Analysis of mammographic microcalcifications using gray-level images and neural networks

At present, Mammography associated with clinical breast examination and breast self-examination is the only effective and viable method for mass breast screening the presence of microcalcifications is one of the primary signs of breast cancer. It is difficult however, to distinguish between benign and malignent microcalcifications associated with breast cancer. Most of techniques used in computerized analysis of Mammographic microcalcification use the shape features on the segmented regions of microcalcifications extracted from the digitized Mammographs. Since Mammographic images usually suffer from poorly defined microcalcifications features, the extraction of shape features based on a segmentation process may not accurately represent microcalcifications. The intensity variations and texture information in the area of interest provide important diagnostic information about the underlying biological process for the benign or malignent tissue and therefore should be included in the analysis. In this paper, we define a set of image structure features for calcification of malignancy. Two categorizes of correlated gray-level images structure features are defined for classifications of "difficult-to-diagnose" cases. The first based features representing the global texture and the wavelet decomposition-based features representing the local texture of the microcalcification area of interest. The second category of statics of the segmented microcalcification regions and the size, number, and distance features in each category were correlated with cluster. Various features in each category were correlated with the section of the best of features of the segmented calcification cluster. Various features in each category were correlated with the biopsy examination results of a lots of "difficult-to-diagnose" cases gray-level images structure information. The selection of the breast features was performed using the multivariate cluster analysis as well as a genetic algorithm (GA)-based search method the neural network and parametric statistical classifiers receiver operation characteristic (ROC) analysis was performed to compare the neural network based classifications with linear and nearest neighbor (KNN) classifiers. Then neural network classifier yielded between results using once combined set of features selected GA based search method for classification of "difficult-to-diagnose" microcalcifications.