Classification of Microcalcification Clusters Using Topological Structure Features

The presence of microcalcification clusters is a primary sign of breast cancer. It is difficult and time consuming for radiologists to diagnose microcalcifications. In this paper, we present a novel method for the classification of malignant and benign microcalcification clusters in mammograms. We analyse the topology of individual microcalcifications within a cluster using multiscale morphology. A microcalcification graph is constructed to represent the topological structure of the cluster and two properties associated with the connectivity are investigated. A multiscale topological feature vector is generated from a set of microcalcification graphs for classification. The validity of the proposed method is evaluated based on the MIAS database. Using a k-nearest neighbour classifier, a classification accuracy of 95% is achieved for both manual annotations and automatic detection results. The obtained area under the ROC curve is 0.93 and 0.92 for the manual and automatic segmentation, respectively.