Can AI Detect Vertical Root Fractures? What a 2026 Research Review Found

Among all the problems an endodontist evaluates, a vertical root fracture is one of the most difficult to diagnose with confidence. It is a crack that begins in the root of a tooth and runs lengthwise — often a hairline split, frequently hidden below the gum, and notorious for hiding from ordinary X-rays. Catching it matters, because a confirmed vertical root fracture usually changes the entire treatment plan and can mean the difference between saving a tooth and losing it.

A 2026 systematic review published in the Journal of Endodontics asked a timely question: can artificial intelligence help detect these fractures? The short answer is that AI shows real promise on high-quality 3D scans — but it is not ready to replace a specialist's judgment.¹

Why Vertical Root Fractures Are So Hard to Catch

Two things make these fractures uniquely tricky. First, the symptoms are vague — lingering discomfort when biting, a deep ache, swelling that comes and goes — and they overlap with many other conditions. Second, the crack itself is often invisible on a standard flat X-ray. Conventional periapical radiographs detect fewer than one in three vertical root fractures unless the X-ray beam happens to line up almost perfectly with the plane of the fracture.¹

This is not a rare problem, either. Reported prevalence ranges from roughly 4% to 32% of affected teeth, and because an estimated 7.5%-10% of adults worldwide have at least one root-filled tooth, hundreds of millions of teeth are potentially at risk over a lifetime.¹ That combination — common, consequential, and hard to see — is exactly why better diagnostic tools matter.

What the 2026 Review Looked At

Researchers from Roseman University of Health Sciences and the Harvard School of Dental Medicine pooled the available evidence on AI for fracture detection. They searched the major medical and dental databases, screened more than 1,500 studies, and narrowed them down to the 7 that met strict quality criteria. Each study was assessed for risk of bias using the QUADAS-2 tool, and the overall strength of the evidence was graded using the GRADE approach.¹

Most of the AI systems studied were convolutional neural networks — the same family of image-recognition models already used to flag cavities and read complex root anatomy on dental images.

What They Found: AI Is Promising on 3D Scans

Across the included studies, AI models detected vertical root fractures with an accuracy ranging from about 75% all the way up to 97.8%, with sensitivity and specificity spanning similarly wide ranges. The single best result — 97.8% accuracy — came from a model applied to carefully prepared cone-beam CT (CBCT) slices.¹

The most consistent theme was that 3D imaging beat 2D imaging. Models trained on CBCT data outperformed those reading flat panoramic or periapical X-rays, and accuracy dropped noticeably when the image resolution was lower or when the software was left to pick the region of interest automatically rather than working from images a clinician had curated.¹ In other words, AI worked best when it had the same kind of high-quality 3D information a specialist relies on — and when a human was still involved in setting it up.

The Important Caveat: The Evidence Is Still Early

Here is the part that matters most, and the part the authors were careful to emphasize. The overall certainty of this evidence was rated low. The studies were small, used datasets assembled in the lab rather than from everyday clinical practice, and were rarely tested across multiple independent sites. High accuracy on a curated research dataset does not automatically translate to a real patient in a real chair.¹

The researchers' conclusion was measured: AI is best understood today as a second reader or triage aid — a tool that can shorten reading times, highlight a subtle fracture a clinician might double-check, and help sort which cases need closer surgical evaluation. It is not a standalone diagnosis. Larger, multi-center studies using real clinical scans are still needed before these tools can be trusted on their own.¹

What This Means for Your Tooth

If you have a tooth that hurts to bite on and no one can quite explain why, this research is encouraging — but the takeaway is not "wait for AI." The takeaway is that the foundation these AI tools depend on is the same foundation a careful endodontic evaluation already uses today: high-resolution 3D imaging combined with the trained eye of a specialist.

At Silicon Valley Endodontics & Microsurgery, suspected fractures are exactly the kind of case where we reach for CBCT 3D imaging and high-power microscope examination together, rather than relying on a flat X-ray alone.² When a fracture is confirmed, we walk you through the realistic options for that specific tooth — and when it is not, that 3D clarity helps us find the real cause and pursue treatment that can actually save the tooth. As AI second-reader tools mature, they will likely become one more layer of confidence on top of that process — not a shortcut around it.

The honest bottom line from 2026: artificial intelligence is becoming a genuinely useful assistant for one of dentistry's hardest diagnostic problems. But for now, the best protection for a cracked or fractured tooth is still an early, thorough evaluation by an endodontist using the right imaging.