See it on your scans
Back to Features Feature — Quality

Retake before you misread

Exposure, positioning, and cropping are checked before analysis runs, so a blurry film gets flagged for a retake instead of confidently misread by a model that shouldn't have been given it in the first place.

Dentist showing a patient a printed radiograph and explaining a finding.
A surprisingly useful first step

Garbage in, confident garbage out

Most of the risk in an AI imaging tool doesn't come from the model being wrong on a good image — it comes from the model being confidently wrong on a bad one. An underexposed bitewing, a cropped panoramic missing the apex of a root, a blurred periapical: any of these can produce a finding that looks just as confident as one from a perfect image.

The quality assistant exists to catch that before it happens. It's the first thing that runs on any uploaded image, and nothing moves to analysis until it passes.

What gets checked

Five things, before anything else runs

  1. 01

    Exposure

    Too dark or too light, and fine detail in enamel and bone starts to disappear — exactly where a small lesion would be.

  2. 02

    Positioning

    Checks that the relevant anatomy is actually captured within the frame, not cut off at the edge.

  3. 03

    Contrast

    Low contrast between tooth structure and surrounding tissue makes early findings especially easy to miss, for a model or a person.

  4. 04

    Cropping

    Confirms the full region of interest is present — a periapical missing the root tip isn't useful for a periapical finding.

  5. 05

    Blur

    Motion or focus blur is flagged before it can be mistaken for — or mask — an actual finding.

Dentist and assistant treating a patient in the operatory, with imaging equipment visible beside the chair.
What happens next

Two outcomes, nothing in between

An image either passes — "Good, ready for analysis" — and moves on to tooth detection and AI analysis, or it doesn't — "Attention required: image quality may limit reliable interpretation" — and the practice knows to retake it before relying on any AI read.

There's no middle state where a marginal image gets analyzed with a quiet asterisk. The check is binary on purpose: a practice shouldn't have to interpret a quality warning to know what to do with it.

Common questions

Common questions about image quality

Does a failed quality check get logged anywhere?
Yes — it's recorded as part of the image's processing history, useful for a practice tracking whether certain equipment or positioning habits produce more retakes over time.
Can the quality threshold be adjusted per practice?
The check is calibrated against what the analysis models need to produce a reliable read, rather than a practice-configurable setting — the goal is a consistent bar, not a variable one.
Does this slow down the workflow?
No — the check runs in the same first-pass window as tooth detection, well within the seconds-not-minutes latency the rest of the pipeline is built around.
See the check in action

Run a few of your own images through it

See it on your scans See all features