A spatial–semantic dual-graph transformer framework for automated dental disease classification from X-ray images

Abstract

Artificial intelligence (AI) is transforming modern healthcare by enabling faster and more accurate diagnostic workflows, particularly in image-based clinical analysis. In dentistry, early detection of dental diseases using X-ray images remains challenging due to subtle anatomical variations, overlapping structures, and low-contrast lesions. To address these limitations, this study proposes an innovative AI-driven framework for automated dental disease classification using the Spatial–Semantic Dual-Graph Transformer Neural Network (SSDGT-Net). The system starts by collecting dental X-ray images and improving their quality through contrast enhancement and noise reduction. Next, the teeth regions are automatically segmented using YOLOv10 to isolate the relevant dental structures. After segmentation, meaningful features such as texture, shape, and edges are extracted using an Entropy-CNN model. These features are combined using a Cross-Attention Feature Fusion Network (CAFF-Net) along with statistical and geometric descriptors to strengthen the overall representation. The fused feature set is then fed into the proposed SSDGT-Net, which classifies dental diseases into cavity, filling, and implants categories. Experimental results demonstrate that the proposed system achieves superior accuracy, robustness, and interpretability compared to existing approaches, making it a promising tool for reliable, real-time dental disease detection in clinical settings.