Design and Development of an Artificial Intelligence (AI)-Based Early Screening Tool for Plaque and Caries Identification in Self-Serving Intraoral Photos "DentAware+"

Abstract

Dental caries and plaque remain among the most prevalent yet preventable oral health problems worldwide, and early self-screening could reduce the delay between disease onset and professional care, particularly where dentist-to-population ratios are limited. This study designs, develops, and evaluates DentAware+, an artificial intelligence (AI)-based mobile screening tool that identifies plaque and caries directly from self-taken intraoral photographs without clinician assistance. The system was developed using a design-science approach comprising a guided self-capture interface, an image-quality pre-check module, and two convolutional neural network (CNN/YOLO-based) inference models. A prospective diagnostic-accuracy evaluation involved 150 participants who each captured six standardized self-taken photographs (900 images total; 630 training, 135 validation, 135 test), independently annotated by three calibrated dentists using ICDAS-II (caries) and the Silness-Löe index (plaque) as reference standards. On the held-out test set, the caries model achieved 91.4% accuracy, 88.6% sensitivity, 91.5% specificity, and an AUC of 0.93, while the plaque model achieved 89.3% accuracy, 85.0% sensitivity, 89.1% specificity, and an AUC of 0.91; agreement with dentist consensus was substantial (Cohen's κ = 0.79 for caries, κ = 0.74 for plaque). The image-quality module prompted retakes for 14.6% of capture attempts. System Usability Scale scores averaged 78.4, indicating good usability. These findings suggest DentAware+ can provide clinically meaningful, accessible early screening from ordinary self-taken smartphone photographs, offering a low-cost triage tool to prompt timely professional consultation.