Wavelet scalar-energy features for recognition of mammographic mass shapes

Abstract

The goal of this study is to analyze the use of wavelet- based shape features for automated recognition of mammographic mass shapes. Two sets of shape features are used. The first set includes wavelet-based scalar-energy features. The mass boundary radial distance memory is decomposed using a discrete wavelet transform. The energy of the coefficients at each scale are computed, and these energy values are then used to form a feature vector. Several mother wavelets are used for the wavelet-based shape features: Daubechies-3 (DB3), DB5, DB7, DB9, DB11, DB13, DB15, Coiflets-3 (C3), C5, Symlets-2 (S2), S4, S6, and S8. The second set includes the following traditional features: radial distance mean, standard deviation, zero-crossing count, roughness index, entropy, and compactness. For each set of shape features, linear discriminant analysis is used to appropriately weight the features, and a minimum Euclidean distance classifier is used to separate the shapes into three classes: round, lobular, and irregular. The classification results, as well as false positive and false negative rates, are compared for each set of shape features.