基于肝脏超声图像的脂肪肝计算机辅助诊断研究

发布时间：2018-06-09 23:59

本文选题：纹理特征 + GLCM矩阵　；参考：《南阳师范学院》2017年硕士论文

【摘要】：随着大数据时代的到来,被数字化后的医学图像数据量和数据维度越来越丰富,从而形成了庞大的医学数字图像库,如肝脏B超图像。如何使图像观察者从大量的肝脏B超图像中,快速识别出某种疾病的图像(如脂肪肝的图像),从而缩减搜索范围,在保障一定准确率的同时,大大减少观察者繁杂的工作量,是一个非常有意义的课题。首先,本研究收集了大量肝脏超声图像,所收集的超声图像是来自于淅川县第一人民医院的临床影像,并由经验丰富的影像学医生对图像进行了分类,确保了数据源的数量和质量。在收集图像的基础上,本研究还对肝脏超声检查的基本原理进行了解,并识别了典型的超声图像切面。同时,也对脂肪肝判别的医疗标准进行了说明,以此为前提条件,提出了利用计算机进行脂肪肝辅助诊断的研判标准。其次,本研究还介绍了图像的预处理方法,并通过手动选取的方法选取了图像的典型局部区域。本研究中确定了利用纹理特征来识别脂肪肝的思路,对于图像中纹理特征的表示采用了灰度共生矩阵(Gray Level Cooccurrence Matrix, GLCM)的方法,同时,本研究就从灰度共生矩阵的特征出发,对共生矩阵的角度、各种间距等参数进行了分析,从共生矩阵本身进行了纹理特征的相关讨论,并且利用零空间的方法对矩阵进行了降阶处理。再次,本研究从灰度共生矩阵中二次提取了特征值,来深刻的反应纹理的特征,并形成向量库。向量库中的每一条向量表示一个图像的纹理信息,由于灰度共生矩阵的参数的不同从而形成多种不同的特征向量模板,分为正常肝脏向量和脂肪肝向量两种类型。最后,比对测试集中的图片特征向量与特征库向量的欧式距离,最近欧式距离图像所在类即为肝脏图像类别。鉴于欧式距离方法的局限性,本文把数据挖掘分类技术中的KNN方法应用到本文,通过大量实验验证比较,这种有效的筛选方法能够解决问题并实现了86.78%以上的准确率。本文的研究成果对医学大数据背景下,在大量的肝脏B超图像中快速识别脂肪肝图像,辅助医生临床诊断,提高识别率具有十分重大的理论价值和现实意义。
[Abstract]:With the arrival of big data era, the digital medical image data volume and data dimension are more and more abundant, thus forming a huge medical digital image database, such as liver B-ultrasound image. How to make the image observer quickly recognize the image of a certain disease (such as fatty liver image) from a large number of liver B ultrasound images, thus reducing the search range, while ensuring a certain accuracy, while greatly reducing the workload of the observer. Is a very meaningful subject. First of all, a large number of liver ultrasound images were collected from the first people's Hospital of Xichuan County, and the images were classified by experienced imageologists. Ensures the quantity and quality of data sources. On the basis of collecting images, the basic principle of liver ultrasound examination was also understood, and typical ultrasound images were identified. At the same time, the medical standard of fatty liver discrimination is explained, and the computer aided diagnosis standard of fatty liver is put forward. Secondly, the preprocessing method of image is introduced, and the typical local area of image is selected by manual selection. In this study, the idea of using texture features to identify fatty liver is determined. The gray level occurrence matrix (GLCMCMx) method is used to represent texture features in images. At the same time, this study starts from the characteristics of gray co-occurrence matrix. Based on the analysis of the angle of the symbiotic matrix and the distance between the parameters, the texture features of the symbiotic matrix are discussed, and the order reduction of the matrix is carried out by using the method of zero space. Thirdly, the eigenvalues are extracted from the gray level co-occurrence matrix, which can reflect the texture features deeply and form a vector library. Each vector in the vector library represents the texture information of an image. Because of the different parameters of the gray level co-occurrence matrix, there are many different feature vector templates, which can be divided into two types: normal liver vector and fatty liver vector. Finally, compared the Euclidean distance between the feature vector and the feature vector in the test set, the nearest Euclidean distance image is the liver image category. In view of the limitation of the Euclidean distance method, this paper applies the KNN method in the data mining classification technology to this paper. Through a large number of experiments and comparison, the effective screening method can solve the problem and achieve the accuracy of more than 86.78%. The research results of this paper are of great theoretical value and practical significance for rapid recognition of fatty liver images in a large number of B-mode ultrasound images of liver, assisting doctors in clinical diagnosis and improving the recognition rate under the background of medical big data.
【学位授予单位】：南阳师范学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R575.5;TP391.41

【相似文献】