AI tooth segmentation is the crucial first step where software analyzes a raw intraoral scan to understand its surfaces before any restoration is designed. Because a scan initially arrives as an undifferentiated mesh, the software must identify each individual tooth, the gum line, and their boundaries—much like how a dental technician intuitively reads a physical model at a glance. Everything downstream depends on this accuracy; a design engine that misreads the prep is essentially designing for a tooth that isn't there.
How does AI segment the teeth in an intraoral scan?
The AI model evaluates the mesh and assigns each section to a specific structure, distinguishing a molar from the gingiva and marking the exact boundary between them. Just as a technician instantly recognizes teeth on a model without conscious effort, segmentation is the software equivalent. To be effective, it must perform reliably on completely new cases—handling crowded arches, worn cusps, heavily compromised preps, and teeth that are far from textbook perfect.
What is supervised learning, and how does it train segmentation?
Supervised learning is a branch of AI that trains on labeled examples. If you show a model enough identified images, it eventually learns the defining features and can recognize them in completely new scenarios. When applied to annotated intraoral scans—where dental experts have explicitly marked each tooth and gum line—the model learns to distinguish an incisor from a molar, and pinpoint exactly where a tooth ends and the soft tissue begins.
This is the same class of AI used to analyze bitewing X-rays for early signs of decay (more on where AI shows up in dentistry). It focuses purely on recognition, not creation. Actually designing the crown requires a different type of AI—generative AI—which we explore further in our guide on how AI crown design works.
Why does the margin line matter more than any other boundary?
While segmentation identifies many boundaries, one stands above the rest: the margin line. As the precise point where the restoration meets the prepared tooth, it is the single most consequential line in the entire workflow. Get it right, and the crown seats cleanly. Get it wrong, and no downstream effort—whether it's the material, the mill, or the polish—can save the fit.
On a clean prep with a distinct finish line, this boundary is obvious to both a trained AI and a human. However, when the margin is subgingival or obscured by tissue and fluid, it becomes a judgment call. This is precisely why margin placement is always worth confirming with your own eyes, rather than accepting on faith, regardless of how capable the software handles the initial proposal.
What makes tooth segmentation succeed or fail?
On the training side, three core elements must align:
- Labeled Scans: These must be annotated by experts who understand dental anatomy. The model can only learn from what was marked correctly.
- Model Design: This determines the AI's ability to pick up on subtle boundaries, unusual anatomy, and damaged preps.
- Computing Power: Sufficient processing power is required to train on massive scan libraries and deliver fast results while the patient case is still open.
On the input side, success comes down to the quality of the scan itself. Missing data from scanner blind spots, a distorted mesh, soft tissue draping over the finish line, or a rushed scan all lead to a compromised input—and a bad input inevitably leads to a guessed output. Since the AI will generate a result either way, it is crucial that the software highlights areas of uncertainty rather than presenting every estimation with absolute confidence. Catching problematic scans before moving into the design phase is an essential discipline known as scan quality control.
What happens after the scan is segmented?
Once the arch is fully separated and the margin is precisely identified, the design phase finally has the context it needs: which specific tooth is being restored, its relationship with neighboring teeth, and exactly where the restoration must meet the prep. From there, a generative AI model can propose a custom crown tailored to that specific mouth, rather than relying on a generic shape.
Tooth segmentation might happen quietly in the background, but it is the foundational step that makes everything else possible. See how the full pipeline runs, from scan to proposal, on the Dentscape CAD page.
Go deeper: