Dental radiology: a convolutional neural network-based approach to detect dental disorders from dental images in a real-time environment

Abstract

Periodontal diseases are among the most prevalent infectious conditions, affecting a major portion of people at times in their lives. As per the World Health Organization (WHO) reports, dental disease diagnosis consumes 5% to 10% of developing economies as healthcare expenditures. Bacteria present in the mouth cause inflammation around the tooth, leading to periodontal disease. In this research paper, three types of dental diseases, i.e., oral cancer, broken teeth, and dental caries, have been classified using the proposed Convolutional Neural Network (CNN)-based model. Moreover, to crossvalidate the results, different pretrained CNN models, i.e. Visual Geometry Group (VGG19), MobileNet-V2, VGG16, Inception-V3, EfficientNet-B3, ResNet-34, Efficient Net-B7, and DenseNet-201 and Support Vector Machine (SVM) have been used for experiments. Experiments have been performed on a locally developed dataset, obtained in the Department of Radiology, Bahawal Victoria Hospital, Bahawalpur, Pakistan. The dataset consists of 1067 dental images of three dental disorders, i.e., oral cancer, broken teeth, and dental caries, labeled by an expert dental surgeon. The proposed CNN model automatically identifies these dental diseases in an automated and noninvasive manner, specifically targeting periodontal conditions. The channel attention module, spatial attention module, squeeze, and excitation block, residual block, convolutional block, and Atrous Spatial Pyramid Pooling (ASPP) module have been applied in the proposed CNN model to extract features. The proposed model, consisting of 24 layers, achieved an impressive classification accuracy of 95.34% for oral cancer, broken teeth, and dental caries.