计算机辅助诊断系统中肝脏B超图像的识别研究

发布时间：2018-03-15 03:09

本文选题：计算机辅助诊断　切入点：肝脏B超图像　出处：《青岛大学》2017年硕士论文　论文类型：学位论文

【摘要】：本文致力于计算机辅助诊断系统(Computer Aided Diagnosis,简称CAD)中肝脏B超图像的识别研究,目标是建立一个可协助临床医生识别多类肝脏B超图像的计算机辅助诊断平台,完成对正常肝、肝硬化、脂肪肝三类B超图像的识别。整个计算机辅助诊断系统主要由感兴趣区域(Region of Interest,ROI)的选取、图像预处理、特征提取和选择以及分类器识别这四个模块组成。首先在有经验的医生的指导下划出肝脏区域,在指定范围内提取感兴趣区域,然后对获得的ROI进行预处理,包括图像去噪和图像增强,接下来对ROI进行特征提取和特征选择得到最优特征子集,最后将特征集合输入设计好的分类器进行分类识别。文章主要研究了肝脏B超图像纹理特征提取和选择算法以及分类器的设计问题。由于肝脏病变形态的多样性,并且没有一定的规律可循,对于纹理特征的获得,本文从灰度、空间、频率三个方面进行考虑,涉及到一阶灰度统计、灰度共生矩阵、灰度差直方图、小波包变换。利用所提取的纹理特征就可以通过分类器对三种肝脏B超图像(正常、脂肪肝、肝硬化)进行分类识别。但是其中有些特征会包含重复信息,对分类效果产生负面影响,所以在特征提取之后需要设计一个特征选择模块,本文引入遗传算法从原始特征集合中选出一个可分性好的最优特征子集,以消除特征冗余。最后将最优特征子集作为分类器的输入。关于分类器的设计,本文在传统神经网络的基础上引入Adaboost理论,构建出一种分类效果较传统分类器更好的强分类器,依据最后的分类结果对CAD的性能做出评价,最终完成肝脏B超图像的识别研究。
[Abstract]:This paper is devoted to the research on the recognition of B-ultrasound images of liver in computer Aided diagnosis system (CAD). The aim is to establish a computer-aided diagnosis platform to assist clinicians in identifying various kinds of B-ultrasound images of liver, and to complete the diagnosis of normal liver. Recognition of three types of B-ultrasound images of liver cirrhosis and fatty liver. The whole computer-aided diagnosis system is mainly selected by region of interest region of interest and image preprocessing. The four modules of feature extraction and selection and classifier recognition are composed of four modules. Firstly, the liver region is drawn under the guidance of experienced doctors, and the region of interest is extracted within the specified range, and then the obtained ROI is preprocessed. Image denoising and image enhancement are included, and then the optimal feature subset is obtained by feature extraction and feature selection for ROI. Finally, the feature set is input into the designed classifier for classification and recognition. This paper mainly studies the texture feature extraction and selection algorithm of liver B-ultrasound image and the design of the classifier. And there are no certain rules to follow, for texture features, this paper from gray, space, frequency three aspects to consider, involving the first order gray statistics, gray level co-occurrence matrix, gray difference histogram, Wavelet packet transform. Using the extracted texture features, we can classify three kinds of liver B-ultrasound images (normal, fatty liver, liver cirrhosis) by classifier. But some of the features will contain repeated information. It is necessary to design a feature selection module after feature extraction. In this paper, genetic algorithm is introduced to select an optimal feature subset with good separability from the original feature set. Finally, the optimal feature subset is used as the input of the classifier. On the design of the classifier, Adaboost theory is introduced based on the traditional neural network to construct a strong classifier with better classification effect than the traditional classifier. According to the final classification results, the performance of CAD was evaluated, and the recognition of liver B-ultrasound image was finally completed.
【学位授予单位】：青岛大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R575;TP391.41

【参考文献】