Automated Caries Detection Under Dental Restorations and Braces Using Deep Learning.

In the dentistry field, dental caries is a common issue affecting all age groups. The presence of dental braces and dental restoration makes the detection of caries more challenging. Traditionally, dentists rely on visual examinations to diagnose caries under restoration and dental braces, which can be prone to errors and are time-consuming. This study proposes an innovative deep learning and image processing-based approach for automated caries detection under restoration and dental braces, aiming to reduce the clinical burden on dental practitioners. The contributions of this research are summarized as follows: (1) YOLOv8 was employed to detect individual teeth in bitewing radiographs, and a rotation-aware segmentation method was introduced to handle angular variations in BW. The method achieved a sensitivity of 99.40% and a recall of 98.5%. (2) Using the original unprocessed images, AlexNet achieved an accuracy of 95.83% for detecting caries under restoration and dental braces. By incorporating the image processing techniques developed in this study, the accuracy of Inception-v3 improved to a maximum of 99.17%, representing a 3.34% increase over the baseline. (3) In clinical evaluation scenarios, the proposed AlexNet-based model achieved a specificity of 99.94% for non-caries cases and a precision of 99.99% for detecting caries under restoration and dental braces. All datasets used in this study were obtained with IRB approval (certificate number: 02002030B0). A total of 505 bitewing radiographs were collected from Chang Gung Memorial Hospital in Taoyuan, Taiwan. Patients with a history of the human immunodeficiency virus (HIV) were excluded from the dataset. The proposed system effectively identifies caries under restoration and dental braces, strengthens the dentist-patient relationship, and reduces dentist time during clinical consultations.