Contextual and Non-Contextual Features Extraction and a Selection Method for Microcalcifications Detection

In this paper a method to extract and build patterns to model microcalcifications from digitized mammography is presented. The proposed method consist in a combination of two steps, in the first one, a feature extraction method is applied using multiscale wavelet image processing, and is combined with a Self Organizing Neural Network to solve the segmentation image problem. In the second one, a feature selection method is considered applying a Generalized Regression Neural Networks (GRNN) in order to obtain an optimal and a short input vector for classifier design proposes. Optimal results were obtained and compared with others.