CariesXplainer: enhancing dental caries detection using Gradient-weighted Class Activation Mapping and transfer learning

Saira Asghar, Junaid Rashid, Anum Masood

Abstract: This paper presents CariesXplainer, a novel approach for dental caries detection that integrates artificial intelligence (AI) with explainable AI (XAI) using Gradient-weighted Class Activation Mapping (Grad-CAM). To the best of our knowledge, this is the first method to integrate Grad-CAM with standard dental imagery, thereby enhancing both diagnostic transparency and accuracy. The CariesXplainer method comprises three key stages: preprocessing, transfer learning, and integration with XAI. The preprocessing module applies sophisticated image enhancement techniques to the dental image to increase the sharpness and quality of the images. The transfer learning stage utilizes the MobileNetV3 model, pre-trained on diverse datasets, and adapts its feature extraction capabilities to the domain of dental imaging. In the final stage, Grad-CAM is applied to the fine-tuned MobileNetV3 model to visualize the localized regions associated with dental caries. Experimental evaluations are performed on a publicly available Kaggle dataset. We compared CariesXplainer with multiple state-of-the-art models, including CNN-RCNN, ResNet-50, VGG-16, and InceptionV3, and with previous studies. CariesXplainer achieves a superior classification accuracy of 99.50%, significantly outperforming existing methods, thereby demonstrating its usefulness and potential for early caries detection and improved patient outcomes. The generated heatmaps are not only visually interpretable, but also provide local cues to the occurrence of diseases, thereby making diagnosis more accurate and timely. An ablation study is also performed to show the robustness of CariesXplainer.

External IDs:dblp:journals/mta/AsgharRM25