OralGuard: Harnessing Inception-ResNet-v2 for Cutting-Edge Oral Health Diagnostics

This paper considers the application of the InceptionResNetV2 architecture in the classification of pathologies of the mouth. Six classes are focused on: calculus, caries, ulcers, gingivitis, discoloration of tooth enamel, and hypodontia. The performance of the model is benchmarked on a set containing 1166 labeled images, taking into account the prevalence and severity of each condition. The results indicate that, among all of the scenarios, InceptionResNetV2 has achieved an accuracy of $\mathbf{9 3 \%}$. Accuracy in distinguishing classes: calculus, 0.76; caries, 0.99; ulcers, 1.00; gingivitis, 0.79; discoloration of tooth, 0.99; and hypodontia, 0.99. Its macro average for precision, recall, and F1-score are 0.91, 0.92, and 0.91, respectively. The results showed that the InceptionResNetV2 model could differentiate between classes for a variety of common oral pathologies, thus being able to be very helpful in improving diagnostic accuracy and efficiency in dental practice. Further studies are required to confirm generalizability of these results to larger and more diverse datasets and to real-world clinical settings.