A retrospective study of 26,550 panoramic radiographs, published online in the Journal of Dentistry on October 17, 2026, found that an AI-derived score built from missing teeth and alveolar bone loss was independently associated with diabetes, hypertension, osteoporosis, ischemic heart disease and hyperlipidemia. The authors describe the score as an exploratory index, its added discrimination over age, sex and body mass index was small, and it has not been shown to diagnose anything. It adds to the evidence that the images your practice already takes carry information about general health.
As of October 2026, much of the conversation about artificial intelligence in dental radiography centres on reading the images themselves. This study asks a different question: can a single number, calculated automatically from a routine panoramic image, tell you anything about a patient's medical history? The results show a statistical association, with important limits on what that means in the operatory.
What did the study measure?
The paper, titled "AI-derived panoramic score and its association with major systemic diseases: A retrospective cross-sectional study," lists its first author at The Catholic University of Korea, with co-authors at Seoul National University School of Dentistry and the University of Iowa College of Dentistry. Two authors list their affiliation as the Artificial Intelligence Research Center of Digital Dental Hub Incorporation in Seoul, a detail worth knowing whenever a study evaluates an AI model.
According to the abstract, the team analysed 26,550 panoramic radiographs from two hospitals. An AI model detected teeth and implants and quantified alveolar bone loss. From those outputs the researchers derived a radiograph-based score from 0 to 100, built from missing teeth and weighted bone-loss severity.
They then tested whether the score was associated with five systemic conditions:
- Diabetes mellitus
- Hypertension
- Osteoporosis
- Ischemic heart disease
- Hyperlipidemia
The analysis used between-group comparisons, prevalence trends across score bins, and logistic regression adjusted for age, sex and body mass index. To judge whether the score added anything beyond those basic covariates, the authors measured the change in the area under the curve (ΔAUC) and applied the DeLong test.
What did the researchers find?
The direction of the result is consistent across all five conditions. Patients with each condition had lower scores than controls (all P<0.001). The largest adjusted difference was for diabetes mellitus (β = −6.25). The prevalence of all five conditions fell as the score rose, and the highest Q1-to-Q5 relative risk was for ischemic heart disease, at 4.73.
After adjustment for age, sex and body mass index, the score remained independently associated with all five diseases, with AUC values of 0.74 to 0.80. That is the headline. The more useful number for clinicians is the next one.
How much did the score add?
The score significantly improved discrimination over covariates alone for diabetes mellitus, hypertension and hyperlipidemia, with ΔAUC values between 0.0015 and 0.0154. It did not significantly improve discrimination for ischemic heart disease or osteoporosis.
Read those two findings together. Ischemic heart disease had the highest Q1-to-Q5 relative risk, yet the score did not add significant discrimination for it once age, sex and body mass index were already in the model. And where the score did add something, the gain in AUC was at most about 0.015. The score's added discrimination beyond age, sex and body mass index was small.
Why does this matter for dental practices in Canada?
The authors describe the score as something that "may serve as an exploratory image-based index, supporting integrated management between dental and medical care." That framing is careful, and it should stay careful when the study is repeated in a lunchroom conversation or a patient newsletter.
Three points are worth taking back to your practice, whether it is in downtown Toronto, Mississauga or a smaller Ontario community.
1. The underlying signals are ones you already record
The score is built from missing teeth and alveolar bone loss. Neither requires AI to see. Your periodontal charting and radiographic findings already capture versions of both. What the AI adds is a standardized, automatic number, which matters most at the scale of thousands of images rather than one patient in the chair.
2. It is an association, not a screening test
The design is retrospective and cross-sectional. It shows that lower scores and systemic diagnoses travel together in these two hospitals' radiographs. It does not show that tooth loss or bone loss causes these conditions, and it does not establish a cut-off at which a dentist should suspect a specific disease. The abstract does not state where the two hospitals are located, and it does not report validation in other populations, including Canadian patients.
3. No clinical product is described
The abstract does not mention a commercial product, regulatory status or availability in Canada. Any AI software marketed to your practice for this purpose would need its own evidence and its own Health Canada status checked before use.
Pro Tip: When you see advanced bone loss or multiple missing teeth on a panoramic image, check that the medical history has been updated within the last 12 months. The study does not tell you to diagnose anything, but it is a reasonable prompt to confirm what the patient's physician already knows.
Pro Tip: If an AI vendor cites this study in a sales presentation, ask two questions: was their product the model tested, and what did the added discrimination (ΔAUC) look like? In this paper it was 0.0015 to 0.0154 for the three conditions where it was significant.
How should dentists talk to patients about this?
Patients may arrive having read headlines that connect oral health and heart disease or diabetes. A study like this one can feed those headlines. A measured reply works best: research continues to find links between the mouth and general health, your radiographs and periodontal records are part of your overall health picture, and your dentist will share relevant findings with you and, where appropriate, suggest you discuss them with your physician.
Avoid language that implies an X-ray can detect diabetes or heart disease. The study does not support that. If the claim reaches your website or newsletter, it also runs into the Royal College of Dental Surgeons of Ontario (RCDSO) Advertising Guidelines, which bar statements that are "unclear, untrue or misleading."
EBIKO Dental will continue monitoring research on AI in dental imaging and its connection to medical care. More from our news desk is at ebiko.ca.
Sources
- Kahm SH, Ko DY, Choi JS, Lee JH, Cho SH, Lee SY. AI-derived panoramic score and its association with major systemic diseases: A retrospective cross-sectional study. Journal of Dentistry, published online October 17, 2026. doi:10.1016/j.jdent.2026.107104 (PMID 42892360; record on Europe PMC)
Frequently Asked Questions
Q: Can a dental panoramic X-ray detect diabetes or heart disease?
No. A 2026 Journal of Dentistry study found that an AI score based on missing teeth and bone loss on panoramic radiographs was statistically associated with diabetes, hypertension, osteoporosis, ischemic heart disease and hyperlipidemia, but the authors describe it as an exploratory image-based index.
Q: How much did the AI panoramic score improve disease prediction?
Only slightly. Beyond age, sex and body mass index, the score significantly improved discrimination for diabetes, hypertension and hyperlipidemia, with AUC gains of 0.0015 to 0.0154, and it did not significantly improve discrimination for ischemic heart disease or osteoporosis.
Q: Is this AI score available to dental practices in Canada?
The study's abstract does not mention a commercial product, regulatory status or Canadian availability. Canadian practices should confirm the evidence and Health Canada status of any AI imaging software before using it.