Empirical Mode Decomposition of Digital Mammograms for the Statistical based Characterization of Architectural Distortion

Among the different and common mammographic signs of the early-stage breast cancer, the architectural distortion is the most difficult to be identified. In this paper, we propose a new multiscale statistical texture analysis to characterize the presence of architectural distortion by distinguishing between textural patterns of architectural distortion and normal breast parenchyma. The proposed approach, firstly, applies the bidimensional empirical mode decomposition algorithm to decompose each mammographic region of interest into a set of adaptive and data-driven two-dimensional intrinsic mode functions (IMF) layers that capture details or high-frequency oscillations of the input image. Then, a model-based approach is applied to IMF histograms to acquire the first order statistics. The normalized entropy measure is also computed from each IMF and used as a complementary textural feature for the recognition of architectural distortion patterns. For evaluating the proposed AD characterization approach, we used a mammographic dataset of 187 true positive regions (i.e. depicting architectural distortion) and 887 true negative (normal parenchyma) regions, extracted from the DDSM database. Using the proposed multiscale textural features and the nonlinear support vector machine classifier, the best classification performance, in terms of the area under the receiver operating characteristic curve (or A(z) value), achieved was 0.88.