基于OCT图像的青光眼病变定量分析研究

发布时间：2018-11-24 19:19

【摘要】：青光眼是一种致盲性特别高的眼部疾病,它能够引起视盘和视杯形态的显著变化。视杯和视盘比值(简称杯盘比)的测量在青光眼的检测中尤为重要。目前较为成熟的青光眼检测方法大都是基于彩色眼底图像的,然而,频谱光学相干层析图像(spectral domain optical coherence tomography, SD-OCT)较眼底图像具有更高的精确度和可靠性,所以,基于SD-OCT图像的杯盘比计算会更准确。且在SD-OCT图像中,杯盘比与视网膜色素上皮层(retinal pigment epithelium, RPE)断点密切相关,因此,本文的研究目的是提出一种基于SD-OCT图像的视网膜色素上皮层断点检测与杯盘比评估方法,来辅助青光眼病变的定量分析。首先,介绍了文章使用的实验数据SD-OCT图像相对于由其他方式获取的图像的优势,及青光眼病变的临床表现,并通过相关文献,介绍了目前青光眼发病部位视神经头区域在眼底图像和SD-OCT图像中的研究现状；接着,阐述了青光眼研究的重要性和本课题的研究意义。然后,给出了本文方法的具体描述,包括断点的初步检测和精确检测,以及每一步的具体处理流程。初步检测时采用了阈值分割和二值形态学、图像拉平的相关知识,但是断点精度有待提高,故又在精确检测时采用了提取样本、确定特征、合成投影图像并限定分类的目标区域、分别利用支持向量机(support vector machine, SVM)和极限学习机(extreme learning machine, ELM)对RPE断点进行识别、利用标签矩阵的梯度信息对异常断点纠正等操作,实现了RPE断点的精确定位,并对断点的实验误差及误差来源进行了相关分析。接着,介绍了一些青光眼病变的诊断标准及可以辅助诊断的相关指标,如盘沿宽度、视网膜神经纤维层(retinal nerve fiber layer, RNFL)厚度、杯盘比等,并对杯盘比在青光眼病变诊断中的意义和本文选取杯盘比作为评估标准的合理性进行了分析,随后利用前文检测出的断点进行了杯盘比计算和误差定量分析。最后,对本文算法所完成的工作做出了总结,简述了本文研究方法的创新点和不足之处,并对下一步需要开展的工作给出了一些展望和建议。
[Abstract]:Glaucoma is a particularly blind eye disease that can cause significant changes in the shape of the optic disc and cup. The measurement of the ratio of cup to disc is particularly important in the detection of glaucoma. At present, the more mature methods of glaucoma detection are mostly based on color fundus images. However, the spectral optical coherence tomography (spectral domain optical coherence tomography, SD-OCT) has higher accuracy and reliability than the fundus images. Cups and disks based on SD-OCT images are more accurate than calculations. In SD-OCT images, the ratio of cup to disc is closely related to the breakpoint of (retinal pigment epithelium, RPE) in the retinal pigment epithelium layer. The purpose of this paper is to propose a method to detect the breakpoint of retinal pigment epithelium and evaluate the cup / disc ratio based on SD-OCT image to assist the quantitative analysis of glaucoma. First of all, the advantages of the experimental data SD-OCT images compared with those obtained by other methods, and the clinical manifestations of glaucoma are introduced. The present research status of optic nerve head area in the fundus and SD-OCT images of glaucoma is introduced. Then, the importance of glaucoma research and the significance of this research are expounded. Then, the detailed description of the method is given, including the preliminary detection and accurate detection of breakpoints, and the specific processing flow of each step. Threshold segmentation, binary morphology and image leveling are used in the initial detection, but the accuracy of breakpoints needs to be improved. The projection image is synthesized and the target area is defined. The support vector machine (support vector machine, SVM) and the extreme learning machine (extreme learning machine, ELM) are used to identify the breakpoints of RPE, and the gradient information of label matrix is used to correct the abnormal breakpoints. The accurate location of RPE breakpoint is realized, and the experimental error and error source of breakpoint are analyzed. Then, the diagnostic criteria of some glaucoma lesions and the relevant indexes that can be used to assist the diagnosis are introduced, such as the width of disc edge, the thickness of retinal nerve fiber layer (retinal nerve fiber layer, RNFL), the ratio of cup to disc, etc. The significance of cup / disc ratio in the diagnosis of glaucoma and the reasonableness of selecting cup / disc ratio as the evaluation criterion are analyzed, and then the cup / disc ratio and error quantitative analysis are carried out by using the breakpoints detected above. Finally, the work of the algorithm is summarized, the innovations and shortcomings of the research methods are briefly described, and some prospects and suggestions are given for the next work to be carried out.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R775;TP391.41

【参考文献】