Characterization of clustered microcalcifications in mammograms based on support vector machines with genetic algorithms

Characterization of microcalcification clusters in mammograms is vital in daily clinical practice. At present, characterization of microcalcification clusters mainly depends upon the experience of experts, which increases workload and difficulty. Therefore, a novel automated method based on computer was presented in this paper. Support vector machines (SVMs) have rarely been applied to characterization of microcalcification clusters. This investigation elucidates the feasibility of SVMs to classify clustered microcalcifications. In addition, genetic algorithms (GAs) are applied to select the parameters of an SVM model. We made an experiment with Mini-MIAS database according to the proposed method. Then the approach was evaluated using receiver operating characteristic (ROC) analysis. The experimental results reveal that the proposed method not only achieves a relative high classification performance but also displays good generalization performance.