Approach for the Detection of Early Carious Lesions Based on Intraoral Photographs Using YOLOv7 and Faster R-CNN (#1207)

Authors

Huertas Acevedo, William

Artica Hurtado, Kevin

Castillo Sequera, José Luis

Wong Portillo, Lenis

Abstract

Dental caries represents a global challenge due to inadequate dental care and excessive sugar consumption. These factors lead to costly dental treatments and impact quality of life, self-esteem, and productivity. In response, an approach for the detection of pre-caries lesions through dental image processing and the use of two of the most commonly utilized Deep Learning architectures in the literature, YOLOv7 and Faster R-CNN, is proposed. The approach unfolds in four phases: (i) dataset acquisition, (ii) architecture development, (iii) performance evaluation, and (iv) results analysis. Both architectures leverage a public dataset comprising a total of 9,327  Intraoral Photographs of teeth classified into three categories: “teeth with caries” (0), “teeth without caries” (1), and “teeth with amalgam” (2). A web-based system was developed with the model that demonstrated superior performance. The findings revealed that the YOLOv7 architecture outperformed Faster R-CNN, achieving an average accuracy of 95.7% in detecting “without caries,” “with caries,” and “with amalgam” teeth.