Comparative Analysis of Different Texture Features in Breast Abnormality Prediction

Abstract

With advancement in thermal imaging technology, thermogram based methods are becoming increasingly popular for non-invasive and early detection of abnormalities. This paper presents a comparative analysis of breast thermogram texture features to classify malignancy in breast tissues. For this, statistical, Gabor, histogram of gradient and local binary pattern texture features is extracted from pre-processed thermograms individually. Using principal component analysis, these feature sets are reduced in dimension and applied to machine learning, support vector machine algorithm for classification of malignancy. In order to analyse the efficiency of feature sets accuracy, sensitivity, specificity and area under curve performance parameters are computed. The results of best feature set are also compared with other state of the art schemes. This work also amends the literature as it compares different types of texture features used for breast thermograms abnormality detection.