视网膜图像解剖结构检测及病变分析研究

发布时间：2018-12-18 03:45

【摘要】：随着人们生活质量的不断提高,糖尿病患者也呈现逐年增多的趋势。糖尿病性视网膜病变作为糖尿病性微血管病变的严重并发症之一,是引起失明以及视力障碍的主要原因。这种病理信息往往可以从视网膜图像中分析得出,然而对于不同的病变阶段,视网膜图像会呈现出不同的病理构造;此外,视网膜图像中的微血管系统也因人而异。这些因素都给本文的研究带来巨大的挑战。本文以视网膜图像作为研究对象,着重对视网膜图像的解剖结构以及病变区域等内容进行分析,主要的研究内容分为以下三个方面:1、视网膜图像预处理。此阶段的主要任务包含视网膜图像背景均衡化,图像噪声去除和感兴趣区域提取三个方面。预处理工作的好坏直接关系到解剖结构检测和病变分析的准确性。2、视网膜图像解剖结构检测。包括血管分割、视神经盘检测以及黄斑区域定位。在视网膜血管分割上,提出一种基于韦伯变换和多尺度空间分析的方法,并通过边界强度和脊强度这两个重要特征对血管像素进行增强;视神经盘作为青光眼神经损害的诊断依据,本文利用其本质的特征并结合Sobel算子、Canny算子以及霍夫变换得到最终的检测结果;对于黄斑区域的工作,文中主要应用模板匹配的方法对其实现精确定位。3、视网膜图像病变区域检测。主要检测的对象为微动脉瘤、出血以及渗出。在微动脉瘤的检测上,运用多尺度高斯核函数进行相似性匹配,接着通过支持向量机对获得的检测点进行分类,该方法的准确率较其他算法相比有着明显的提升。对于视网膜图像中的出血检测,将基于多尺度线性结构的数学形态学方法检测得到的候选区域,通过SLIC算法处理成超像素块并计算其分类特征,利用灰度值投票算法进行分类。最后,针对渗出的检测,先是通过滑动窗口以及形态学重建的方法获得渗出候选区域,然后根据超像素区域计算其对应的特征,并设计人工神经网络分类器进行分类。
[Abstract]:With the continuous improvement of people's quality of life, diabetes patients also show an increasing trend year by year. As one of the serious complications of diabetic microangiopathy, diabetic retinopathy is the main cause of blindness and visual impairment. This kind of pathological information can often be obtained from the retinal image, but for different pathological stages, the retinal image will present different pathological structure. In addition, the microvascular system in the retinal image is different from person to person. These factors bring great challenges to the research of this paper. In this paper, the retina image is taken as the research object, and the anatomical structure and pathological area of the retinal image are analyzed. The main research contents are as follows: 1. Retina image preprocessing. The main tasks of this stage include retinal image background equalization, image noise removal and region of interest extraction. The quality of preprocessing is directly related to the accuracy of anatomical structure detection and pathological analysis. 2. Retinal image anatomical structure detection. These include vascular segmentation, optic disc detection, and macular localization. In retinal vascular segmentation, a method based on Weber transform and multi-scale spatial analysis is proposed, and the pixels are enhanced by boundary strength and ridge strength. Optic disc is the basis for the diagnosis of glaucoma nerve damage. This paper uses its essential characteristics and combines Sobel operator, Canny operator and Hough transform to get the final detection results. For the work of macular region, the template matching method is mainly used to locate the macular area accurately. 3. Microaneurysms, bleeding and exudation were mainly detected. In the detection of microaneurysms, the similarity matching is carried out by using multi-scale Gao Si kernel function, and then the detection points are classified by support vector machine. The accuracy of this method is obviously improved compared with other algorithms. For the detection of hemorrhage in retinal images, the candidate regions detected by mathematical morphology method based on multi-scale linear structure are processed into super-pixel blocks by SLIC algorithm and their classification features are calculated, and the classification is carried out by using gray-scale voting algorithm. Finally, for the detection of exudation, the candidate seepage regions are obtained by sliding window and morphological reconstruction, and then the corresponding features are calculated according to the super-pixel region, and an artificial neural network classifier is designed for classification.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R770.4

【相似文献】