Detection of clusters of microcalcifications using a K-nearest neighbour rule with locally optimum distance metrics

A method is proposed for the detection of clusters of microcalcifications. The method first segments the image into suspected regions using morphological filters and a new region growing to derive two boundaries for each region. Then a KNN classifier with two different distance measures, Euclidean distance and locally optimum distance measures, is considered for the task of classifying the regions as normal or MC. The last step of the algorithm uses a hierarchical nearest mean clustering method to find the location of clusters of MCs. The performance of the method on a set of normal and abnormal images is then presented