IDENTIFICATION OF DENTAL CARIES USING A STRUCTURED DEEP LEARNING APPROACH

As is well known, early treatment for dental caries can prevent many subsequent issues. This study applies advanced computational intelligence techniques to address the critical task of dental caries detection and localization. In the initial step, the YOLOv7 model is used to detect and segment the full-tooth dental periapical images into single-tooth dental periapical images. Subsequently, the crown region in each single-tooth dental periapical image is segmented to prepare the data for further analysis. This segmentation enables the InceptionV3 deep learning model to efficiently conduct preliminary caries classification, generating dental caries periapical images that facilitate caries localization through YOLOv7. Additionally, to improve the accuracy of dental caries classification and localization, this study trains the deep learning models for upper and lower teeth separately. Hence, our classification models can achieve F1-score of 92.5% and 91.9% for upper teeth and lower teeth, respectively, while our localization model can achieve an average F1-score of 81.1% with a precision of 80.5% and a recall of 82.3%. For the localization of early symptoms of caries, this study achieves an average precision of 80.8%, recall of 94.1%, and F1-score of 87%. When compared to other studies, our proposed approach demonstrates better adaptability and accuracy in locating caries for more complex images. This integration of AI techniques within computational intelligence highlights its transformative potential for improving clinical outcomes in dental biomechanics and diagnostics.