Detection of Periodontal Bone Loss (PBL) based on panoramic radiographs is very important in a clinical setting; however, it is still a difficult task due to anatomical overlaps, inconsistent image quality, and the subjectivity of manual interpretation. A Transformer-Based Bilateral Encoder–Decoder (Tr-BLatED) architecture with a specification for automatic tooth recognition and PBL measurement in digital radiographs is introduced in this article. The outlined setup features dual-path feature encoding, where the transformer path obtains global contextual relationships while the CNN-based path gets fine-grained local structures. A specially designed bilateral decoder thus merges the two complementary representations to not only delineate the tooth boundaries but also localize the cemento-enamel junction (CEJ) and alveolar crest (AC) as well as calculate tooth-level bone loss. Different measures, such as Mixup augmentation, weight decay regularization, and stratified k-fold cross-validation, are taken to stabilize the method and are applied to the combined dataset made from three different sources. The method proposed test on the three benchmark datasets and a multi-center test set from the outside. The accuracies of 99.04%, 98.99%, 99.10%, and 94.2% of external validation were attained, respectively. These outcomes indicate the model’s soundness, great diagnostic accuracy, and potentiality for real-world clinical application. The proposed Tr-BLatED framework is a significant step toward the realization of dependable, automated periodontal assessment.
Kotwal et al. (Sat,) studied this question.