ROI-Guided CNN-Based Classification of Dental Caries in Panoramic X-Ray Images

Dental caries is one of the most common oral health problems, and early detection is essential to prevent progression and tooth loss. Conventional diagnostic approaches, such as manual interpretation of radiographs, are prone to human error and strongly depend on the examiner's experience. In recent years, deep learning methods have shown promising results in improving diagnostic accuracy and reducing variability. This study proposes a region of interest (ROI)-guided convolutional neural network (CNN) model for the automatic detection of dental caries in panoramic radiographs. The methodology involved dataset preparation, patch extraction and labeling of both healthy and carious regions, and application of ROI filtering to remove irrelevant features. To address class imbalance, healthy samples were systematically extracted by excluding overlapping areas with carious annotations. A lightweight CNN architecture was trained on these balanced datasets, incorporating convolutional layers, pooling operations, and dropout regularization to improve generalization. The proposed model achieved a classification accuracy of ۹۹% in distinguishing carious from healthy regions. Furthermore, evaluation on full panoramic radiographs demonstrated strong generalization, with the model successfully localizing carious lesions in realistic clinical scenarios. These findings suggest that the ROI-guided CNN framework provides a robust and efficient solution for caries detection.

کلیدواژه ها:

Dental caries ، Deep learning ، Convolutional neural network ، Panoramic radiographs ، Computer-aided diagnosis