Breast cancer is one of the leading causes of death among women worldwide. Early detection of breast cancer significantly reduces the mortality rate. Computer aided diagnosis (CAD) systems assist the clinicians for early detection, however they are still far from perfection due to morphological diversity of abnormalities in mammograms. In this study morphological component analysis (MCA) has been implemented to remove noises and extract texture information. Two different dictionaries i.e. Local discrete cosine transform (LDCT) and Curvelet transform via wrapping (CURVwrap) are used with 50, 100, 200 and 300 iterations to obtain morphological decomposition into two parts namely piecewise smooth part and the texture part .The piecewise smooth part for each iteration are kept and features were extracted using curvelet coefficients. The Classification performance metrics are measured by 10 fold cross validation rule using Simple logistic, Multilayer perceptron (MLP), Support Vector Machine (SVM) and k-nearest neighbors (KNN) classifiers. Simple logistic classifier produces maximum accuracy of 86.6 % with sensitivity and specificity values of 82.04 and 88.4% respectively.
Curvelet Transform; Morphological Component Analysis; Mammogram Classification; Statistical Features.