Classification of Microcalcifications

   The microcalcifications detected by the computer are analyzed for their likelihood of malignancy. Morphological features that describe the size, shape, and density of the individual microcalcifications and their variation in a cluster are extracted from the digitized image. Texture features are extracted from the tissue region containing the cluster. A linear discriminant classifier is trained to classify the malignant and benign clusters in the multi-dimensional feature space [10].

Fig. 6. ^ Schematic diagram for computer classification of microcalcifications.

Malignant Cluster: Intraductal Carcinoma Benign Cluster: atypical lobular and     atypical ductal hyperplasia Fig. 7. ^ Examples of microcalcification clusters used in the classification study. In each case, two views were included: craniocaudal view (left), mediolateral view (right). Fig. 8. < Comparison of the average ROC curve for seven MQSA- approved radiologists and the ROC curve of our computer algorithm in classification of 112 clusters.