Explainable Deep Learning Framework for Periodontal Diagnosis in Dental Imaging

Abstract Periodontal disease is a chronic inflammatory condition impacting tooth-supporting structures. Early detection via radiographic analysis is essential to prevent progression and tooth loss. However, radiograph interpretation remains subjective and reliant on clinical expertise. This study presents an automated diagnostic framework using a Convolutional Neural Network (CNN) combined with Explainable Artificial Intelligence (XAI) through Gradient-weighted Class Activation Mapping (Grad-CAM). The system classifies periapical dental radiographs into periodontal and normal categories. The proposed CNN includes four convolutional layers with ReLU activation, max-pooling, and two fully connected layers. Data augmentation improved generalizability, and training employed the Stochastic Gradient Descent with Momentum (SGDM) algorithm. The model achieved a 94.17% classification accuracy with stable convergence. Grad-CAM heatmaps highlighted relevant regions influencing the model's decisions. In periodontal cases, alveolar bone loss was consistently identified, while normal images showed no pathological patterns. This visualization enhances interpretability and clinical trust. The integration of CNN and Grad-CAM offers a reliable, transparent diagnostic tool for dental radiology and supports the ethical adoption of AI in healthcare.