Linear Discriminant Based Mammographic Tumor Classification Using Shape Descriptors
J. Kilday, F. Palmieri, M.D. Fox · 2005
Linear discriminant based techniques were applied to distinguish three distinct tumor types commonly found in mammographic images. Four key features were extracted semi-automatically from each image to provide information on tumor shape. Reduction of the feature space was performed via linear discriminant analysis thus eliminating correlation between features and reducing computational complexity of the classifier. The classification scheme consisted of a simple metric which measured the distance between the discriminant scores in the new feature space and each class mean. One lesion was misclassified out of the 15 analyzed in this preliminary study yielding a misclassification rate of 6.7%.