You finish eighteen months in clear aligners, your teeth are finally straight — and then you spot them: two little dark notches at the base of your lower front teeth, where solid pink gum used to be. Dentists call them open gingival embrasures. Everyone else calls them black triangles, and once you have noticed yours, you cannot un-notice them.
The 30-Second Version
- Black triangles — dark gaps near the gumline between teeth — are a common and stubbornly hard-to-reverse side effect of clear aligners, especially on the lower front teeth.
- A 2025 study in Progress in Orthodontics built a machine-learning model that flags, before treatment even starts, who is likely to get one.
- Blending the Invisalign ClinCheck simulation with a few clinical factors reached an AUC of 0.88 — measurably better than either ingredient alone.
- The honest caveat: one retrospective, single-clinic, Invisalign-only study, no external validation, and it still misses about a third of real cases.
It sounds like the end of post-treatment regret. But a prediction is only ever as trustworthy as the study underneath it. This model comes from Guifeng Li and colleagues at Nanjing University, published in Progress in Orthodontics in October 2025. They looked back at 297 adults treated with Invisalign — 134 of whom developed black triangles between the lower central incisors — and asked a pointed question: can we tell, from the digital plan alone, who is heading for one?
The study, in one glance
This was a retrospective cohort study: the team mined completed treatment records rather than running a live trial. They began with 23 candidate features per patient — drawn from the ClinCheck digital setup and the clinical chart — then used a statistical filter called LASSO regression to keep only the six that carried real predictive weight: age, the gingival papilla angle, interproximal reduction (the deliberate slimming of teeth), crown shape, and two kinds of planned tooth movement. Crucially, they also measured the gap area that ClinCheck’s own 3D simulation predicted for the finished result, and folded it into a combined model presented as a nomogram — a printable scoring chart a clinician can actually use at the chairside.
Why the combined model is the interesting part
On its own, the clinical checklist reached an AUC of 0.815, and the ClinCheck simulation reached 0.860 (AUC runs from 0.5, a coin toss, to 1.0, perfect). Fold them together and the combined model climbed to 0.880 — a statistically significant improvement over both. The single strongest signal was ClinCheck’s predicted gap area: each unit of increase multiplied the odds of a real black triangle almost sixfold. In plain terms, the software’s own 3D preview already “sees” the problem coming, and adding a handful of patient factors sharpens the picture. The model was also stable: a bootstrap re-test of a thousand resamples barely moved the number (corrected C-statistic 0.888), and calibration was tight, with predictions off by an average of just 1.2 percentage points. It is one more sign of AI creeping across orthodontics — from aligner design to placing cephalometric landmarks on a scan — this time pointed at an outcome patients care about intensely.
Can AI really predict black triangles — or just the easy ones?
Here is the number the headline tends to skip. The combined model’s specificity was 0.90 — it was good at reassuring the people who would stay clear. But its sensitivity was only 0.66, meaning it missed roughly one in three of the patients who actually went on to develop black triangles. For a tool meant to warn you before you commit to treatment, missing a third of true cases is not a footnote — it is the whole question. A green light from this model is reassuring, not a guarantee.
⚠ One clinic, one aligner brand, one look back
Every patient came from a single hospital in Nanjing, every case used Invisalign, and the data were gathered by looking backward at completed treatments. The authors themselves flag that they left out known risk factors such as gum biotype, and that the number of predictors was large relative to the sample — a recipe for a model that flatters itself on home turf and may stumble elsewhere.
Think of it as a weather forecast trained on one city’s history: reliable for tomorrow in Nanjing, unproven the moment you carry it somewhere else.
The model was never tested on an outside dataset — the gold standard that separates a promising idea from a clinic-ready tool. It was validated only by resampling its own data, which measures consistency, not generalisability. And because black triangles were judged from photographs by human raters rather than a hard physical measurement, the very thing the AI learned to predict carries a dose of human subjectivity.
Who owns the prediction — the software or the clinician?
A nomogram is refreshingly transparent: you can see every factor and exactly how much it counts, unlike a black-box neural network. But transparency is not the same as authority. The score estimates a probability, not a verdict, and it says nothing about what to do next. Deciding whether a 40% risk justifies more enamel reduction, a different tooth-movement sequence, or simply a frank pre-treatment conversation is still a clinical judgement — and still the clinician’s responsibility, not the software’s.
Who actually gets this — and who is left out?
The model is bolted to Invisalign’s ClinCheck, so its immediate benefit flows to patients who can afford a premium, brand-name aligner system and to clinics already invested in that software. Whether the same predictions hold for cheaper aligners, for teenagers, or for different populations with different gum anatomy is genuinely unknown. A prediction tool that only works inside one commercial ecosystem risks widening, not narrowing, the gap between the well-resourced smile and everyone else’s.
What this means for you
If you’re a patient
Considering clear aligners — especially a non-extraction case with crowded lower front teeth? Ask your orthodontist to walk you through the ClinCheck preview and your black-triangle risk before you start. Older age, thin gums and heavily rotated teeth push the odds up. Knowing early lets you weigh options like tooth re-shaping; it will not make the risk zero.
If you’re a clinician
Treat a tool like this as a structured second opinion, not an oracle. The finding that ClinCheck’s predicted gap area tracks real outcomes is worth acting on — but a low predicted risk should not override a thin-biotype, older patient sitting in your chair. Until it is externally validated on your aligner system and population, use it to open conversations, not to close them.
The bottom line
The most honest reading of this study is not “AI can predict black triangles.” It is “the digital plan you already have contains more warning than we were reading out of it.” That is genuinely useful — a well-built assistant pointing at a risk the eye tends to miss. But an assistant that catches two-thirds of cases in a single clinic is a prompt for a conversation, not a promise about your smile. The forecast is worth hearing; the decision is still human.
Frequently asked questions
What exactly is a black triangle (open gingival embrasure)?
It is the small dark gap that can appear between two teeth just below where they touch, when the gum papilla no longer fills the space. It is common after orthodontics — especially clear aligners on the lower front teeth — and it tends to be difficult to reverse without extra treatment such as tooth re-shaping or bonding.
How accurate is the AI at predicting them?
In this single study, the combined model scored an AUC of 0.88 and correctly reassured about 90% of people who stayed clear, but caught only about 66% of those who actually developed black triangles — so it misses roughly one in three real cases. It is a useful signal, not a guarantee.
Does this mean clear aligners cause black triangles?
Not on their own. Black triangles come from a mix of factors — age, gum shape, crown shape, crowding and tooth movement — and can follow any orthodontic treatment. Research cited in the study reports they appear more often on the lower front teeth after clear aligners than after fixed braces, which is part of why predicting them matters.
Can I use this tool at my next dental visit?
Not as a finished product. It was built and tested at one hospital on Invisalign cases and has not been externally validated, so it is a research model, not an approved app. But the underlying idea — reading your ClinCheck simulation for black-triangle risk before you begin — is something you can raise with your orthodontist today.
Source & author credit
This article interprets, and does not reproduce, the following peer-reviewed study. All figures are the authors’ original findings.
Li G, Guo F, Chen J, Li H, Lei L. A novel machine-learning-based model for prediction of open gingival embrasures between mandibular central incisors after clear aligners treatment: a retrospective cohort study. Progress in Orthodontics. 2025;26:39. DOI: 10.1186/s40510-025-00584-0
ORCID — no ORCID iDs were registered with Crossref for this article’s authors, so none are listed here (a Decadentry sourcing rule: we never invent identifiers).
Published open access under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 (CC BY-NC-ND 4.0) licence. Decadentry is an independent educational publication and is not affiliated with the study’s authors. Reviewed against the primary source per our editorial standards.
Leave a Reply