Turning a dental scan into clinical understanding.
A dental scan is not a picture — it is 3D structured data with clinical meaning. Our research builds deep-learning models for that specific structure: what each surface is, where the margins run, how the arches meet, and whether the capture is even usable.
Deep learning · 3D dental anatomy · production-validated
What we work on
The preparation & margin
Finish line · reduction · undercuts
Locating the finish line to within microns — chamfer, shoulder, or feather edge — and reading reduction and undercuts. Where the margin is clean it is traced; where the prep is ambiguous the region is flagged, not guessed.
Emergence & fit
Emergence profile · contacts · clearance
Evaluating the emergence profile and interproximal contacts that decide whether a restoration seats and holds — the difference between a two-minute delivery and a remake.
Occlusion & articulation
Centric · excursions · interferences
Reconstructing how the arches meet — centric stops and excursive interferences — from the digital bite, so occlusal problems are caught before milling, not in the chair.
Scan integrity
Distortion · drag · margin capture
Recognizing the capture failures every lab knows on sight — distortion, drag, incomplete margin capture, insufficient clearance — at intake, while the patient is still there.
Anatomy-aware proposals
Adjacent · antagonist · arch form
Design grounded in the patient's own dentition — adjacent teeth, antagonist, arch form — not a generic library tooth dropped into a socket.
How a model earns trust
Cases
Real production work — the marginal, messy scans, not textbook examples.
Ground truth
Labeled by experienced technicians and clinicians — people who'd catch it by hand.
Validate
Scored against that expert judgment. A model earns trust the way a new tech does.
Defer
On an ambiguous case it says so and routes to a person. It never fakes certainty.
Compound
Every case labeled and reviewed sharpens the next one.
Measured, not asserted.
We hold every model to production standards and measure it against clinician ground truth. A capability ships when the numbers earn it — and where a model is uncertain, it says so and defers to a person, rather than guessing with confidence.
Margin agreement
How close the model's finish line sits to one an experienced technician would draw — in microns.
Warranted rescan rate
Of the scans we send back, the share a clinician confirms genuinely needed it.
Interferences caught pre-milling
Occlusal problems surfaced digitally before they reach the mill — or the mouth.
Remake root-cause coverage
Share of the failure modes that actually drive remakes that we detect at intake.
Serious AI research and real dental craft rarely sit under one roof. Here, they do — our models are built by researchers and validated by the technicians and clinicians who do this work every day.
Grounded in production
Proven on real cases, against real deadlines.
Our models are trained and validated on production work through our partnership with the Spectrum Killian Dental Lab Alliance — real prescriptions, real preps, real remakes — then shipped inside ClearPath and Orovia.