An enhanced tooth segmentation and numbering according to FDI notation in bitewing radiographs

Bitewing radiographic imaging is an excellent diagnostic tool for detecting caries and restorations that are difficult to view in the mouth, particularly at the molar surfaces. Labeling radiological images by an expert is a labor-intensive, time-consuming, and meticulous process. A deep learning-based approach has been applied in this study so that experts can perform dental analyzes successfully, quickly, and efficiently. Computer-aided applications can now detect teeth and number classes in bitewing radiographic images automatically. In the deep learning-based approach of the study, the neural network has a structure that works according to regions. A region-based automatic segmentation system that segments each tooth using masks to help to assist analysis as given to lessen the effort of experts. To acquire precision and recall on a test dataset, Intersection Over Union value is determined by comparing the model's classified and ground-truth boxes. The chosen IOU value was set to 0.9 to allocate bounding boxes to the class scores. Mask R–CNN is a method that serves as instance segmentation and predicts a pixel-to-pixel segmentation mask when applied to each Region of Interest. The tooth numbering module uses the FDI notation, which is widely used by dentists, to classify and number dental items found as a result of segmentation. According to the experimental results were reached 100% precision and 97.49% mAP value. In the tooth numbering, were obtained 94.35% precision and 91.51% as an mAP value. The performance of the Mask R–CNN method used has been proven by comparing it with other state-of-the-art methods. • An enhanced segmentation network on bitewing dental radiographs. • Extraction of dental anchors as a result of two-stage region proposal networks architecture. • Working with 1200 data gathered from real-world data. • Achieving efficient accuracy with hyper-parameter settings.