Classifying Dental Caries in X-Ray Images With Neural Networks Based on Texture Feature Approach

Oral health is essential for overall well-being, affecting physical, social, and economic aspects of life, with dental caries being the most prevalent oral disease worldwide. Early detection and accurate classification of dental caries are crucial to prevent complications. This study focuses on classifying dental caries from X-ray images into four classes based on G.V. Black’s classification. Texture features were extracted using Gray Level Co-occurrence Matrix (GLCM), Gray Level Run-Length Matrix (GLRM), Gray Level Difference Matrix (GLDM), and Law’s Energy Features, resulting in 49 features. The classification was performed using Scaled Conjugate Gradient Backpropagation (SCG) and Gradient Descent with Momentum and Adaptive Learning Rate (GDMALR) algorithms using 5, 10, and 15 hidden node configurations. SCG outperformed GDMALR, with the 10-hidden-node SCG configuration achieving the highest testing accuracy of 97.1%, validation accuracy of 95.7%, and training accuracy of 95.9%. In contrast, GDMALR achieved its best performance with the 15-hidden-node configuration, yielding 73.9% training accuracy, 72.5% validation accuracy, and 71.0% testing accuracy. Although SCG required more epochs, it showed better generalization, making it more effective for classifying dental caries and highlighting the potential of machine learning in enhancing early diagnosis and treatment planning for oral health conditions.