【作者】 柏杏丽；

【导师】 钱旭；

【作者基本信息】 中国矿业大学（北京） ， 计算机应用技术， 2010， 博士

【摘要】 乳腺癌是严重危害女性身心健康的常见恶性肿瘤之一,由于其致病原因还未完全揭晓,目前预防与治疗乳腺癌的关键还在于“早发现,早诊断,早治疗”,钼靶X射线摄影是目前临床乳腺癌检测的主要手段。本文研究乳腺钼靶X射线图像普查和检测的计算机辅助诊断方法,根据乳腺X射线图像的特点,采用独立成分分析技术求出图像的基图像,构造图像的子空间,对图像感兴趣区域在子空间的投影系数进行识别；对提取的图像特征基于粗糙集进行优化选择,两种技术有机结合有效实现了图像的分割和特征提取；基于经典支持向量机原理,结合模糊集理论,采用模糊支持向量机对图像的特征集进行有效分类,对不同的样本采用不同的权系数,提出一种基于样本密度的权系数确定方法,达到消除噪声与野值样本影响的目的；提出一种基于数据融合的双视图乳腺X射线图像检测方法,在同侧乳腺组织的不同视图之间建立参考坐标系,提取区域不变性特征和相似度特征进行检测,提高了检测的灵敏度和特异度；最后采用接受者操作特性曲线对检测效果进行比较评价。更多还原

【Abstract】 Breast cancer is one common malignancy which seriously damages the health of women, as causes not yet fully known, the key of the current prevention and treatment of breast cancer lies in "early discovery, early diagnosis and treatment", mammography X-ray photography is the primary means of clinical breast cancer detection. In this paper, it researches on X-ray used in mammography screening and testing of computer-aided diagnostic methods, according to the characteristics of breast X-ray images, obtained using independent component analysis image-based image, the image of the sub-space structure, the image region of interest in sub-space projection factor identification; on the extraction of image feature set based on rough sets to optimize the combination of two technologies to achieve effective image segmentation and feature extraction; support vector machine based on the classical theory, fuzzy set theory, fuzzy support vector set of image features for effective classification of different samples of different weights, is proposed based on the sample method of determining the right density, to eliminate noise and outliers in the sample affect the purpose; proposed based on data fusion two-view breast X-ray image detection in the ipsilateral breast tissue between the different reference frame the view to extract invariant features and similarity features of the region to detect, to improve the detection sensitivity and specificity; final action by the recipient characteristic curve was evaluated on the test results compared.更多还原