Hybrid XAI and deep learning architecture for trustworthy dental diagnostics

Abstract

Dental periodontal disease is a persistent an inflammatory disorder affecting tooth supporting tissues and stays a main motive of tooth loss. Although dental radiographs are essential for early diagnosis, their interpretation is often subjective and inconsistent due to reliance on clinician expertise. This study proposes an automated and interpretable diagnostic framework using a convolutional neural network (CNN) integrated with gradient-weighted class activation mapping (Grad-CAM). The CNN performs binary classification of periapical radiographs into periodontal and normal categories, while Grad-CAM provides visual explanations of the model’s decision-making process. Experimental results show that the proposed model achieves a classification accuracy of 94.17%, indicating reliable diagnostic performance. The generated heatmaps consistently highlight clinically relevant regions, particularly alveolar bone loss in periodontal cases, whereas normal images exhibit no pathological activation. These findings demonstrate that the proposed CNN–Grad-CAM framework enhances both diagnostic accuracy and interpretability. The study contributes a transparent and trustworthy artificial intelligence solution to support objective periodontal disease diagnosis in dental radiology.