Support vector machine-based diagnostic classification of caries and healthy teeth in panoramic orthopantomography images

Abstract

Aim: This study proposes a support vector machine (SVM)-based classification model for detecting dental caries in panoramic dental orthopantomography (OPG) images. The primary objective is to develop a computationally efficient, reproducible, and clinically adaptable method that enables the automatic differentiation between carious and healthy tooth structures.

Methodology: The images used in the study were first subjected to standard resizing and normalization steps. The extracted features were then prepared for classification through feature scaling, feature selection, and dimensionality reduction processes. Hyperparameter grid search, stratified k-fold cross-validation, and decision threshold optimization were applied during the model development process. Medically meaningful metrics, including accuracy, precision, sensitivity, F1 score, area under the receiver operating characteristic (ROC) curve (AUC), and the confusion matrix, were used in the performance evaluation.

Results: The SVM model demonstrated balanced and stable performance throughout cross-validation. ROC and Precision–Recall analyses showed that the model has a high capacity to distinguish between decayed and healthy classes. Decision threshold scanning revealed that false-positive and false-negative rates can be optimized according to clinical requirements.

Conclusion: The results show that the proposed SVM-based approach offers a reliable and feasible method for automatic caries detection from panoramic OPG images. This study highlights that traditional machine learning methods, when combined with structural feature engineering, are suitable for supporting dental diagnosis processes. Future studies should test the model on larger datasets and perform multi-classification of different dental pathologies.