Evaluation of deep learning systems in detection of dental caries in panoramic radiography

The aim of the study is to compare traditional diagnostic methods and deep learning methods, which is a subbranch of artificial intelligence, in the detection of dental caries and to evaluate the effectiveness of caries diagnosis with artificial intelligence method. In the study, panoramic radiographs of 101 patients were selected as "caries/no caries" in a joint council formed by two physicians specialized in Oral and Maxillofacial Radiology and a research assistant continuing their education in Oral and Maxillofacial Radiology, and the teeth were labeled as healthy and caries. The 5000 data obtained were divided into 80% training and 20% test data. ResNet50 neural network architecture was applied to build the model. Our deep learning model showed an accuracy of 0.82. Among the performance metrics of the model, the PPV value was 75.8%, the NPV value was 92%, the sensitivity was 94%, and the specificity was 70%. The AUC value was found to be 82%. Our convolutional neural network model performed efficiently in detecting caries lesions on panoramic radiographs.