New AI Reads Dental Scans to Spot Jaw and Bite Problems

If you've ever had braces, you know the drill: a dentist studies your teeth, measures angles on a scan, and decides how bad your bite really is. That measuring is slow, and two dentists can look at the same scan and reach different conclusions. Researchers have now built an AI system, TeethGNN, that does this grading automatically.

The system works from cone-beam CT scans (a 3D X-ray of your mouth, common in orthodontics). Instead of a human drawing lines and angles, the AI predicts the key measurements itself, then combines those numbers with the raw image using a graph neural network — think of it as a map where each tooth is a dot and the AI studies how the dots relate to each other. The team tested it on their own patient scans and got roughly 77% accuracy, with a stronger 89.6% score on its ability to rank severity correctly, beating other published methods.

Why should you care? Orthodontic treatment is expensive — often thousands of dollars — and part of that cost is the expert time spent diagnosing. If software can do the first pass reliably, appointments could get shorter, waiting lists could shrink, and patients in places with few orthodontists could get a faster read on whether they need care.

The honest catch: 77% accuracy means the AI would be wrong roughly one time in four, which is nowhere near good enough to replace a dentist. The tests were also run on the researchers' own dataset, not on scans from a wide range of clinics and X-ray machines, so real-world performance is still unproven. Expect this to arrive first as a tool that flags cases and double-checks a human, not one that makes the call alone.