乳腺癌X线图像的钙化点计算机辅助检测

Computer Aided Detection of Calcifications in Mammograms

作者: 专业:生物医学工程 导师:刘惠 年度:2010 学位:硕士  院校: 大连理工大学

Keywords

Mammograms, Calcification, BP neural network, image segmentation

        乳腺癌是女性常见的恶性肿瘤之一,严重威胁着女性健康,及早发现和治疗是降低乳腺癌死亡率的关键。微小、颗粒状的钙化点是早期乳腺癌的一个重要特征,据统计30%-50%的乳腺恶性肿瘤伴有钙化点。乳腺X线摄影简单经济,对早期乳腺癌的病理表现有明显表征,因此借助乳腺X线影像检测钙化点是乳腺癌早期诊断的重要技术。但是钙化点小且形状不规则,和一些致密组织及血管等的阴影十分相似,大部分钙化点在人眼识别时不易被察觉,使得钙化点的诊断成为医生的棘手问题。随着现代医学成像技术的迅猛发展,借助计算机及人工智能技术的钙化点计算机辅助检测成为可能,而且为医生提供了一个有价值的参考意见,减少医生在大范围内查找病变区域的工作量;另一方面使X线的影像学诊断更为客观化,减少因为医生经验不足造成的误诊和漏诊。因此,利用计算机辅助手段提高钙化点检测的准确率是目前众多学者研究的热点。本文以钙化点检测展开,对早期乳腺癌中钙化点的计算机辅助检测技术进行探索,依照乳腺X线图像的数据准备、乳腺区域提取、特征提取及分类、钙化点区域提取和定位的次序,针对乳腺X线图像进行钙化点的检测研究。主要研究内容和成果有以下几个方面:(1)实现了乳腺区域的自动提取,降低了背景对钙化点检测可能造成的干扰,并且能够有效的降低运算量,为后续的钙化点检测做好准备。(2)由于钙化点微小,与背景对比度差,本文采用了对像素进行特征提取的方法,此外根据医生提供的标准区域的特点,提出进行钙化点的分割以选取正样本像素;对样本进行特征提取及选择。(3)设计和训练神经网络分类器,基于选择出的特征子集构建神经网络,在训练网络时,采用了样本分组模式,通过校验样本优化分类性能,最终选出最优分类器。(4)用设计出的分类器对乳腺X线图像进行测试,通过对伪钙化点的形状分析后,较好的去处了伪钙化点,提高了检测的准确率,并实现了计算机辅助检测的目的。在上述思路下,本文以DDSM数据库和MIAS数据为基础,进行分析处理后,依乳腺区域提取、特征提取、神经网络设计及图像测试的步骤,进行最后试验,验证了算法的可行性和有效性,达到了计算机辅助检测的目的。
    Breast cancer is one of the most common malignant tumors in women, which is a serious threat to women’s health. But the etiology of breast cancer is not explicit now. Recently there is a trend of increasing incidence. So early diagnosis and treatment of breast cancer is a key ingredient of any strategy designed to reduce breast cancer mortality. Small, granular calcification is an important feature of early breast cancer. According to related statistics, 30%-50% of breast cancer is associated with calcification. Mammography is the first choice for early detection of breast cancer for it’s simple and economic. So early detection of calcification in mammography is a key technology for early diagnosis of breast cancer. But calcifications are usually very small and their shapes are irregular. They are also very similar to the shadow of some dense tissues or blood vessels, so only a small part of the calcifications can be identified through human eyes. In most cases, calcifications cannot be easily detected, making it a tough problem for physicians. Thanks to the rapid development of modern medical imaging technology, it’s possible to accomplish computer-aided detection of calcification with the use of computer and artificial intelligence technology. On the one hand it provides valuable advice and reference and helps reduce the work of finding lesion area for doctors; On the other hand, it makes mammograms diagnosis become more objective, and helps to reduce the resulting misdiagnosis and missed diagnosis when the doctors are lack of experience. Now, improving the accuracy of detection of calcification through image processing techniques is a research focus.This paper is an investigation for calcification detection in early breast cancer. We’ll investigate computer-aided detection of calcification in mammograms. We finished the preparation for mammogram data, extracted the breast region. Then we extracted the calcification regions and labeled them, and show the effectiveness of the algorithm by experimental results. In this paper, the research content and results are the following parts:(1) Achieve extraction of breast region automatically, and reduce the interference which may be caused by background in calcification detection, and can reduce the amount of computation effectively, make a nice prepare for the following calcification detection.(2) For calcifications are always small, and have a low contrast with background, so use the pixel-level feature extraction.Make calcifications segmentation in standard area provided by doctors, and select samples, and do calcifications feature extraction and selection in two kinds of samples, and use forward sequence to find out the most optimal feature subset. (3) Design and train neural network classifier, construct neural network based on the selected feature subset, obtain the optimal classifier, and analyze the performance of the classifier.(4) Use the optimal classifier to test breast X-ray images, after analyzing the shape of pseudo-calcifications, we can remove them better, and improve the accuracy of detection, and achieve the purpose of computer-aided detection.Based on the above ideas, this paper analyze and process the data based on DDSM and MIAS database, follow the steps of breast region extraction, feature extraction, neural network design and image test to conduct the final experiment, verify the feasibility and effectiveness of the algorithm, and achieve the purpose of computer-aided detection.
        

乳腺癌X线图像的钙化点计算机辅助检测

摘要4-5
Abstract5-6
1 绪论9-16
    1.1 研究的背景及意义9
    1.2 乳腺癌影像学诊断方法9-11
    1.3 计算机辅助诊断的研究状况11-12
    1.4 钙化点检测算法研究12-13
    1.5 论文的内容及框架安排13-16
2 乳腺区域提取16-26
    2.1 乳腺区域分割的目的和难点16-18
    2.2 阈值分割和形态学处理18-20
    2.3 乳腺区域提取算法20-25
        2.3.1 乳腺区域提取方法20-21
        2.3.2 乳腺区域提取实现21-25
    2.5 本章小结25-26
3 钙化点的特征提取和选择26-34
    3.1 样本选择26-27
    3.2 特征提取27-31
    3.3 特征选择31-33
    3.4 本章小结33-34
4 基于神经网络的特征分类34-42
    4.1 分类器介绍34-35
    4.2 BP神经网络分类器35-38
    4.3 BP神经网络分类器设计及评价38-41
        4.3.1 BP神经网络的设计38-39
        4.3.2 性能评估39-41
    4.4 本章小结41-42
5 钙化点检测实验及结果分析42-53
    5.1 乳腺X线图像获取42-46
        5.1.1 数据库的介绍43-44
        5.1.2 图像格式转换44-46
        5.1.3 图像采样和量化46
    5.2 钙化点提取46-52
        5.2.1 钙化点检测46-47
        5.2.2 钙化点识别47-52
    5.3 本章小结52-53
结论53-54
参考文献54-57
攻读硕士学位期间发表学术论文情况57-58
致谢58-60
        下载全文需10


本文地址:

上一篇:陶瓷微细结构的生坯加工方法研究
下一篇:H.264编码标准中帧内预测及全零块预判算法研究

分享到: 分享乳腺癌X线图像的钙化点计算机辅助检测到腾讯微博           收藏
评论排行
公告