AI + clinician, in the loop

Where every correction
trains the next model.

OrthoStudio loads intraoral-scan STLs and automatically identifies the arch, segments and numbers every tooth, traces anatomical landmarks and computes measurements — with every AI result reviewable, and every expert correction captured as attributed ground truth.

Invitation-only · de-identified data · your work follows your account

AI-segmented dental arch with per-tooth colours and landmark points
arch · segmented · numbered · landmarked
The five pillars

One scan in. A reviewed, measurable case out.

Trained models do the first pass; you stay in command of every result. Nothing is ever fabricated — an estimate is flagged, a missing landmark reads as missing, never guessed.

🧭

Arch & dentition

Upper vs. lower and permanent vs. mixed, identified on load and cross-checked against a 2D second opinion.

🦷

Segmentation & numbering

Per-vertex tooth segmentation with Palmer / FDI / Universal numbering — correct it once and it stays put.

📍

Landmark tracing

Per-tooth clinical landmarks — contacts, cusps, fossae, gingival margins, palate anatomy — placed against your confirmed segmentation.

📐

Measurements

Widths, arch dimensions and study metrics computed from the landmarks you approved, in millimeters.

🔎

Second opinion

An independent verification layer flags order, symmetry and plausibility issues — it advises, it never overrides.

🔒

Private by design

Scans are de-identified before they leave your browser; the patient-name key never reaches the server.

The learning loop

Your expertise becomes the training set.

This isn't autopilot — it's a research instrument. Every renumber, re-segmentation, boundary drag and landmark adjustment is recorded under your account, timestamped and pooled (de-identified, consent-gated) to retrain the models that assist the next case.

Upload

Drop an STL or a whole folder. It's auto-de-identified and analyzed in seconds by your own AI engine.

Review & correct

Fix segmentation, numbering and landmarks with direct-manipulation tools. Your word is final and sticky.

Contribute ground truth

Corrections export as attributed, trainer-ready samples — including blind multi-rater duplicates for inter-rater analysis.

De-identified only Consent-gated training Per-user isolation Every result reviewable
Ready when you are

Sign in and open your first scan.

Accounts are created by your study administrator. Have one? Jump straight in.