-
摘要频域光学相干断层（Spectral Domain Optical Coherence Tomography, SD-OCT）视网膜图像在眼科疾病的诊断和治疗中有着重要的作用。图像中某些重要特征反映了疾病程度和位置。本文选取糖尿病性视网膜病变（Diabetic Retinopathy, DR）中的渗出（图像中的高信号亮斑）作为研究对象。首先使用一系列层分割算法限制亮斑提取区域，再通过自适应阈值法和区域生长法分割出亮斑，并对误差较大的进行手动修正。通过统计分析四个病变类别40只眼睛的亮斑情况发现，一般其面积和数目越大，则病变越严重。通过方差分析法选出影响病变比较显著的特征，根据这些特征运用libsvm建立初步预测模型，并采用留一交叉验证法评估模型，精度达到70%。27106
- 上一篇：IDEF0+Björk科学文献传播过程与传播效果形成的描述模型研究
- 下一篇：JSP+mysql学生食堂点餐管理系统设计
关键词 频域光学相干断层，糖尿病性视网膜病变，渗出，高信号亮斑，层分割，预测模型
毕业论文设计说明书外文摘要
Title Research of the relationship between features of Optical Coherence Tomography Retinal Images and Diabetic Retinopathy
Abstract
Spectral domain optical coherence tomography retinal images play a vital role in
the diagnosis and treatment of retinal diseases. Some important features in images
reflect the degree and places of diseases. In this paper, we research the exudation
(bright speckles)in diabetic retinopathy. First we limit target regions by layer segmentation methods and then extract bright speckles by self-adaption threshold
and region merging. Meanwhile we correct some segment results by hand. Through statistics and analyses to bright speckles of 40 eyes, we discover that if the image has more speckles, the disease is more severe. We find several remarkable speckle features by the analysis of variance. According to these features, we utilize libsvm to build a prediction model to predict diseases. At last, we use leave-one-out validation method to assess the model and the accuracy acquired is 70%.
Keywords Spectral Domain Optical Coherence Tomography, Diabetic Retinopathy, exudation, bright speckles, layer segmentation, prediction model
目次
1 绪论 1
1.1 光学相干断层视网膜图像相关背景介绍 1
1.2 SD-OCT分层结构介绍 2
1.3 视网膜层分割的研究现状 3
1.4 糖尿病性视网膜病变相关介绍 4
1.5 糖网图像特征与病变关系研究意义及现状 4
1.6 论文的组织结构 5
2 高信号亮斑提取区域的限制 7
2.1 图像的预处理 7
2.2 內界膜及NFL/GCL的分割 8
2.3 IS/OS边界的分割 12
2.4 本章小结 14
3 基于自适应阈值和区域生长的亮斑提取 15
3.1 基于自适应阈值的图像分割 15
3.2 区域生长算法 18
3.3 亮斑分割修正 19
3.4 本章小结 21
4 实验统计分析 22
4.1 实验方法 22
4.2 实验结果分析 23
4.3 本章小结 26
结论 27
致谢 28
表1 DME组眼睛亮斑各特征统计结果 31
表2 NPDR组眼睛亮斑各特征统计结果 32
表3 PDR组眼睛亮斑各特征统计结果 33