Abstract: Breast cancer is one of the most common causes of cancer fatality in women. Early detection and treatment are keys to preventing breast cancer from spreading. Digital mammograms are one of the most effective means for detecting possible breast abnormalities at early on stages. Digital mammograms supported with Computer Aided Diagnostic (CAD) systems help the radiologists in taking reliable decisions within short span of time. The proposed CAD system analyses and combines statistical texture features obtained from gray level correlated matrices, for the better classification of mammograms. The Probabilistic Neural Network classifier is used to analyze 256 real mammogram images acquired from different hospitals. The database contains of 136 normal and 130 abnormal i.e., in MLO or CC view. The specific dataset is carefully selected such that the abnormality is apparent in one view and subtle in other due to its complex texture. The proposed system gives 94.76% and 91.31% youdenís ratio, 97.41%, 97.69% accuracies for the two different views respectively. The Youdenís ratio of 94.30 and accuracy 97.14% is attained for normal - abnormal datasets discrimination. The study reveals that features extracted in unified statistical texture domain with PNN classifier proves to be a promising tool for analysis of mammograms irrespective of their appearance and views and Youdenís ratio is the optimum performance measure with justness to sensitivity and specificity.

Keywords: CC, gray level statistics, Mammograms, MLO, PNN, Youdenís ratio.