See it on your scans
Back to Oralytix How it works

The path a scan takes, and the rule that never changes

This is the literal path a radiograph takes through Oralytix, from the moment it lands in storage to the moment a finding is written into the chart. Every stage either checks the image or hands a decision back to the dentist — nothing in between.

Dental team member holding up a monitor showing four panoramic and periapical radiograph panels.
Before analysis

Getting the image right, before anything reads it

Every scan that reaches Oralytix passes through two checks before the AI forms any opinion about it. First, an image quality assessment looks at exposure, positioning, cropping, and blur — the same things a dentist would notice if a bitewing came back slightly off. If something's wrong, the image is flagged for a retake instead of being confidently misread.

Second, automatic tooth detection identifies and numbers every visible tooth. This sounds minor, but it's what makes everything downstream possible: a finding, a note, or a historical comparison is only useful if it's attached to a specific tooth, not a vague region of an image.

Why this order matters

Running quality and detection first means the analysis model never has to guess at what it's looking at, or work from an image that shouldn't have been analyzed in the first place.

Gloved hands holding a periapical radiograph up to a monitor, using a dental probe to point at a tooth root.
The read

What the model actually looks for

With a quality-checked, tooth-numbered image, the AI analysis stage runs a first pass across the full scan looking for candidate findings — caries, existing restorations, periapical changes. It isn't asked to catalog everything; it's asked to shortlist what's worth a second look.

Every candidate is drawn directly onto the image as a highlighted region, not written up as a separate line item in a report. Click a highlighted area, and a panel opens with the tooth number, a plain description of the finding, and the model's confidence.

Sample finding
Tooth 16 — possible carious lesion
Localized radiolucency on the mesial surface. Recommend visual and tactile exam to confirm.
87% model confidence
Dentist holding two printed panoramic radiographs up to an overhead light for side-by-side comparison.
Context

What changed since the last visit

If the patient has prior studies on file, Oralytix aligns them with the current image automatically — no manually flipping between two windows. The system looks for regions that appear different between studies and marks them for review, the same way a dentist would notice a shadow that wasn't there two years ago.

A note like "potential change, region of tooth 36" points to exactly where to look. The comparison itself never makes the call on what changed and why — that stays with the dentist, who can pull up side-by-side, overlay, or difference views to check it directly.

Dentist showing a patient a printed radiograph and explaining a finding.
The review

Where the dentist takes over

This is the step that makes everything before it safe to ship: every AI finding sits in a queue with exactly three possible actions, and nothing moves past this point without one of them.

  1. Confirm

    The finding stands

    It's written into the chart as a confirmed finding, in the dentist's own phrasing where relevant.

  2. Dismiss

    Not clinically relevant

    The candidate is logged and cleared — it never reaches the patient record as a finding.

  3. Modify

    Close, but not quite right

    The dentist edits the finding before it's saved — for example, correcting "possible caries" to "existing restoration."

After the decision

Documentation and history update themselves

Once a finding is confirmed, Oralytix drafts a structured chart entry from it — turning a quick note like "46 looks suspicious, compare with previous X-ray" into a complete entry the dentist edits before saving: "Tooth 46 demonstrates an area requiring clinical review; comparison with the previous radiograph is recommended." Three dentists with three different shorthands end up with consistent-looking charts.

At the same time, the patient's imaging timeline updates automatically. The next time this patient comes in, whoever opens the chart starts with full context already assembled — not a folder of loose images to sort through again.

The rule that doesn't move

AI assists. Dentist decides. Nowhere in these nine steps does a finding reach the patient chart without a human decision attached to it. That's not a policy layered on top of the product — it's steps 7 and 8 of the pipeline itself.

Questions practices ask

Common questions about how it works

How long does a first-pass review actually take?
Analysis typically returns in seconds once an image is uploaded — the design goal throughout the pipeline is to keep the model out of the way of chairside time, not add a waiting room.
What happens if the image quality check fails?
The image is flagged before any analysis runs, with a note on what to fix — exposure, positioning, or cropping. It's re-taken like any other unusable film; the AI never analyzes an image it's flagged as unreliable.
Does Oralytix ever alter the original image?
No. Findings are drawn as an overlay layer on top of the original radiograph. The source image stored on file is never modified.
Can a dentist skip the review step?
No — this is enforced by design, not left as a setting. A finding cannot be written to the chart without passing through confirm, dismiss, or modify.
See it end to end

Walk through this pipeline with your own image types

A live walkthrough runs a real bitewing, periapical, and panoramic study through every step described here.

See it on your scans See all features