A Computer-Aided Diagnosis System for Dental Diseases Using Deep Learning and Panoramic Radiographs

Diagnosis of multiple oral diseases such as caries and infections, fractures, and impacted teeth depends on dental panoramic radiographs (orthopantomograms, OPGs) but their manual analysis is subjective, time consuming and prone to inter-clinician variation.To overcome these challenges, this study presents a transformer based deep learning approach for automated, precise classification of dental diseases in panoramic X-rays.Two architectures, EfficientNet-B3 and Vision Transformer (ViT-B/16), were trained and evaluated on an augmented dataset covering six diagnostic categories. The Vision Transformer demonstrated superior performance over the CNN-based EfficientNet, achieving an accuracy of 91.76%, precision of 89.90%, recall of 88.40%, specificity of 98.22%, and F1-score of 88.50%.Analysis of the confusion matrix indicated excellent discrimination between classes, particularly for Healthy and Carious Teeth, while ROC curves highlighted the ViT’s balanced sensitivity and specificity across all categories.The results show that transformer models have the ability to extract local and global spatial variations in dental radiographs that assist in identification of sophisticated anatomical patterns with reliability. The framework offers a basis on which AI-assisted dental diagnostics can be performed to offer a higher level of accuracy, less clinician workload, and better evidence-based treatment planning.