2. Auto-contour result
Pick a case on the left and click Run auto-contour to see the CT with AI-generated organ contours overlaid.
Live demo · Board preview
Pick a public CT case. Watch Veridose generate 100+ anatomy contours, compare against ground truth, and read the cryptographic release-attestation signed by an ABR-certified medical physicist.
Running on RTX 5070 Ti via Tailscale Funnel · Model: TotalSegmentator-v2 · Apache-2.0 weights · First inference per case ≈ 30–60 s; cached after.
Loading…
Every TG-51 / TG-142 calibration measurement is signed by the physicist's hardware credential and appended to a hash-chained ledger. The chain itself is the legal defense — any later challenge of "did you calibrate this correctly" is answered by cryptographic record. If a measurement fails tolerance, the linac is INTERLOCKED out of clinical service automatically until recalibrated.
Loading chain…
What you saw in the viewer above is the AI's voxel-precise segmentation. Below is the production handoff to the clinic's treatment-planning software. DICOM RT-STRUCT is the universal interchange format every TPS imports natively — the same format Radformation, Limbus AI, and MVision use.
The clinic's physicist opens their TPS, clicks File → Import DICOM RT-STRUCT, and picks the file. Works in every commercial TPS — no Veridose plugin needed. The demo bundle below contains the source NIfTI; the production Veridose Workstation generates the matching DICOM RT-STRUCT against the patient's actual CT.
The Veridose Workstation runs locally on the clinic PC and uses Varian's Eclipse Scripting API to drop contours directly into the open plan as ROIs. Zero file movement. The physicist sees the AI's ROIs alongside their own, can edit them, accept them, or reject them. Roadmap: shipping week 4–6.
Same automated path for RaySearch RayStation via its IronPython scripting interface. Roadmap: shipping week 6–8.
Mean DSC on the public TotalSegmentator hold-out set. Production benchmarks compare against published Radformation, Limbus AI, and MVision AI numbers for the same anatomies.
Loading benchmarks…