乳腺X线图像肿块分类方法研究
Research on benign and malignant masses classification in mammogram

摘要 乳腺X线图像肿块的良恶性分类是计算机辅助诊断的研究内容之一，本文对乳腺X线图像肿块边缘分割及不同特征的肿块良恶性分类进行研究.基于最大化分割后图像类间方差的思想，提出了一种改进控制标记分水岭方法完成粗分割，然后采用无边缘活动轮廓(CV)模型对粗分割结果进行修正.为了验证不同特征在肿块良恶性分类中的性能，对现有形状特征、纹理特征在不同分类器下的分类性能进行测试.实验在开源数据库DDSM上验证，结果表明，在通过自动分割方法得到肿块边缘的情况下，纹理特征的分类性能更好.
Abstract：The classification of mammographic masses into malignant or benign is one of the important contents in CAD (Computer-Aided Diagnosis) systems. In this paper, mass contour segmentation and mass classification under different features are studied. Based on the idea of maximizing the between-cluster variance of the segmented images, a modified marker controlled watershed segmentation algorithm is proposed and employed to give the coarse segmentation. Then CV (Active Contour without Edge) model is used to refine the coarse segmentation. The classification performance of existing shape features and texture features under different classifiers is tested for the purpose of validating how different features perform in the malignant-benign classification. The proposed method is evaluated on a public database, DDSM (Digital Database for Screening Mammography). The results show that automatic segmentation can get texture features with better classification performances.