Answers are grounded in this paper and cite their sources.
Objective To train, validate, and evaluate a cellular phone–based AI system for image-level classification of visible dental calculus and gingivitis in conscious dogs. Methods This multicenter retrospective study used dental photographs collected from 12 veterinary hospitals (2022 through 2024). Dental photographs were obtained from conscious dogs using cellular phone cameras. A single veterinarian assigned reference labels using standardized criteria. Separate binary datasets for dental calculus and gingivitis were split by dog into training, validation, and independent test datasets. A convolutional neural network was trained using transfer learning, and final models were selected by validation loss. Results A total of 3,170 dogs were included in this study. For dental calculus, 2,771 images were analyzed: 2,451 for training and validation and 320 in the independent test dataset. For gingivitis, 1,684 images were analyzed: 1,434 for training and validation and 250 in the independent test dataset. On the independent test datasets, the dental calculus model achieved precision of 0.977 (95% CI, 0.941 to 0.991), sensitivity of 0.965 (95% CI, 0.926 to 0.984), specificity of 0.973 (95% CI, 0.932 to 0.989), accuracy of 0.969 (95% CI, 0.943 to 0.983), and F 1 score of 0.971. The gingivitis model achieved precision of 0.912 (95% CI, 0.855 to 0.948), sensitivity of 0.950 (95% CI, 0.901 to 0.976), specificity of 0.881 (95% CI, 0.807 to 0.929), accuracy of 0.920 (95% CI, 0.880 to 0.948), and F 1 score of 0.931. Conclusions The AI system demonstrated high classification performance for visible dental calculus and gingivitis in conscious dogs. Clinical Relevance This system may serve as an adjunctive screening and triage tool but is not intended to replace comprehensive oral examination or dental radiography.
No takes yet. Share an insight, caveat, or question.
Kim et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: