AI Reads Dental X-Rays: What the FDA Clearances Show

If your dentist pulls up a bitewing and the software has already drawn a box around a shadow between two molars, you have met dental imaging AI. It is spreading fast through South Florida practices, it is regulated by the FDA, and the evidence behind it is thinner than the marketing suggests. This is what the record shows as of early 2026.

How much of it there is

A narrative review published on December 19, 2025 in the International Dental Journal searched the FDA’s 510(k), De Novo and PMA databases through July 2025 and counted thirteen companies offering twenty-nine FDA-cleared AI products for dental imaging [1]. The tasks span caries (cavity) detection, periodontal bone-level assessment, cephalometric analysis for orthodontics, multi-pathology detection, automated charting and three-dimensional segmentation of CBCT scans [1].

The pace has not slowed since the search closed. On December 18, 2025, Pearl announced FDA clearance for its Second Opinion platform on panoramic radiographs, adding caries, periapical radiolucencies and impacted third molars on the wide-view image to its earlier clearances on bitewing, periapical and CBCT imaging; the clearance was supported by a standalone performance study and a fully crossed multi-reader, multi-case study [2].

The numbers the FDA saw

Regulatory clearance for these tools rests on standalone accuracy and, for the stronger submissions, on reader studies that measure whether dentists find more disease with the software than without it. From the review [1]:

• Better Diagnostics reported 89.2 percent sensitivity and 99.5 percent specificity for caries on bitewings.
• Overjet Caries Assist reported sensitivity between 72 and 79.4 percent with specificity of 98 to 99.4 percent across image types; its pediatric module reported 83.9 percent sensitivity and 97.5 percent specificity.
• Videa Caries Assist reported 70.8 percent standalone sensitivity, and in the reader study the aided figure of merit rose to 0.739 from 0.667 unaided.

Read those carefully. High specificity means the software rarely flags healthy tooth structure. Sensitivity in the seventies means it still misses a meaningful share of real lesions on its own. The value, on the FDA’s own evidence, is as a second reader that nudges the dentist to look again, not as a replacement for the dentist looking.

What the FDA did not see

The review’s central criticism is not about the algorithms but about what surrounds them. The authors point to a lack of multi-center clinical trials to confirm generalizability, to the risk of algorithmic bias from nonrepresentative training data, and to the absence of standardized postmarket surveillance once a product is in thousands of operatories [1]. Most strikingly, several of the cleared platforms had no peer-reviewed publications supporting their clinical use at the time of the review, despite regulatory clearance [1]. Company press releases, including Pearl’s, describe improvement as “significant” without publishing the magnitude [2].

That is not an argument against the technology. It is an argument for asking what a given result means before a treatment plan is built on it.

What it means for a patient in Miami

Dental AI is being adopted quickly in Miami-Dade and Broward, often by larger groups and DSO-affiliated offices that can spread the licensing cost across many chairs. Three practical points from the desk:

1. A box on the screen is a prompt, not a diagnosis. Ask the dentist what they see, and whether the finding is visible without the overlay.
2. Early enamel lesions flagged by software do not automatically need a filling. Remineralization, fluoride and monitoring remain the standard of care for many early lesions; a flag should start that conversation, not end it.
3. If a treatment plan grows sharply after an AI-assisted exam, a second opinion is reasonable and cheap. The software is designed to raise sensitivity; the human judgment about what to treat is still the dentist’s, and yours.

The desk’s reading: FDA clearance tells you a product performed acceptably on the images it was tested on. It does not yet tell you how it performs in a busy Hialeah practice with a different X-ray sensor and a different patient population. Until the multi-center trials the reviewers call for are published, treat dental AI as a well-regulated second pair of eyes with a known tendency to under-call and over-flag, and keep the decision about your tooth where it belongs.

Sources

1. FDA-Approved AI Solutions in Dental Imaging: A Narrative Review of Applications, Evidence, and Outlook. International Dental Journal 76(1):109315, December 19, 2025 (PMC12775797).
2. Pearl press release, December 18, 2025 — Pearl expands dental AI capabilities with FDA clearance for panoramic X-rays.