Everything conventional auto-contouring makes hard, done differently.
Six design decisions, each taken against a specific limitation of how auto-contouring is usually delivered.
Conventional tools
- Locked to a workstation. Delivered as a TPS plugin — accessible only from the specific system where it's installed.
- Semi-automatic. A clinician must place seed points or draw an initial contour; every structure still needs manual initialisation.
- Generic models. Pre-trained on public or pooled datasets — the same model shared across all institutions, tuned to none.
- Vendor lock-in. Clinicians must be inside the TPS to trigger contouring, tied to a specific vendor and version.
- Case by case. A clinician must be present to initiate each scan — no way to queue a full day's caseload.
- Fixed local compute. Processing degrades under caseload volume or complex anatomy.
ContourOAI
- Device-agnostic. Fully browser-based — access from any computer or tablet. No software to install, no TPS required.
- Fully automatic. Upload a CT scan and receive complete 3D structures across all slices with zero user input. Upload, review, ready.
- Your private model. Train on your own annotated cases — owned exclusively by you, reflecting your patients and protocols, never shared.
- DICOM-native. Send a CT series straight from PACS or scanner; RT Structs return automatically. No plugin, no vendor lock-in.
- Queue the whole day. ContourOAI processes every scan overnight, back to back, so your team arrives to ready-to-review contours.
- Cloud GPUs on demand. Head & Neck OARs complete in under 3 minutes — power no local workstation can match.
Beyond contouring
One platform, not a point tool.
Auto-contouring is the entry point. The same workspace handles correction, manual contouring, training data creation, and custom model deployment.
Contour Station
A cloud-based annotation workspace serving three purposes in one tool — correcting AI output, contouring from scratch, and producing training data.
- Slice-by-slice brush and eraser correction
- Full manual contouring of any structure
- Interpolation between edited slices
- Structure addition, splitting and merging
Custom model training
ContourOAI is not a closed black box. Annotate your own cases, fine-tune on them, and deploy the result alongside the base OAR models.
- Any structure — including GTV and CTV volumes
- Captures your protocols and delineation style
- Selectable at contouring time
- Institution-private, never shared
Research & education
A complete research instrument, from DICOM ingestion through to annotated dataset export — and a teaching baseline for residents and fellows.
- Batch processing across large cohorts
- DSC and volumetric metrics per structure
- Side-by-side trainee, AI and consultant comparison
- Integrated anonymisation pipeline
The annotation interface used clinically is the training data tool. Annotated cases flow directly into the custom model pipeline, so there is no separate labelling workflow to build or maintain.