超声甲状腺结节自动分级相关技术研究

发布时间：2018-04-18 11:13

本文选题：超声图像 + 甲状腺结节　；参考：《电子科技大学》2017年硕士论文

【摘要】：近年来,甲状腺结节发病率逐年上升,引起越来越多的关注。及时地诊断出甲状腺结节的性质对于治疗方案的选择和预后都具有重要的意义。由于具有实时性、廉价性、无损性、可重复性、灵敏度高的优点,超声检查特别是常规超声检查成为甲状腺结节诊断的首选影像学方法。但是由于超声图像往往成像质量较低,而且甲状腺结节的超声表现复杂多样,给甲状腺结节的诊断带来了很大的困难。计算机辅助诊断不但可以提高放射医师的分析诊断效率,同时还可以提高诊断的准确性。因此,超声甲状腺结节计算机辅助诊断的研究日益受到大家的重视。本文主要针对超声甲状腺结节计算机辅助诊断的相关技术进行研究,研究内容主要包括图像预处理、结节分割、特征提取以及良恶性分类和分级四个部分。本文的主要贡献如下:(1)考虑到超声甲状腺结节分割的复杂性,提出一种基于活动轮廓模型的半自动结节分割算法,得到了较好的分割效果。(2)在结节的特征提取方面,不仅采用了若干传统的形状特征和纹理特征,而且根据甲状腺诊断标准中的声像图特征设计了若干有效的统计特征。(3)设计与比较了两个甲状腺结节良恶性分类算法。考虑到特征的维数较高,在算法中引入了特征选择。第一个算法采用基于过滤式特征选择与支持向量机分类器;第二个算法采用了LASSO逻辑回归,利用L1正则化在构建分类模型过程中实现特征选择。(4)基于TI-RADS诊断标准设计了一种基于顺序回归的超声甲状腺结节良恶性分级算法。
[Abstract]:In recent years, the incidence of thyroid nodules has increased year by year, attracting more and more attention.The diagnosis of thyroid nodules in time is important for the choice of treatment and prognosis.Because of its advantages of real time, low cost, nondestructive, repeatability and high sensitivity, ultrasound, especially conventional ultrasound, has become the preferred imaging method for the diagnosis of thyroid nodules.However, the imaging quality of ultrasound images is often low, and the ultrasonic manifestations of thyroid nodules are complex and diverse, which brings great difficulties to the diagnosis of thyroid nodules.Computer aided diagnosis can not only improve the efficiency of radiologist's analysis and diagnosis, but also improve the accuracy of diagnosis.Therefore, more and more attention has been paid to the computer aided diagnosis of thyroid nodules.In this paper, the relevant techniques of computer-aided diagnosis of thyroid nodules with ultrasound are studied, including image preprocessing, nodule segmentation, feature extraction, classification and classification of benign and malignant thyroid nodules.The main contributions of this paper are as follows: (1) considering the complexity of ultrasonic thyroid nodule segmentation, a semi-automatic nodule segmentation algorithm based on active contour model is proposed.Not only some traditional shape features and texture features are adopted, but also some effective statistical features are designed according to the sonographic features of thyroid diagnostic criteria.) two classification algorithms for benign and malignant thyroid nodules are designed and compared.Considering the high dimension of features, feature selection is introduced into the algorithm.The first algorithm uses filter based feature selection and support vector machine classifier, the second uses LASSO logic regression.Using L1 regularization to realize feature selection in the process of constructing classification model. Based on TI-RADS diagnostic criteria, a sequential regression based classification algorithm for benign and malignant thyroid nodules was designed.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R581;R445.1;TP391.41

【相似文献】