Hybrid two-stage CNN for detection and staging of periodontitis on panoramic radiographs.

Background: Periodontal disease is an inflammatory condition causing chronic damage to the tooth-supporting connective tissues, leading to tooth loss in adults. Diagnosing periodontitis requires clinical and radiographic examinations, with panoramic radiographs crucial in identifying and assessing its severity and staging. Convolutional Neural Networks (CNNs), a deep learning method for visual data analysis, and Dense Convolutional Networks (DenseNet), which utilize direct feed-forward connections between layers, enable high-performance computer vision tasks with reduced computational demands. This study aims to evaluate the performance of a hybrid two-stage CNN integrating Mask R-CNN with DenseNet169 for detecting and staging periodontitis in panoramic radiographs. Methods: A total of 600 panoramic radiographs were divided into training (70 %), validation (10 %), and testing (20 %) datasets, with an additional 100 external radiographs used as a final testing set. Four types of annotations were applied: tooth segmentation, radiographic bone loss (RBL), cementoenamel junction (CEJ) area, and periodontitis staging (normal, stage 1, 2, 3, and 4). Mask R-CNN was employed for segmentation training to detect teeth, CEJ, and RBL, while DenseNet169 served as the classifier for periodontitis staging. Results: The hybrid two-stage CNN achieved a periodontitis staging performance on the external testing set with specificity and accuracy of 0.88 and 0.80, respectively. Conclusion: These results demonstrate the potential of this hybrid two-stage CNN model as a diagnostic aid for periodontitis in panoramic radiographs. Further development of this approach could enhance its clinical applicability and accuracy.