Color Image Segmentation of Dental Caries Using U-Net Enhanced with Residual Blocks and Attention Mechanisms

The rapid progress of artificial intelligence and deep learning has greatly influenced several areas, particularly in healthcare, with computer vision displaying outstanding capabilities in diagnosing and treating patients. Using image analysis in dentistry shows great potential for improving oral healthcare through the identification of dental conditions. The goal of this research is to identify dental calculus and cavities by separating color images into two classes using various U-Net model designs. Three U-Net models are suggested: one with a VGG16 backbone, another with attention gates to highlight important features, and a third combining residual blocks with attention gates. The research made use of a collection of RGB images taken by a dental expert with an intraoral camera within a six-month period. The outcomes show that advanced U-Net structures perform exceptionally well in segmenting dental problems from color images, with the attention residual U-Net achieving a top training accuracy of 84.8%. This study highlights the possibility of creating mobile dental care systems that provide convenient, customized, and effective oral healthcare services.