Computer-Aided Classification of Breast Tumors Using the Affinity Propagation Clustering
To improve the accuracy and sensitivity of the breasttumor classification based on ultrasound images, a computer-aidedclassification algorithm is proposed using the AffinityPropagation (AP) clustering. Five morphologic features and threetexture features are extracted from each breast ultrasoundimage. The AP clustering with an empirical value of preferenceis used as the primary classification to classify tumors into fiveclusters with each cluster enjoying specific and similar featuresand holding a certain probability to be malignant. Then, fiveclusters are further classified into benign and malignantaccording to their feature distribution. The proposed system isvalidated by experiments of 132 cases (including 67 benigntumors and 65 malignant ones) with its performance comparedwith those of popular methods such as the back propagationartificial neural network (ANN), the self-organizing mappingANN and the support vector machine (SVM). Results show that the proposed system which needs no training procedure performswell in the ultrasonic classification of breast tumors with thehighest accuracy of 94.7%.
Yanni Su Yuanyuan Wang
Department of Electronic Engineering Fudan University Shanghai 200433, China
国际会议
成都
英文
1-4
2010-06-18（万方平台首次上网日期，不代表论文的发表时间）