AI for Dental X-Rays in 2026: What FDA-Cleared Tools Detect and How Accurate They Are

The AI overlay on a patient's bitewings shows three colored boxes your own read did not flag. Do you restore, watch or ignore them? The sales demo rarely shows what the FDA actually cleared or how the tool performed in testing.

This guide goes to the primary sources: FDA 510(k) summaries checked on October 1, 2026, plus peer-reviewed trials and systematic reviews. Below: what each tool is cleared to detect, how it performed, where it falls short, what the ADA says about billing, and how to judge a vendor.

Quick Answer

FDA-cleared dental X-ray AI tools are computer-aided detection aids, not diagnosticians. Pearl, Overjet and VideaHealth hold 510(k) clearances for marking caries, periapical radiolucency, calculus and other findings on bitewings and periapicals; Overjet and VideaHealth also have bone-level measurement tools; Diagnocat and Dentsply Sirona are cleared for periapical radiolucency on CBCT. In company reader studies, dentists found more lesions with AI but usually lost some specificity, so every flag still needs your clinical confirmation.

Which AI X-Ray Tools Are FDA Cleared, and for What?

Caries-detection software was a class III device until 2020, when the FDA reclassified dental caries image analyzers, along with mammography, breast ultrasound and lung nodule tools, into class II (product code MYN), opening the 510(k) route. The first AI caries tools all named Carestream's Logicon Caries Detector, approved through premarket approval (P980025), as their predicate. A 2026 review counted 29 FDA-cleared standalone, cloud-based dental imaging AI products from 13 companies through July 2025 and found that several newer platforms lacked independent clinical validation.

The table covers the radiograph tools a general practice is most likely to be offered. Results are from the companies' own studies as summarized for the FDA: manufacturer data.

Read the indication, not the brochure. Videa Dental Assist lists panoramic images, but only for charting past treatment and anatomy, and its periapical radiolucency findings apply only from age 22. Pearl has since added clearances for panoramic radiographs from age 16 (K250525) and for caries on periapicals (K243234).

How Accurate Were They in the FDA Studies?

Caries

In Overjet's 2022 study, 13 US dentists read 352 bitewings with and without the software, 30 days apart; sensitivity for secondary caries rose less than overall, from 49.8% to 63.0%. On its own, the software detected 72.0% of carious surfaces with 98.1% specificity. Videa's study found improved AFROC scores for each of its 21 readers, but its standalone image-based positive predictive value (the share of flagged images that truly had caries) was 59.5%.

Periapical radiolucency

Overjet's periapical study used a reference standard set by 3 endodontists. Sensitivity rose, but readers' specificity fell from 83.2% to 76.1%. On CBCT, Diagnocat raised readers' AUC from 0.894 to 0.921. DS Core Detect raised tooth-level sensitivity from 0.421 to 0.649; at its high-sensitivity setting the software alone found 0.82 of lesions larger than 8 mm³ but only 0.65 of those 8 mm³ or smaller.

Bone levels

Bone-level tools measure rather than diagnose. Videa Perio Assist (K223296, February 6, 2023) reported 92.8% sensitivity and 89.4% specificity on bitewings, but on periapicals its CEJ-to-bone-level line missed the specificity target where overlapping teeth obscured landmarks.

What Do Independent Studies Show?

A 2024 meta-analysis of 21 studies of approximal caries on bitewings pooled sensitivity at 0.94 and specificity at 0.91, yet positive predictive value ranged from 0.15 to 0.87; the authors advise verifying positive findings to prevent unnecessary treatment. Another 2024 review rated most of its 14 studies at unclear or high risk of bias and found sensitivity of 0.71 for enamel caries against 0.84 for dentin. For periapical radiolucency, a 2024 meta-analysis pooled sensitivity at 0.94 and specificity at 0.96, but nearly all studies tested AI models, not fully fledged software. For marginal bone loss, a 2025 review of 64 studies pooled sensitivity at 92.3% and specificity at 91.7%; no study was performed in a clinical setting.

External tests of real products are rarer. At a Veterans Affairs center in Los Angeles, two unnamed FDA-cleared systems were checked in 90 patients against 6 to 12 months of clinical follow-up: caries specificity was 80.17% and 83.76%, negative predictive value 96.92% and 97.36%. In a King's College London study, Diagnocat read periapical radiographs (outside its US clearance) against CBCT: its sensitivity was 47.9%, against 65.3% for experienced endodontists, with similar specificity. The pattern repeats: these tools are better at ruling disease out than ruling it in.

What Are the Limitations?

More flags can mean more drilling

In the randomized trial of 22 dentists, AI support increased non-invasive but also invasive treatment decisions. A companion cost-effectiveness analysis from the same group found identical tooth retention (a mean of 49 years) and nearly identical costs, 330 euros with or without AI, because more invasive treatment cancelled out the accuracy gain.

Specificity tends to drop

In four of the FDA reader studies above, readers' specificity fell with AI, by up to 7.1 percentage points. Because most surfaces are sound, small drops add up.

Results may not travel to your sensor

In Videa's study, standalone performance on the Schick 33 sensor was lower (AFROC 0.608 against 0.740 overall). Overjet does not guarantee accuracy below 500 by 500 pixels, and a 2026 study found higher-resolution full-mouth series improved Diagnocat's accuracy for caries by 4.3% and calculus by 9.8%.

No lesion depth

Pearl's summary states that its device does not predict lesion depth or severity. A box tells you where to look, not whether to restore.

Can You Bill for AI Analysis?

Not with a dedicated code. The ADA's Code Maintenance Committee rejected seven proposed CDT codes for AI-aided image analysis, 22 votes against, noting that these procedures are reportable with current image capture with interpretation, or interpretation and report only, codes. The ADA's announcements of CDT 2026 and CDT 2027 changes mention no AI code. For lab coding, see CDT Codes for Dental Lab Work.

How Should You Evaluate a Vendor?

1. Get the K number and read the indication in the FDA 510(k) database: image types, findings, patient ages and intended users.
2. Ask for the reader study: number of readers and images, and sensitivity and specificity with and without AI, not only an AFROC score.
3. Match your sensors and resolution to those tested.
4. Ask for independent validation. ADA Technical Report No. 1109:2025 calls for an independent validation dataset kept by a third party that is not an AI manufacturer.
5. Ask how the model changes. Overjet's 2026 clearance includes a predetermined change control plan to cut false positives and negatives using real-world data; ask how you will hear about version changes.
6. Ask where images are processed. Pearl and Overjet describe cloud processing; review data terms and the business associate agreement.
7. Pilot it on cases you have already diagnosed and count how many flags you confirm.

How Does AI Connect to Restorative Planning and Lab Cases?

Several tools flag existing work: Pearl marks discrepancy at the margin of an existing restoration, and Videa marks restorative imperfections. Those flags can turn into replacement crowns, so confirm them clinically first. Bone-level measurements can help document periodontal support before you choose between a bridge and an implant crown. AI is also entering the lab side; see AI Crown Design in Dental Labs.

Working With Elegant Dental Laboratory

• Full-service lab at 2308 McDonald Ave, Brooklyn, NY, founded in 2007: more than 35 technicians, every case made in house.
• Digital scans reviewed before production; STL and PLY files accepted from all major intraoral scanners, with direct connections for iTero, 3Shape TRIOS, Dentsply Sirona, Medit and others (submit a digital case).
• Zirconia, IPS e.max, PFM and implant restorations, with a 5 business day standard turnaround on most restorative cases and rush options.
• Free pickup and delivery by our own drivers in all five New York City boroughs and 13 northern New Jersey counties, and UPS labels for practices anywhere (send a case). Call 877-335-5221.

FAQ

Is any AI X-ray tool FDA cleared to diagnose caries on its own?

No. They are cleared as detection or measurement aids that do not replace the dentist's review or clinical judgment.

Which is most accurate: Pearl, Overjet or Videa?

The FDA files cannot say. Each company tested its own images with different metrics, so the numbers are not comparable.

Is there a CDT code for AI analysis?

No. The ADA's Code Maintenance Committee rejected proposed AI codes, saying the work is reportable with existing interpretation codes.

Does AI work on panoramic images and CBCT?

Only where cleared: Pearl has a panoramic clearance from age 16, Videa uses panoramics for charting only, and Diagnocat and DS Core Detect read CBCT for periapical radiolucency.

Can AI lead to overtreatment?

It can. In a randomized trial, dentists using AI found more enamel lesions and chose invasive treatment more often. Confirm every flag clinically.

Reviewed by Gary Fingerman, founder and CEO of Elegant Dental Laboratory, Brooklyn, New York (founded 2007). Last reviewed: October 1, 2026. FDA clearances checked October 1, 2026; study values are from FDA 510(k) summaries (manufacturers' own testing) or published research.