New AI tool provides early warning for dog dental disease

Waltham Petcare Science Institute researchers, in collaboration with Queen Mary University of London, created a new risk assessment model to enhance existing tools to better predict canine periodontal disease. Published in Frontiers in Veterinary Science, the research aims to bring this level of predictive technology to companion animals. ¹

"This breakthrough is more than just a scientific achievement; it's a potential gamechanger for canine oral health. Supported by AI, we are moving from detection to prediction, empowering veterinarians and owners with the ability to provide proactive and personalized oral care. This could fundamentally change the health and wellbeing of our canine companions," stated Stephen Makin, BSc (Hons), PhD, in an organizational release.¹

The research

The researchers constructed a directed acyclic graph (DAG) to identify risk factors associated with periodontal disease in canine patients by combining a literature review with domain expertise from veterinarians who specialize in canine oral health. To create the DAG, researchers utilized an iterative process to identify directional putative causal relationships between variables including signalment factors (age and breed), preventive interventions (professional treatment and home care), diagnostic indicators, and clinical observations.² They also included temporal considerations, such as the timing of observed clinical signs, to ensure variable definitions aligned appropriately with downstream causal pathways.

After the DAG was built, the appropriate data were determined and relevant variables extracted. The researchers utilized observational real-world datasets made up of electronic health records from Banfield Pet Hospitals, to which owners consented to anonymized data retention and use in research at registration, as well as client questionnaires obtained from pet owners registered with NomNomNow, Inc.¹

The study produced a network containing 19 nodes, 45 edges, and 101 total states, requiring 33,231 conditional probabilities.¹ The nodes represented four categories: clinical signs, dental hygiene practice, disease-diagnosis relationships, and pet attributes. Validation against observational data identified two additional relationships: age affecting both dental hygiene and clinical signs. Validation also suggested an increased likelihood of biofilm presence with good dental hygiene, but this was rejected based on clinical implausibility and the sparseness of available dental hygiene data.

The DAG was then parameterized into a fully operational Bayesian network, a type of artificial intelligence model that can analyze complex relationships between multiple risk factors,¹ drawing on electronic health records from over nine million dogs, alongside demographic data, owner-reported surveys, and structured input from veterinary dental specialists. The completed network spans 19 interconnected nodes, each capable of functioning as both an input and an output, allowing the model to generate individualized risk predictions as well as infer likely underlying factors from observed clinical presentations.

The model was tested against multiple independent datasets, demonstrating consistent performance across three of the four validation cohorts. In the cohort where performance was relatively weaker, a prospective study dataset, most misclassifications involved cases with concurrent gingivitis, missing breed information, or known high-risk breed profiles, suggesting the errors reflected case complexity rather than a deficiency in the model's underlying structure.

The findings

Age emerged as a strong and consistent predictor, with periodontitis probability rising from near zero in puppies to approximately 47% by 13 years of age.² Small-breed and brachycephalic dogs showed elevated risk, though the model highlighted important nuances: apparent breed-level rates are substantially influenced by age distribution within breed populations, underscoring the risk of applying population-level statistics to individual patients without accounting for confounding variables.

The presence of gingivitis increased the probability of periodontitis more than 300-fold compared to its absence. Alongside poor dental conformation and visible biofilm accumulation, gingivitis was identified as a potentially valuable early biomarker — one that is readily assessable during routine conscious examination and, with appropriate guidance, even by attentive owners at home. The authors suggest these factors represent not only diagnostic indicators but viable targets for intervention before irreversible periodontal attachment loss occurs.

Reference

1. A first for companion animals: AI-supported risk assessment will provide early warning of periodontal disease in dogs. News release. April 23, 2026. Accessed April 30, 2026. https://prnmedia.prnewswire.com/news-releases/a-first-for-companion-animals-ai-supported-risk-assessment-will-provide-early-warning-of-periodontal-disease-in-dogs-302752058.html
2. O'Flynn C, Wright H, O'Rourke A, et al. Risk assessment for canine periodontal disease using a hybrid causal Bayesian network. Front Vet Sci. 2026;13. doi:10.3389/fvets.2026.1781228