An Automated Self-Attention Enhanced Deep Learning Model for Dental Caries Detection

To improve the efficacy of detecting dental caries in panoramic X-ray images using deep learning (DL) models, this study assesses how effective attention processes are. To strengthen the model's resilience panoramic radiographs were gathered, the dataset consists of 1160 images after data augmentation where a 7:2:1 ratio of images was taken for training, validation, and testing. NasNetMobile and VGG16 were utilized as foundational models, incorporating attention techniques to concentrate on crucial areas in the images. The VGG16 model with attention consistently performs better than the NasNetMobile model with an accuracy of 98.2%, according to experimental data, proving its improved capacity to classify caries in panoramic X-ray photographs. These results highlight the importance of attention processes in DL model optimization for medical image analysis tasks, particularly in the context of dental caries detection.