Detection of microcalcifications in mammograms by the combination of a neural detector and multiscale feature enhancement

Abstract

We propose a two steps method for the automatic classification of microcalcifications in Mammograms. The first step performs the improvement of the visualization of any abnormal lesion through feature enhancement based in multiscale wavelet representations of the mammographic images. In a second step the automatic recognition of microcalcifications is achieved by the application of a Neural Network optimized in the Neyman-Pearson sense. That means that the Neural Network presents a controlled and very low probability of classifying abnormal images as normal. © Springer-Verlag Berlin Heidelberg 2001.