Edge Weighted Local Texture Features for the Categorization of Mammographic Masses
In this paper, two novel features—discriminative robust local binary pattern (DRLBP) and discriminative robust local ternary pattern (DRLTP), extensions of local binary and local ternary pattern (LBP and LTP), respectively, are introduced for the characterization of mammographic masses as benign or malignant. Unlike LBP and LTP, DRLBP and DRLTP conserve the edge information along with the texture which delivers more discriminating potential when tested on two benchmark databases, namely, the mini-MIAS and DDSM. To evaluate the efficacy of the proposed features, four distinct classifiers: fisher linear discriminant analysis, artificial neural network, random forest, and support vector machine have been employed with 10-fold cross-validation after selecting the optimal subset of attributes using stepwise logistic regression. The best area under the receiver operating characteristic curve of 0.98 with an accuracy of 92.10% has been achieved with the mini-MIAS database while the same for the DDSM database is 0.97 with accuracy 90.94%.