摘要
视网膜图像的血管提取对心脑血管等疾病的诊断、治疗与评价具有重要的临床应用价值。为解决目前视网膜血管分割算法中存在的分割精度低(特别针对病变图像)等问题,提出基于先验知识随机游走模型的视网膜血管分割方法。在分析视网膜血管特征的基础上,构建归一化梯度向量散度场,针对高、低对比度血管采用不同的定向拉普拉斯算子提取血管中心线,并将先验知识随机游走模型应用于图像分割,实现对比度低、边界微弱的视网膜血管提取。采用STARE视网膜图像库进行分割精度测试,结果表明本算法精度相对已有算法明显提高,特别针对带有病变的视网膜图像,算法的有效性得到了验证,可满足临床处理的要求。
Segmentation of retinal blood vessels is significant to diagnosis, treatment and evaluation of the cardiocerebro-vascular diseases. To overcome shortcomings of the existed segmentation algorithms, a novel method based on prior knowledge random walks model was proposed in this paper. Considering the model of the blood vessels, divergence field of normalized gradient vector was constructed, while high-contrast and low-contrast vessel centerlines were extracted using different oriented Laplasian operators. These centerlines were selected as the prior knowledge of random walks model to realize segmentation of blood vessels with low contrast and faintness edges. Experiments were made under STARE public retinal image database. The proposed method achieved better segmentation accuracy than those from existed algorithms. Moreover, the efficiency could satisfy the requirement of clinic diagnosis.
出处
《中国生物医学工程学报》
CAS
CSCD
北大核心
2009年第4期501-507,共7页
Chinese Journal of Biomedical Engineering
基金
新世纪优秀人才支持计划(50051)
国家自然科学基金资助项目(60872081)
关键词
血管分割
视网膜图像
先验知识
随机游走模型
blood vessels segmentation
retinal images
prior knowledge
random walks model
