A New Feature Extraction Approach Using Contourlet Transform and T-Test Statistics for Mammogram Classification

Abstract

In this study, a CAD system is recommended for the classification of mammographic images. The images are classified as normal-abnormal and benign-malignant. The proposed system consists of three basic steps: the feature extraction, determination of the distinguishing capabilities of the features and selection, and classification. The distinguishing capabilities of the features mean determining the best or optimal features. Thanks to this determination, mammograms could be put into classes with high accuracy. The determination process is carried out using thresholding and t-test statistics. Classification is performed repeatedly for all threshold values using support vector machine. Among the obtained results of the classification, the optimal feature set, which has the best classification performance, is selected. Finally, to evaluate the optimal feature set, classification carries out applying 5-fold cross-validation.