Artificial intelligence that reads dental X-rays is now a regular part of practice. A July 2026 analysis by the American Dental Association’s Health Policy Institute (HPI) found that 22.8 percent of responding dentists use AI for imaging and diagnostics. Fewer than 5 percent use it to recommend treatment. That gap matches what the federal clearances actually allow. The cleared tools are built to point at things on an image. They are not built to decide what is wrong with the patient.
This article covers what the U.S. Food and Drug Administration has cleared for dental radiograph analysis, what the public records say those products do, and where their limits are.
The regulatory category
Most dental caries-detection software is cleared through the FDA’s 510(k) process as a Class II “medical image analyzer,” under 21 CFR 892.2070, product code MYN. The regulation’s own wording sets the boundaries. It describes a prescription device intended to “identify, mark, highlight” possible abnormalities on images that have already been taken. It says the device “is not intended to be used for triage, or to recommend diagnosis,” and is not meant to replace review by a qualified clinician.
The regulation also sets special controls. Manufacturers must provide clinical performance data showing the device improves reader performance in its intended population. Their labeling must state the intended patients, the compatible imaging equipment, and warnings about situations where the device may fail or underperform, such as poor image quality.
Examples from the FDA database
The following clearances were confirmed in the FDA’s 510(k) database. Indications and study figures come from the public 510(k) summaries. They are examples, not a complete list of cleared products.
Overjet Dental Assist (K210187, May 19, 2021)
Dental Assist measures mesial and distal bone levels on bitewing and periapical radiographs to support interproximal bone measurements. It is cleared under a different image-processing regulation (21 CFR 892.2050). Its indicated population is patients 22 or older with no remaining primary teeth. The summary says it “should not be used in-lieu of full patient evaluation or solely relied upon to make or confirm a diagnosis.”
Overjet Caries Assist (K212519, May 10, 2022)
Caries Assist detects and outlines suspected caries on bitewing radiographs in patients 18 and older with permanent dentition. In the reader study submitted to the FDA, 13 U.S. dentists read 352 radiographs. Their overall sensitivity rose from 57.9 percent unaided to 76.2 percent with the software. Specificity dipped from 99.3 percent to 98.4 percent. The labeling says the tool does not replace a complete review, the patient’s history or clinical findings.
Pearl Second Opinion CC (K242522, January 16, 2025)
This version of Pearl’s Second Opinion software draws outlines around suspected caries on bitewing and periapical images. It is indicated for patients 19 and older and for permanent teeth only, and is used as a second reader. Its submission relied on a standalone study of 500 images rather than a study comparing aided and unaided dentists. That study reported lesion-level sensitivity of 90 percent and an average of 1.34 false-positive marks per image. The 510(k) summary states that the device “does not replace professional clinical judgment.”
Videa Dental AI (K251002, September 19, 2025)
VideaHealth’s clearance has one of the broader scopes in the category. It covers bitewing, periapical and panoramic images and patients as young as 3, with narrower age limits for some findings. The suspected findings it flags include caries, calculus, periapical radiolucency, widened periodontal ligament, furcation and restorative imperfections. It also labels existing restorations and normal anatomy. In the reader study, 20 readers reviewed 378 radiographs. Aided reads improved significantly for two findings, caries and periapical radiolucency.
What the tools do
- Draw attention. They put boxes or outlines on regions that may contain a finding so the clinician looks again.
- Measure. Some quantify structures, such as bone levels, in millimeters.
- Visualize for patients. Marked-up images are often used chairside to explain findings. HPI found 13.2 percent of dentists use AI to explain clinical findings to patients.
What they don’t do
- Diagnose. By regulation and by labeling, the dentist makes the diagnosis, using the clinical exam and the patient’s history.
- Work on every patient or image. Each clearance is limited to specific image types, age ranges and dentition. A caries tool cleared for adult bitewings has not been cleared for a child’s periapical image.
- Cover every finding equally. A product may be cleared for some findings and not others. Even in the Videa reader study, significant improvement was shown for two of the suspected-finding categories.
- Come without trade-offs. Higher sensitivity can mean more false marks. That shows up in the small specificity drop in the Overjet study and the per-image false-positive rate in Pearl’s study. In practice, each flag still needs a clinical decision.
- Guarantee real-world results. The figures above come from manufacturers’ controlled studies submitted for clearance. Performance in a particular office depends on its imaging hardware, image quality and patients.
How dentists are using it
HPI’s July 2026 analysis, drawn from its national dentist panel, found that 43.3 percent of dentists use AI for at least one task and another 26.4 percent plan to. Imaging and diagnostics was the most common use, followed by insurance verification (13.6 percent). Dentists were much more reluctant about treatment recommendations: 82.6 percent said they do not plan to use AI for that. The concerns respondents cited included AI replacing human judgment and errors that could lead to overtreatment or misdiagnosis.
Questions for a vendor
The public record answers most of the questions an owner should ask before buying. For any product, ask for the 510(k) number and look it up in the FDA’s database. Read the indications for use: which image types, which ages, which findings. Ask whether the clearance was supported by a reader study, which measures dentists with and without the tool, or only a standalone test of the algorithm. Ask how the software handles images from your sensors and panoramic unit. Finally, decide in advance how the practice will document that a dentist, not the software, made each diagnosis.
Sources
- ADA Health Policy Institute, HPI Perspectives: Dentists’ AI usage and attitudes (July 2026)
- 21 CFR 892.2070, Medical image analyzer (Legal Information Institute)
- FDA 510(k) database, K210187 Overjet Dental Assist
- K210187 510(k) summary text (Innolitics FDA database mirror)
- FDA 510(k) database, K212519 Overjet Caries Assist
- K212519 510(k) summary text (Innolitics FDA database mirror)
- FDA 510(k) database, K242522 Pearl Second Opinion CC
- K242522 510(k) summary text (Innolitics FDA database mirror)
- FDA 510(k) database, K251002 Videa Dental AI
- K251002 510(k) summary text (Innolitics FDA database mirror)
- Dental Products Report, “Overjet’s AI Caries Assist Platform Receives FDA Clearance” (May 16, 2022)
