Computer-Aided Classification of Breast Tumors Using the Affinity Propagation Clustering

To improve the accuracy and sensitivity of the breast tumor classification based on ultrasound images, a computer-aided classification algorithm is proposed using the Affinity Propagation (AP) clustering. Five morphologic features and three texture features are extracted from each breast ultrasound image. The AP clustering with an empirical value of "preference" is used as the primary classification to classify tumors into five clusters with each cluster enjoying specific and similar features and holding a certain probability to be malignant. Then, five clusters are further classified into benign and malignant according to their feature distribution. The proposed system is validated by experiments of 132 cases (including 67 benign tumors and 65 malignant ones) with its performance compared with those of popular methods such as the back propagation artificial neural network (ANN), the self-organizing mapping ANN and the support vector machine (SVM). Results show that the proposed system which needs no training procedure performs well in the ultrasonic classification of breast tumors with the highest accuracy of 94.7%.