An end-to-end deep-learning system for segmentation and classification of dental caries from radiovisiography images.

Background: Dental caries remains one of the most widespread oral health issues globally, where timely and accurate diagnosis is critical to prevent tooth loss and preserve oral function. While Radiovisiography (RVG) is widely used for caries detection due to its noninvasive nature, manual interpretation is subject to diagnostic variability and time constraints. This study prsents a comprehensive deep learning (DL)-based pipeline designed to automate the segmentation and classification of dental caries from RVG images with high precision. Materials and Methods: A retrospective dataset comprising 1200 anonymized RVG radiographs was curated and annotated under the guidance of two experienced oral radiologists. The proposed approach integrates a U-Net-based convolutional architecture for precise pixel-wise segmentation of carious lesions, followed by a ResNet-50-based convolutional neural network for multi-class classification into three categories: No caries, enamel caries, and dentin caries. The dataset was divided into training (70%), validation (15%), and testing (15%) sets. Performance metrics included Dice coefficient for segmentation, and accuracy, sensitivity, specificity, and F1-score for classification. Results: The segmentation model attained a Dice score of 0.89, indicating a strong correspondence with expert-annotated lesion boundaries. The classification module attained an accuracy of 93.2%, sensitivity of 91.4%, specificity of 95.1%, and an F1-score of 0.92. Class-wise accuracies were 95% (no caries), 92% (enamel caries), and 91% (dentin caries). In addition, the proposed framework significantly reduced diagnostic time and eliminated inter-observer variability. Conclusion: This study introduces a novel, fully automated artificial intelligence (AI)-based system for the accurate segmentation and classification of dental caries in RVG images. The DL framework demonstrates strong potential as a clinical decision support tool, enhancing diagnostic consistency and efficiency in dental radiology. The findings pave the way for integrating AI into routine dental diagnostics, ultimately leading to better patient care and optimized clinical workflows.