Wavelet-based Feature Analysis for Classification of Breast Masses from Normal Dense Tissue

Automated detection of masses on mammograms is challenged by the presence of dense breast parenchyma. The aim of this study was to investigate the feasibility of using wavelet-based feature analysis for differentiating masses, of varying sizes, from normal dense tissue on mammograms. The dataset analyzed consists of 166 regions of interest (ROIs) containing spiculated masses (60), circumscribed masses (40) and normal dense tissue (66). A set of ten multiscale features, based on intensity, texture and edge variations, were extracted from the ROIs subimages provided by the overcomplete wavelet transform. Logistic regression analysis was employed to determine the optimal multiscale features for differentiating masses from normal dense tissue. The classification accuracy in differentiating circumscribed masses from normal dense tissue is comparable with the corresponding accuracy in differentiating spiculated masses from normal dense tissue, achieving areas under the ROC curve 0.895 and 0.875, respectively.