Applying feature selection for effective classification of microcalcification clusters in mammograms

Classification of benign and malignant microcalcification clusters (MCC) in mammograms plays an essential role for early detection of breast cancer in computer aided diagnosis (CAD) systems, where feature selection is desirable to improve both the efficiency and robustness of the classifiers. In this paper, three approaches are applied for this task, including feature selection using a neural classifier, a clustering criterion and a combined scheme. To evaluate the performance of these feature selection approaches, a same neural classifier is then applied using the selected features and the classification results are then compared. In our dataset in total 748 MCC samples are detained from the
well-known DDSM database, where 39 features are extracted for each sample. Comprehensive experiments with quantitative evaluations have demonstrated that the best classification rate can be achieved using 15-20 selected features. Also it is found that applying features selected from clustering rules can yield better performance in separate and combined scheme.