DentXAI: An Explainable AI System for Real-Time Dental Lesion Detection and Diagnosis

Abstract

Dental caries and other oral pathologies remain among the most prevalent chronic diseases globally, affecting billions of individuals and imposing substantial healthcare burdens, particularly in resource-constrained settings. Traditional diagnostic approaches, while effective, are often time- intensive, subject to inter-observer variability, and reliant on expert availability. This study presents DentXAI, a comprehensive explainable artificial intelligence framework designed to detect and classify dental diseases from both intraoral photographs and panoramic radiographs. Our system employs YOLOv8 architecture combined with Layer-wise Relevance Propagation (LRP) and Occlusion detection methods to provide transparent, interpretable diagnostic insights. We developed three distinct models: a binary cavity detection model achieving 90% accuracy on intraoral images, a three-class severity classification model (severe, moderate, low) demonstrating 65% accuracy with a mean Average Precision of 0.41 at IoU 0.5, and a comprehensive 31-class multi-pathology detection model for panoramic radiographs trained on approximately 10,000 images. The integration of explain ability techniques addresses the critical "black box" limitation of deep learning models in clinical applications, fostering trust and enabling clinicians to understand the model's decision-making process. Initial results demonstrate promising performance across multiple dental conditions including caries, root canal treatments, periapical lesions, crowns, fillings, and missing teeth. However, the study reveals significant challenges in multi-class classification accuracy, particularly for underrepresented classes, highlighting the necessity for larger, more balanced datasets. Future work involves expanding our dataset through collaboration with The Oxford Dental College Bangalore and refining the model architecture to improve diagnostic precision. This research contributes to the growing body of evidence supporting AI-assisted dental diagnostics while emphasizing the importance of transparency and clinical interpretability in medical AI systems.