彩色盘周眼底图的豹纹状眼底自动分级算法

发布时间：2018-06-02 06:19

本文选题：豹纹状眼底 + 脉络膜血管　；参考：《计算机辅助设计与图形学学报》2017年06期

【摘要】：豹纹状眼底常出现在近视眼与老年性退化的眼底中,它是临床上诊断视网膜-脉络膜病变的重要参考.为了辅助近视眼等疾病的临床诊断,提出基于脉络膜血管提取的盘周豹纹状眼底自动分级算法.首先根据所提取的感兴趣区域的直径将眼底图尺寸归一化,以适用于不同分辨率的彩色眼底图;然后基于主成分分析法对视盘中心进行定位,并以视盘中心为中心将眼底图分为上、下、鼻和颞4个象限;再提出描述脉络膜血管透见程度的3类特征:红色通道中的亮度均值、脉络膜血管面积比例以及脉络膜血管密度,并依次对4个象限进行特征提取;最后应用基于信息增益率的C4.5决策树算法将每个象限脉络膜血管透见程度分为0~3级,并将4个象限的分级结果累计,将豹纹状眼底自动分为无、轻度、中度和重度4个级别.用文中算法测试了130幅眼底图,豹纹状眼底自动分级的平均一致率可达84.7%;实验结果表明,该算法能较有效地实现豹纹状眼底的自动分级,为诊断与豹纹状眼底相关的疾病提供量化描述依据.
[Abstract]:Leopard-like fundus is often found in myopia and senile degenerative fundus. It is an important reference for clinical diagnosis of retina-choroidal lesions. In order to assist the clinical diagnosis of myopia and other diseases, an automatic classification algorithm of peri-disc leopard print fundus was proposed based on choroidal blood vessel extraction. First of all, according to the diameter of the extracted region of interest, the size of the fundus image is normalized to apply to the color fundus image with different resolutions, and then the center of the optic disc is located based on principal component analysis. At the center of the optic disc, the fundus image was divided into four quadrants: upper, lower, nasal and temporal quadrants, and then three types of characteristics were proposed to describe the degree of choroidal vascular penetration: the luminance mean in the red channel, the area ratio of choroidal vessels and the density of choroidal vessels. Finally, the C4.5 decision tree algorithm based on the information gain rate is used to divide the degree of choroidal vascular penetration into 0 and 3 levels, and the grading results of the four quadrants are accumulated. The leopard print fundus is automatically divided into 4 grades: none, mild, moderate and severe. 130 eye fundus images were tested with the algorithm, and the average consistency rate of leopard print fundus automatic classification was 84.7.The experimental results show that the algorithm can effectively realize the automatic classification of leopard print fundus. To provide quantitative description for diagnosis of diseases related to leopard print fundus.
【作者单位】：北京理工大学信息与电子学院;北京同仁医院北京市眼科研究所;
【基金】：国家自然科学基金(81271650)
【分类号】：R770.4;TP391.41

【相似文献】