Graph representation of the oral cavity as a method for screening the general condition of the patient

AIM. To analyze the ability of the oral cavity, represented as a patient–tooth graph and processed by a graph neural network, to reflect systemic health, using the detection of concomitant cardiovascular disease as an example.

MATERIALS AND METHODS. On a large population-based cohort, each patient was represented as a heterogeneous patient – tooth graph encoding periodontal status and dentition configuration, and a Heterogeneous Graph Transformer with attention was trained. The capacity of the structure to carry a systemic signal was probed two ways: ablation of the graph structure while keeping the aggregate DMFT index and the graph contribution in a dental-chair scenario (“age, sex + tooth graph” vs “age and sex only”). The cardiovascular example comprised atherosclerotic cardiovascular disease (ASCVD) and chronic heart failure (CHF); discrimination was assessed by AUC with the DeLong test and applicability by an operating point.

RESULTS. The cardiovascular signal resided in the graph structure: removing the tooth structure while keeping the DMFT index significantly reduced AUC for both cardiovascular outcomes – CHF (−0.024; p = 0.026) and ASCVD (−0.011; p = 0.032). In the dental-chair scenario the graph yielded an AUC gain of +0.083 for CHF (p = 2 × 10^-⁴) and +0.030 for ASCVD (p = 1.5 × 10^-⁵); at ~0.80 sensitivity the model reached 0.70–0.74 specificity and 0.97–0.99 negative predictive value.

CONCLUSION. This is the first confirmation that a graph representation of the oral cavity can be “read” as a map of systemic health: the cardiovascular signal lies in the structure of the dentition and is irreducible to aggregate indices. Periodontal examination is applicable as a screening point for concomitant cardiovascular disease with referral to a physician.

1. Sanz M., Marco Del Castillo A., Jepsen S., Gonzalez-Juanatey J.R., D’Aiuto F., Bouchard P. et al. Periodontitis and cardiovascular diseases: Consensus report. J Clin Periodontol. 2020;47(3):268–288. https://doi.org/10.1111/jcpe.13189

2. Tonetti M.S., D’Aiuto F., Nibali L., Donald A., Storry C., Parkar M. et al. Treatment of periodontitis and endothelial function. N Engl J Med. 2007;356(9):911–920. https://doi.org/10.1056/NEJMoa063186 (Erratum in: N Engl J Med. 2018;378(25):2450. https://doi.org/10.1056/NEJMx180022).

3. Fröhlich H., Herrmann K., Franke J., Karimi A., Täger T., Cebola R. et al. Periodontitis in chronic heart failure. Tex Heart Inst J. 2016;43(4):297–304. https://doi.org/10.14503/THIJ-15-5200

4. Nabila S., Choi J., Kim J.E., Hahn S., Hwang I.K., Kim T.I. et al. Bidirectional associations between periodontal disease and systemic diseases: a nationwide population-based study in Korea. Sci Rep. 2023;13(1):14078. https://doi.org/10.1038/s41598-023-41009-4

5. Shefov V.Y., Orekhova L.Y., Loboda E.S., Shefova A.V. Limitations of linear statistical methods for detecting associations between dental status and systemic patient’s health. Parodontologiya. 2026;31(1):61–76. (In Russ.) https://doi.org/10.33925/1683-3759-2026-1210

6. Losev F.F., Sorokina A.A., Salakhov A.K., Dokin S.P. The use of artificial intelligence in modern dentistry in the Russian Federation. Stomatology. 2024;103(5):42–45. (In Russ.) https://doi.org/10.17116/stomat202410305142

7. Oisieva K.Sh., Rozov R.A. Artificial Intelligence in Dentistry: A Sign of the Times. Stomatology. 2025;104(1):87–92. (In Russ.) https://doi.org/10.17116/stomat202510401187

8. Kinane D.F., Stathopoulou P.G., Papapanou P.N. Periodontal diseases. Nat Rev Dis Primers. 2017;3:17038. https://doi.org/10.1038/nrdp.2017.38

9. Hu Z., Dong Y., Wang K., Sun Y. Heterogeneous graph transformer. In: Proceedings of the Web Conference 2020 (WWW ‘20), Taipei Taiwan, April 20–24, 2020. New York, NY: Association for Computing Machinery; 2020, pp. 2704–2710. https://doi.org/10.1145/3366423.3380027

10. Schlichtkrull M., Kipf T.N., Bloem P., van den Berg R., Titov I., Welling M. Modeling relational data with graph convolutional networks. In: Gangemi A., Navigli R., Vidal M.-E., Hitzler P., Troncy R., Hollink L. et al. (eds) The Semantic Web (ESWC 2018). Lecture Notes in Computer Science, vol 10843. Cham: Springer; 2018, pp. 593–607. https://doi.org/10.1007/978-3-319-93417-4_38

11. Collins G.S., Reitsma J.B., Altman D.G., Moons K.G.M. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): The TRIPOD statement. Ann Intern Med. 2015;162(1):55–63. https://doi.org/10.7326/M14-0697

12. DeLong E.R., DeLong D.M., Clarke-Pearson D.L. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44(3):837–845. https://doi.org/10.2307/2531595

13. Vickers A.J., Elkin E.B. Decision curve analysis: A novel method for evaluating prediction models. Med Decis Making. 2006;26(6):565–74. https://doi.org/10.1177/0272989X06295361