基于CT的肝细胞性肝癌辅助诊断系统中的关键方法研究

发布时间：2017-12-31 14:42

本文关键词：基于CT的肝细胞性肝癌辅助诊断系统中的关键方法研究　出处：《南阳师范学院》2017年硕士论文　论文类型：学位论文

【摘要】：据调查研究统计,近年来肝癌严重威胁着人类的健康,其不但是主要的癌症之一,还是致死率最高的肝脏疾病。在肝癌的类别中肝细胞性肝癌最为严重。其治疗效果差的主要原因是病情确诊和治疗的延迟,在众多的肝细胞性肝癌普查手段中总存在着遗漏和误诊。于此,医学图像计算机辅助诊断技术应运而生,它可以归属于数字图像处理和模式识别的一部分,在实际的疾病普查工作中可以提供量化诊断,为医生提供有参考性的辅助诊断依据,在疾病的影像学诊断中前景广阔,有着极其积极的意义。文章深入研究了国内外学者在肝部疾病辅助诊断上的理论和研究方法,总结和吸收了目前图像处理在医学行业上应用的模式,同时针对肝细胞性肝癌CT图像的计算机辅助诊断方法进行了学习和研究,通过计算机图像处理技术对肝部CT图像进行了特征提取、特征选择和分类比对,对基于CT的肝细胞性肝癌辅助诊断方法进行了改进。主要做的工作有以下四点：1.针对当前肝部纹理特征向量的提取方法进行了研究,不止局限于只提取以往研究提出的“影响大”的如常用的基于灰度共生矩阵的四个特征向量,而是对基于纹理的特征向量进行了比较全面的提取,包括基于灰度直方图的六大特征、基于灰度共生矩阵不常用的八大特征以及基于灰度行程矩阵的五大特征三大方面共提取了27个特征向量用于纹理分析；2.在特征选择方式上,参考以往运用较多的主成分分析,针对主成分分析提取主成分含义不明、不同个案主成分提取不同的缺憾,提出了一种改进后的主成分分析法,在提取的特征上由原始特征的线性组合变为对原始特征的筛选,选出具有实际意义、含义明确的原始特征向量以用于分类器的训练；3.在分类上运用了支持向量机的方法构造分类器,采用全样本、随机抽样和交叉验证的方法分别对分类器进行训练和验证,并运用同一数据源来与改进前的研究方法进行实验对比；4.构建了可视化用户操作界面,演示了肝细胞性肝癌辅助诊断技术日后运用的形式。操作界面包括对CT图片的预处理、可疑图像区域的截取、有效特征求值和CT图像的初步诊断等功能。在大型数据库构建后的未来,操作界面将为医疗工作者提供更加方便快捷的服务。在实验结果上,文章的研究方法在肝细胞性肝癌和正常肝CT图像的分类上准确率为92.9%,高于改进前89.2%的准确率。而研究提出的可视化操作界面也具备一定的可实现性,可以为医疗工作者提供可参考的量化信息,对于医学图像辅助诊断研究有着积极的意义和推动作用。
[Abstract]:According to the survey statistics, in recent years, liver cancer is a serious threat to human health, but not the main one of the highest death rate of cancer, or liver disease. In HCC category in hepatocellular carcinoma is the most serious. The treatment of poor effect is the main reason of delayed diagnosis and treatment, in many liver cells liver cancer census means there exist omissions and misdiagnosis. From this, the computer emerged medical image aided diagnosis technology, a part of it can be attributed to the digital image processing and pattern recognition, in the actual disease census work can provide quantitative diagnosis for doctors, provide a reference basis for auxiliary diagnosis, broad in diagnosis the prospect of disease imaging, has the extremely positive significance. This paper focuses on the theory and research methods in diagnosis of liver diseases on the summary and absorption current Application of image processing in the medical industry mode, at the same time of the study and Research on computer aided diagnosis of hepatocellular carcinoma CT images, through computer image processing technology is used to extract the feature of liver CT image, feature selection and classification on hepatocellular carcinoma diagnosis method based on improved CT. The main works are the following four points: 1. the liver texture feature vector extraction methods were studied, not limited to only extract the previous research proposed "effect" as the four feature vector based on GLCM is used, but the texture feature vector were extracted based on a more comprehensive, including six features based on gray histogram, gray level co-occurrence matrix and eight features not commonly used five features of gray run length matrix three aspects were based on The extraction of 27 feature vectors used for texture analysis; in the 2. feature selection methods, according to the former by using principal component analysis of many principal components based on principal component analysis is ambiguous, different extraction defects of different cases of principal component, puts forward a kind of improved principal component analysis method in feature extraction on linear combination from the original features into the selection of original features, which is of practical significance, the original feature vector meaning clear for classifier training; 3. in the classification by using the method of support vector machine classifier, using the full sample, random sampling method and cross validation were trained and validated on the classifier, and using the same data source and research methods improved before the experiment; 4. construct visual user interface, demonstrates the technology on hepatocellular carcinoma diagnosis after operation Use form. Interface includes pretreatment of the CT images, suspicious image region interception, effective evaluation and characteristics of CT image of the preliminary diagnosis. In the future after the construction of a large database, the operation interface will provide more convenient service for medical workers. In the experimental results, the article research methods in the classification of hepatocellular carcinoma and normal liver CT image on the accuracy rate is 92.9%, higher than the previous 89.2% accuracy improvement. And study the visual operation interface also has proposed can be realized to some extent, can provide quantitative information for reference for medical workers, which is of positive significance to research and promote the medical image aided diagnosis role.

【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】：TP391.41;R735.7;R730.44

【相似文献】