ODIN — the open dental outcomes registry — Dentum
ODIN · Open Dental Intelligence Network
Pilot phase — now enrolling clinics

The open registry of dental outcomes

CBCT scans and X-rays are taken every day — and die in local archives. ODIN pools them into a de-identified outcomes registry: a clinic contributes data and gets an AI risk score, an honest benchmark, and a shared model that improves with every case.

For clinics, dentists, researchers and implant manufacturers. Pilot: 3–5 clinics in Uzbekistan.

What contributors get

AI risk score

Upload a case — get a risk estimate for peri-implantitis and implant failure. A prototype model (AUROC ≈0.77 on retrospective data), not a medical device: the clinician always decides.

“My clinic vs the network” benchmark

Your implant metrics against the network — private, visible only to you. See where you are strong and what could improve.

Showcase in the Dentum catalog

A priority clinic page in the catalog (1461 clinics, 21 cities) and a registry-contributor badge.

Data & co-authorship

Access to the de-identified pool for your own research and co-authorship in registry publications.

How it works
01

A case in 90 seconds

A short chairside form — implant, site, risk factors — plus the image. The patient's informed consent is a required field: no consent, no record.

02

De-identification at the door

Everything personal is stripped before storage: DICOM tags, EXIF metadata. CBCT volumes must pass defacing — without it 3D is not accepted.

03

Prediction and the outcome loop

The registry returns a risk score. In 6–12 months we remind you to record the outcome — the shared model keeps learning from those labels.

Open to the world — closed to leaks

Not one binary “access” but four tiers: the world gets the value — models, benchmarks, APIs — not patients' scans.

T0

Open

Who: everyone, no registration

What is shared: aggregate statistics, model weights, synthetic data, the leaderboard

T1

Registered

Who: click-through data use agreement (DUA)

What is shared: de-identified tables and CT-derived features — no pixels

T2

Credentialed

Who: signed DUA + ethics approval

What is shared: de-identified DICOM/CBCT for research; every access is audited

T3

Federated

Who: contributor clinics

What is shared: data never leaves the clinic — the model travels to it; only weights come out

Raw images default to T2/T3 only. “Openness” means open models and aggregates — not patients' faces.

Protocol principles

Consent first

Patients give informed consent to de-identified use and can withdraw it at any time.

The registry knows no names

No names, phone numbers or exact birth dates — a random case identifier. Booking and the registry are separate, unlinkable circuits.

De-identified before stored

De-identification happens at ingest, not “later”: tags, metadata, CBCT defacing. Non-de-identified data is never stored.

Data stays in Uzbekistan

Required by the personal-data law: raw data only on servers in Uzbekistan. Until UZ infrastructure is confirmed, we accept de-identified data only.

A passport for every record

Provenance, consent scope and de-identification level are fixed in a tamper-evident log — records are verifiable.

Ethics and a human in the loop

Full launch only after ethics-committee approval. AI predictions support the clinician's decision — they never replace it.

We are in the pilot phase: refining forms and process with the first clinics. The upload portal, access tiers and open datasets launch in stages — honestly, without rushing.

Apply for the pilot

We will get in touch within 1–2 business days, walk you through the process and tailor it to your practice. Application data is used only to contact you about the ODIN pilot.

FAQ
Is sharing patient data legal?

Only with patient consent and only de-identified. Raw identifiable data never enters the registry; consent can be withdrawn at any time.

What about CBCT — can't a person be identified from 3D?

Yes, via facial reconstruction — which is why every 3D volume must pass defacing before storage. Without defacing, CBCT is not accepted at all.

Who owns the data?

The clinic and the patient. The registry receives de-identified usage rights within the consent scope; every record carries a data passport with provenance and consent scope.

How much does it cost?

Free for contributor clinics: the risk score and benchmark are the value exchanged for contribution.

Is the AI score a diagnosis?

No. It is decision support (a prototype, not a medical device); the clinical decision always stays with the dentist.

Why open?

Data collected by the network benefits everyone: open weights, aggregates and benchmarks raise the standard of care — while raw data stays protected behind access tiers.

The first clinics set the standard

The pilot is co-design: we adapt collection to your workflow, and your cases drive the first version of the registry.

Join the pilot

For clinics: the pilot offer on one page →

For developers: API (DICOMweb, OpenAPI) →

Manage & withdraw consent →

Book a visit

Our AI assistant in Telegram picks a clinic and books you in a minute — 24/7.

Continue in Telegram

Instant · AI · 24/7

or
No Telegram?

We reply on Telegram or by phone during business hours (9:00–19:00).