Ocotillo Family Dentistry patients: We've consolidated at 10450 E Riggs Rd, Suite 118. Same Dr. Suneeta Annamareddy, same insurance, same care. Learn more →
Natural Smiles Dentistry logo — Chandler Sun Lakes AZ dentist

Natural Smiles Dentistry · Blog

How AI-assisted dental X-rays catch what the human eye misses

What FDA-cleared dental AI actually does on a bitewing, what it doesn't, how it changes a routine cleaning visit at our Chandler office, the radiation question parents keep asking, and where this technology genuinely earns its keep for Sun Lakes seniors.

By Dr. Suneeta Annamareddy, DDS ~6 min read

Every set of bitewings taken at our Chandler office in 2026 is read twice. Once by the dentist sitting next to the patient, the way it's always been read. And once by an FDA-cleared software model trained on millions of labeled radiographs. The two readings agree the large majority of the time. When they don't agree, the disagreement is almost always interesting — a small cavity between two molars that the human eye glossed past, a hairline crack the AI flagged with low confidence and the dentist confirmed, calcified plaque on a root the patient hadn't noticed in years. This post is the plain version of what AI-assisted X-ray review actually does, what it can't do, and why we adopted it.

What "AI-assisted X-ray" actually means in a dental chair

The category of software you'll hear about — names like Pearl, Overjet, Videa, and a small handful of others — runs a trained computer-vision model on the X-ray immediately after it's captured. The model returns a visual overlay on the same image: colored outlines around areas where the software believes it sees a cavity, bone loss, a calculus deposit, an existing restoration, a fracture, or other findings, each with a confidence score. The dentist still reads the radiograph independently. Then we compare the two readings on the chair-side monitor with the patient watching. When the AI flags something the dentist already saw, it's confirmation. When it flags something the dentist didn't, we zoom in together and decide whether it's a real finding.

Three things are worth being clear about up front. First, this is software with FDA 510(k) clearance for clinical use in dentistry — it's not a research tool or a black box. The clearances spell out exactly which findings the model is permitted to report. Second, the dentist is still the one making the diagnosis and the treatment plan; the AI provides a second opinion, not a verdict. Third, the patient sees the overlay too. The radiograph stops being a private document the dentist interprets out loud and becomes a shared image we can both look at and discuss.

What the AI catches that the human eye genuinely misses

Three categories show up over and over in the published sensitivity studies and in our own day-to-day Chandler experience:

  • Early interproximal cavities — the small radiolucent areas between two molars that haven't yet reached the enamel-dentin border. These are the cavities a careful dentist catches at the next visit when they've grown another half-millimeter. Catching them earlier means a smaller filling or, in some cases, the chance to remineralize without drilling at all.
  • Subtle bone loss over time — the AI compares today's bitewing to the same patient's prior bitewings and quantifies the millimeter-level change in alveolar bone height. The human eye is good at noticing large changes; the software is much better at noticing slow, year-over-year changes. This matters most for early periodontal disease and for monitoring Sun Lakes patients with controlled gum disease.
  • Calculus and root-surface deposits in places the explorer didn't reach during the clinical exam. The AI doesn't replace the hygienist's tactile exam — it backs it up.

Independent peer-reviewed studies on caries-detection AI report a roughly 25–40% increase in detection sensitivity compared to a single dentist's reading alone, with specificity holding in the 80–90% range. That's the published average across multiple models; our own anecdotal experience in Chandler is consistent with the lower-half of that range — meaningful, not magical.

What it doesn't catch — and the false-positive question

The honest list of limitations, because this matters:

  • The AI cannot diagnose pain. A tooth can hurt with a perfectly clean radiograph, and a radiograph can show a cavity in a tooth that doesn't hurt at all. The clinical exam — what hurts when you bite, what hurts when cold water hits it — still drives much of the decision.
  • It can flag things that aren't there. The published false-positive rate for current caries-detection models is in the 10–20% range. We see this on Mondays — the AI sometimes calls a normal enamel shadow a cavity. That's why the second reading (the dentist) is the one that decides whether anything gets treated.
  • It doesn't replace 3D imaging for implant planning, complex root anatomy, or impacted third molars. Those still need a cone-beam scan. See our About page's premium-tech section for how AI bitewing review and 3D CBCT fit together in our Chandler office.
  • It doesn't read pediatric primary teeth as well as it reads adult teeth. Most models were trained predominantly on adult dentition. For children, the dentist's eye is still the primary reader.

If somebody is selling AI dental imaging as the cavity truth-detector, that's marketing language, not clinical accuracy. Used well, it's a careful second opinion that catches early findings the eye glossed past. Used poorly, it becomes a reason to over-treat. The dentist decides which.

What a routine cleaning visit actually looks like with AI in the loop

The minute-by-minute flow at our Chandler office, for a standard recall exam with bitewings:

  • 0:00 — Hygienist exam and cleaning. Nothing changes here. The tactile exam, the periodontal probing, the cleaning. The AI doesn't touch any of that.
  • 0:35 — Bitewings. Four small films, one of each back quadrant. Modern digital sensors mean the radiation dose for a full set of four bitewings is comparable to a one-hour transcontinental flight at cruising altitude. The image appears on the screen in about two seconds.
  • 0:36 — AI overlay loads. The software processes the image in roughly five to ten seconds and returns a color-coded overlay. The dentist already has their own read in progress.
  • 0:40 — Dentist exam with patient watching the monitor. We walk through each tooth, look at the AI's flagged areas, talk through which findings are real and which aren't, and decide the plan together. The patient sees the same image we see — usually for the first time in their dental life.
  • 0:55 — Plan and out. If the visit surfaced a small new cavity, the patient knows about it today and can decide whether to schedule the filling before they leave the chair.

The radiation question parents keep asking

AI X-ray review doesn't change the radiation exposure at all — the AI reads the same image the dentist reads. No extra films, no different sensor, no additional dose. The software is purely an image-analysis layer running after the X-ray is already captured.

On the underlying radiation question: a single digital bitewing in 2026 delivers a dose roughly 1/100th of a chest X-ray, and most adults at low cavity risk only need bitewings every 24–36 months. For kids and adolescents with active decay history, the frequency is more like every 12 months. We follow the American Dental Association's published guidelines on radiograph frequency, which are explicitly risk-based — higher-risk patients more often, lower-risk patients less. For more detail on what a kid's first-visit X-ray decision actually looks like, see our Chandler family dentist page.

Where AI X-ray review genuinely earns its keep for Sun Lakes seniors

Two patient situations make AI bitewing review especially useful, and both are common in our Sun Lakes practice:

  • Root-surface decay on older teeth. When gums recede over decades, the softer root surface gets exposed and decays faster than enamel. The cavities there are subtle on a radiograph — they don't have the high-contrast shadow a classic enamel cavity has. AI detection sensitivity on root caries is meaningfully higher than human-only reading in the published studies. For Sun Lakes patients in their 70s and 80s, this is the single most useful thing the technology does.
  • Monitoring borderline restorations year over year. Old crowns and large fillings from decades ago slowly develop tiny gaps at the margin where new decay can start. The AI compares today's image to last year's same image and quantifies the change. The human eye doesn't reliably notice a 0.3 mm change between two annual visits; the software does. For the practical Sun Lakes-specific framing on age and dentistry, our Sun Lakes dentist page walks through the broader 55+ care framework.

AI in a dental emergency

Same-day emergency visits are where AI X-ray review speeds the triage decision. A patient walks in Saturday morning with a throbbing molar. We take a periapical film, the AI flags a periapical radiolucency at the root tip, the dentist confirms — that's a tooth that needs a root canal, not a watch-and-wait. The decision to call the on-call endodontist or to start the root canal in-house gets made fifteen minutes earlier than it would have with film-and-dentist alone. For the broader picture of what a real emergency visit looks like at our Chandler office, see emergency dental in Chandler.

Patient privacy and data — the question I'd want answered

If I were the patient, here's what I'd want to know. The AI models cleared for clinical use in dentistry process the radiograph either entirely on a local server in the office or through a HIPAA-Business-Associate-Agreement cloud service that doesn't retain identifiable patient data after processing. The radiograph itself stays in the practice's existing dental record system; the AI doesn't create a parallel patient database. Ask your dentist which model they use and whether the inference runs locally or via BAA-covered cloud — both are valid; the answer should just exist.

What to ask your dentist if your office doesn't have it yet

AI-assisted radiograph review is not yet standard in every Chandler general practice — adoption is roughly a quarter of offices in 2026 and climbing. If your current dentist doesn't have it, that's not a reason to switch offices on its own. The clinical-exam-plus-experienced- dentist combination has worked well for a century, and the AI is an incremental improvement, not a categorical one. The right question to ask is whether the office's radiograph frequency and review protocol matches the ADA's risk-based guidelines, and whether the dentist sits down to walk you through the films with you on the chair-side monitor. That conversation is the actual value — the AI just makes the conversation a little more thorough.

If you'd like to see the AI overlay on your own bitewings — or you have a recent X-ray from another office and want a second opinion before committing to treatment — call us at 480-840-1101. Bring the films on a USB stick or email them ahead; we can usually have the AI read on screen before you sit down. We're a Chandler dentist who'd rather show you the data than ask you to take our word for the diagnosis.

About the author

Dr. Suneeta Annamareddy, DDS has practiced in Chandler since 2006 and owns Natural Smiles Dentistry at 10450 E Riggs Rd, Suite 118 — a family, cosmetic, and restorative practice serving Chandler and Sun Lakes. The office uses AI-assisted bitewing review on every set of X-rays, alongside 3D CBCT imaging, same-day CEREC crowns, Invisalign, dental implants, sedation, and an in-house dental plan for the uninsured.

See your own X-rays the way the AI sees them.

AI-assisted bitewing review on every set, 3D CBCT imaging, same-day CEREC crowns, sedation available, evening and Saturday hours at 10450 E Riggs Rd. Most major PPOs accepted; in-house plan for the uninsured.