邻域约束模型的眼底图像硬性渗出聚类检测方法

Clustering Detection Method of Hard Exudates in Fundus Image Based on Neighborhood Constraint Model

硬性渗出是糖尿病视网膜病变的重要表现和诊断依据.针对硬性渗出检测容易受到图像背景和噪声干扰的问题,提出基于邻域约束模型的眼底图像硬性渗出聚类检测方法.首先设定检测区域,结合区域像素的灰度和空间信息定义目标检测函数,通过迭代计算完成图像的聚类分割;然后计算邻域的灰度差异,将最大灰度变化作为相似性判决的约束条件,进而判定每个聚类图像是否属于硬性渗出.在公开的眼底图像数据库上进行实验的结果表明,该方法能有效地识别和检测眼底图像中可能存在的硬性渗出,对正常图像的判断正确率达到90%,对存在病变图像的检测灵敏度和阳性预测值分别达到79%和81%,有助于眼底疾病的计算机辅助诊断.

Hard exudates are important manifestations and diagnostic bases for diabetic retinopathy.To solve the problem of easy disturbance of hard exudates detection by image background and noise,a hard exudates clustering detection method based on neighborhood constraint model is proposed.Firstly,the detection area is set.The target detection function,which is defined by the gray and spatial information of pixels,is used to complete the clustering segmentation of the image by iterative calculation.Then the gray differences of the neighborhood are calculated,and the greatest gray change is used as the constraint condition of the similarity decision to determine whether each cluster image is hard exudates.The performance of the method is verified on the open eye image databases.The results show that the method can effectively identify and detect the possible hard exudation in the fundus image.The accuracy of the normal image reaches 90%,and the sensitivity and positive predictive value for hard exudates achieve 79%and 81%,respectively.The method is thus proved conducive to the computer-aided diagnosis of the fundus diseases.