Abstract

Image mining is an emerging research field in the digital era. Medical image mining is critical and challenging as medical images contain vital information for characterizing health disorders. Research studies on dental images using data mining classification techniques have revealed infected tooth issues, enabling an accurate automatic interpretation of diseases in clinical imaging. In this study, we present an image mining approach that uses a reduced feature set to classify dental images as normal and caries-infected. Our approach first extracts features from dental images using four feature extraction methods: Local Binary Pattern (LBP), Gray Level Co-occurrence Matrix (GLCM), Gray Level Run Length Matrix (GLRLM) and Local Binary Gray Level Co-occurrence Matrix (LBGLCM). The obtained feature sets are reduced using principal component analysis and then each respective reduced set is subjected to various classification techniques to identify the normal and caries infected dental images. The AdaBoost classifier with the reduced feature set obtained by LBGLCM, GLCM and LBP methods achieved the highest accuracy of 99.7%, 98.7% and 90.8% respectively, whereas the multilayer perceptron layer classifier with the reduced feature set obtained by GLRLM method achieved the best accuracy of 97.9%, demonstrating the efficacy of our reduced textural feature set based approach for dental image classification.