基于图像的青光眼识别方法研究

发布时间：2018-02-11 08:53

本文关键词： 视盘分割视杯分割区域生长算法山谷差值准则　出处：《沈阳工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：数字眼底图像可以用于诊断诸如青光眼、眼底黄斑性病变等眼底疾病,也可以为诊断糖尿病、高血压等疾病提供参考依据。通过计算机辅助自动检测得到患者眼底图像中的参数,既可以为医生在诊断疾病时提供方便,也可以大大缩短时间达到高效医学检测疾病的效果。利用图像处理技术来增强图像特征、自动识别图像中目标区域并分割提取出眼底目标图像、自动测量形态参数,能够较客观、准确、迅速地为临床诊断提供可靠的依据。视盘和视杯的参数是识别青光眼的重要眼底结构特征。通过计算机辅助自动检测得到杯盘比(CDR)这一眼底图像参数,可以便捷地进行青光眼病筛查。本文提出了一套完整的筛查方法:首先进行了通道提取及图像增强等预处理,然后应用阈值分割算法、边缘检测算法、霍夫变换算法、分水岭分割算法进行处理,分割出了视盘和视杯,并可自动检测出视盘视杯的区域面积,最终得出杯盘比。与传统算法进行比较,阈值分割算法更为简便并且计算准确率相对更高。而后本文改进了传统区域生长算法,用以自动检测并分割出视杯。考虑到传统的区域生长算法在选取种子点时不精确、自适应性差、分割结果不正确的缺陷,本文通过计算感兴趣区(ROI)的几何中心并结合中心亮度作为选取种子点的标准;用5*5模板对眼底图像进行均值滤波,提出山谷差值准则和8邻域连通准则对眼底图像进行种子合并,最终准确分割出视杯。本文应用该方法,对高分辨率眼底图像(HRF)数据库中15张青光眼眼底图像和15张健康眼眼底图像逐张进行检测,实验结果表明,该算法能快速、有效地自动检测出眼底图像中的视杯并将其正确的分割出来,与传统算法相比较该算法稳定可靠,并具有较好的准确性、特异性和敏感性。
[Abstract]:Digital fundus images can be used to diagnose fundus diseases such as glaucoma, fundus macular disease, or diabetes. The parameters of the patient's fundus image can be obtained by computer aided automatic detection, which can provide convenience for the doctor in diagnosing the disease. The image processing technology is used to enhance the image features, identify the target area in the image automatically, segment and extract the fundus target image, and measure the morphological parameters automatically. It can provide reliable basis for clinical diagnosis. The parameters of optic disc and cup are important fundus structure features for identifying glaucoma. The fundus image parameters of cup / disc ratio (CDR) can be obtained by computer aided automatic detection. This paper presents a set of complete screening methods: firstly, we preprocess channel extraction and image enhancement, then apply threshold segmentation algorithm, edge detection algorithm, Hough transform algorithm, etc. The watershed segmentation algorithm is used to segment the disc and cup, and the area of the cup can be detected automatically. Finally, the ratio of cup to disc is obtained, which is compared with the traditional algorithm. The threshold segmentation algorithm is simpler and the calculation accuracy is higher. Then the traditional region growth algorithm is improved to detect and segment the viewing cup automatically. Considering that the traditional region growing algorithm is not accurate in selecting seed points. In this paper, the geometric center of ROI (region of interest) is calculated and the center luminance is used as the criterion of selecting seed points, and the mean value of the fundus image is filtered by the 5 ~ (5) template. The valley difference criterion and 8 neighborhood connectivity criterion are proposed to combine the seeds of the fundus image, and finally to segment the visual cup accurately. This method is applied in this paper. Fifteen glaucoma fundus images and 15 healthy eye fundus images in HRF database were detected. The experimental results show that the algorithm is fast. Compared with the traditional algorithm, the algorithm is stable and reliable, and has better accuracy, specificity and sensitivity.
【学位授予单位】：沈阳工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】