUR_PACS_CLIENT_ID

POST /v1/radiology/scan

Ingests a raw de-identified chest or skeletal X-ray image array (DICOM format) into Veltneon's local de-identification buffer. Encrypts pixel coordinates prior to model parsing.

Request Payload Schema

{
  "scan_id": "scan_991823",
  "imaging_modality": "CR",
  "body_part": "chest",
  "dicom_pixel_array": "BASE64_ENCODED_PIXEL_MATRIX...",
  "meta": {
    "device_manufacturer": "GE_Medical_Systems",
    "exposure_index": "EI_241"
  }
}

Response Schema

{
  "scan_id": "scan_991823",
  "status": "ingested",
  "cache_duration_ms": 12000,
  "de_identification_status": "verified_phi_free"
}

POST /v1/radiology/evaluate

Runs AI radiology neural network models on a cached DICOM scan, returning anomalous bounding coordinates and diagnostic indicators.

Request Payload Schema

{
  "scan_id": "scan_991823",
  "minimum_confidence_threshold": 0.85
}

Response Schema

{
  "status": "evaluation_complete",
  "inference_time_ms": 124,
  "anomalies_detected": [
    {
      "finding_type": "Pulmonary Nodule",
      "confidence": 0.964,
      "bounding_box_coordinates": {
        "x_min": 1024,
        "y_min": 850,
        "x_max": 1150,
        "y_max": 960
      },
      "literature_evidence": [
        "Fleischner Society Guidelines for Incidental Pulmonary Nodules (2025)"
      ]
    }
  ]
}

PACS & DICOM Support

Auravita interfaces directly with PACS routers using standard DICOM C-STORE and C-FIND commands, automatically routing incoming CR/DX/CX scans to Veltneon’s local GPU containers.

Webhooks & EHR Write-back

Configure webhook targets inside your client dashboard. Upon evaluation completion, the platform returns an encrypted JSON containing bounding boxes and diagnostic summaries formatted as standard HL7/FHIR resources for direct write-back to Epic/Cerner charts.