Classifying clusters of microcalcification in digitized mammograms by artificial neural network
Ana Cláudia Patrocínio, Homero Schiabel · 2002
Computer-aided diagnosis (CAD) schemes have presented good results in aiding the early diagnosis of breast cancer. Artificial neural networks (ANN) have been successfully used in CAD classifiers, with success in the classification. The classification of clustered microcalcification has been made from an individual microcalcification analysis. In this work, a classification regarding the characteristics determined only from the cluster itself and discarding the characteristics analysis and extraction from individual microcalcification, was made in two classes: non-suspect and suspect types. Dismissing microcalcification individual features for the network input allows one to eliminate procedures intended to separate each structure from the whole image. The classifier using ANN shows the geometric descriptors efficiency for characterizing microcalcification clusters as well as the influence of features extracted from images known as "age" and "density". The best data shows 92% of correct results, with A/sub z/=0.96.