Architecture

Three tiers, one direction of travel.

1 · Hospital infrastructure

Your existing systems remain fully on-premises. No changes to clinical workflow, no new hardware for clinicians.

  • PACS — Orthanc, Sectra, Synapse
  • TPS — Eclipse, RayStation, Monaco
  • CT and MRI modalities

2 · On-premises gateway

A lightweight agent inside your network. It anonymises, encrypts and routes — and it is the only thing that talks to the cloud.

  • Anonymises before any cloud transfer
  • TLS 1.3 encrypted tunnel
  • DICOMweb STOW-RS / WADO-RS
  • Routes RT Struct back on completion

3 · Orcrist cloud

GPU-accelerated inference. Anonymised DICOM in, RT Struct out — and nothing kept once the job is finished.

  • Deep learning GPU inference (< 3 min)
  • Zero data retention post-inference
  • Region-appropriate infrastructure

Only anonymised DICOM crosses the perimeter. Patient-identifiable data is removed inside your network, before transfer — and zero identifiable data is retained in the cloud after processing.

Clinical safety

No RT Struct leaves without a clinician's sign-off.

The review gate is a hard architectural constraint, not a configurable preference. It cannot be bypassed, which is what keeps ContourOAI strictly a decision-support tool.

Per-structure approval

Sign-off happens structure by structure, not once for a whole set — so nothing is approved by omission.

Audit log

Every approval is timestamped and identity-linked, giving a complete record of who accepted what and when.

Version tracking

AI-generated contours and clinician-edited contours are tracked as distinct versions, so the difference is always visible.

Deployment

Cloud, region-pinned, or fully on-premises.

Patient data privacy

Full anonymisation before transfer, GDPR-compliant handling, and zero identifiable data retained after processing.

Cloud data residency

Deploy on region-specific infrastructure to meet residency obligations — EU, Mumbai or Hyderabad, depending on your requirements.

Full on-premises option

For air-gapped environments, ContourOAI deploys entirely on your own hospital GPU servers — same workflow, zero cloud dependency.

No TPS plugin required

Pure DICOMweb. Works with any DICOM-compatible PACS or TPS, with no vendor agreements and no plugin installs.

Questions

Governance and ownership

Is our training data used to improve models for other customers?

Never. Custom models are institution-private: your training data stays in your environment and is never used to train another customer's model. A model fine-tuned on your cases is available only to you.

How is contouring quality measured?

DSC and volumetric metrics are computed per structure and built into the platform, so quality can be benchmarked on your own cases rather than taken on trust.

Batch processing across large cohorts and an integrated anonymisation pipeline mean the same tooling supports prospective and retrospective clinical studies with full audit trails.

What is ContourOAI's regulatory position?

ContourOAI operates strictly as a decision-support tool. The mandatory clinician review gate is enforced architecturally and cannot be bypassed, so no AI-generated structure reaches a treatment plan without explicit clinician approval.

For current certification status in your jurisdiction, please contact us directly at support@orcrist.ai.