Automated Classification of Periodontal Bone Loss in Periapical Radiographs Using Convolutional Neural Networks

Objective: Periodontal diseases, including periodontitis, affected approximately 951.3 million people worldwide in 2021. Early detection and accurate diagnosis are crucial for determining effective treatment plans. Radiographic examination, particularly periapical radiographs, is a valuable method for assessing periodontal bone loss (PBL). However, classification of PBL stages is often influenced by observer subjectivity due to the complexity of manual measurement. This study aimed to develop and evaluate convolutional neural network (CNN) models for automated PBL classification in periapical radiographs. Material and Methods: A dataset of 600 periapical radiographs was used, consisting of 480 training images, 60 validation images, and 60 testing images. DenseNet-121, DenseNet-201, VGG-Net 16, and VGG-Net 19 models were implemented for PBL classification. Based on the 2017 World Workshop on the Classification of Periodontal and Peri-Implant Diseases and Conditions, the data were categorized into normal, stage I, stage II, and stage III/IV. The model development included data augmentation, training, and model evaluation using accuracy, sensitivity, and specificity metrics. Results: DenseNet-121 achieved the highest performance among the evaluated models, with an accuracy of 83.33%, sensitivity of 100%, and specificity of 60%. Conclusion: All CNN models demonstrated good performance in classifying PBL in periapical radiographs, with DenseNet-121 showing the best performance, thus indicating strong potential as an automated tool for PBL staging. Increasing the dataset size may further improve performance and reliability for clinical applications.