Abstract:  Breast cancer is the most common cancer among women. With the development of medical treatment, the diagnosis and treatment techniques to breast cancer have improved, but it is difficult to make a timely diagnosis in the early stages of breast cancer, which leads the mortality of breast cancer still high. In view of this phenomenon, this paper studies a classification method of mammography based on nonsubsampled contourlet transform (NSCT). The method firstly decomposes the region of interest (ROI) of mammography based on the multiresolution NSCT, then uses the Zernike moments (Z-Moments) to extract the features of the image which is decomposed by the NSCT. Secondly, for each feature matrix extracted from the ROI, this method decomposes the feature matrix to get the important features which can be extracted to summarize the global characteristics through the singular value decomposition (SVD). This method combines both texture and shape features, and uses the support vector machines (SVM) classification algorithm to define the mammogram images as normal, benign and malignant, achieves the detection and classification of breast lesions. The experimental results show that the accuracy rate reaches 96.76%, and training time greatly reduces. Compared with the state-of-the-art methods, marked improvements are achieved in both accuracy and time efficiency.

Key words: nonsubsampled contourlet transform (NSCT), Zernike moments, breast cancer, mammography, support vector machines (SVM)

摘要： 乳腺癌是女性最为常见的一种癌症。虽然随着医疗的发展，乳腺癌的诊断和治疗技术都有所提高，但是由于不能在乳腺癌早期及时做出诊断，导致乳腺癌的死亡率依然很高。针对此现象，对基于非下采样轮廓变换法（nonsubsampled contourlet transform，NSCT）的乳腺X线图像的分类方法进行了研究。该方法首先对乳腺X线图像的感兴趣区域（region of interest，ROI）进行多分辨的NSCT分解，然后用泽尼克矩（Z-Moments）对NSCT分解后的图像进行特征提取；其次对每一个感兴趣区域所提取的特征形成的矩阵进行奇异值分解（singular value decomposition，SVD），以提取重要的可以概括全局的特征。该方法组合了纹理和形状特征，使用支持向量机（support vector machines，SVM）分类算法将乳腺X线图像分类为正常、良性和恶性，实现了乳腺病变的检测和分类。通过实验可以看出，该方法的准确率达到了96.76%，并且训练时间大大减少，与其他目前最先进的方法相比，在准确率和时间效率上都取得了显著的成效。

关键词: 非下采样轮廓变换（NSCT）, 泽尼克矩, 乳腺癌, 乳腺X线图像, 支持向量机（SVM）