Document Type : Research Article
Authors
1
Department of Engineering and Technology, University of Mazandaran, Iran,
2
​Faculty of Engineering & Technology, University of Mazandaran
3
Department of Oral and Maxillofacial Radiology, Babol University of Medical Sciences, Babol, Iran
10.22080/frai.2025.29272.1013
Abstract
This research presents a hierarchical approach using deep learning methods to categorize panoramic dental images and identify various dental conditions. The images were initially preprocessed, during which the areas of interest were cropped and labeled meticulously. During the initial stage, a pre-trained deep learning model was utilized to differentiate between images with caries and those without caries. Afterwards, in the second stage, the images without caries were classified into healthy and amalgam-filled categories. Following its preparation, the dataset was preprocessed to a great level of accuracy, after which a number of models were tested and compared. Among those under test, DenseNet121 performed better in the initial stage of caries detection, with a total accuracy rate of 83.95%. Also, in the subsequent stage, EfficientNet-B4 performed best in the detection of amalgam fillings and differentiating them from healthy dentition, with an accuracy rate of 98.28%. This research shows that hierarchical methods, when integrated with deep learning models, can yield accurate and trustworthy outcomes in dental image assessment. Moreover, the application of various deep neural network structures is very helpful to enhance classification precision. The results indicate that the suggested method shows significant potential for the automatic detection of dental abnormalities, such as caries and amalgam fillings, thus serving as a complementary tool in clinical diagnostic procedures. In summary, this study presents an efficient and scalable solution for the analysis of complex dental images, which can be integrated into artificial intelligence-based healthcare systems.
Keywords
)