论文部分内容阅读
硬性渗出物是糖尿病视网膜病变(DR)的早期病症,是糖尿病性黄斑水肿的最主要表现,因此对硬性渗出物的准确检测具有重要的临床意义。提出一种基于背景估计和SVM分类器的眼底图像硬性渗出物检测方法。首先通过背景估计,得到包含亮目标的前景图;然后利用基于Kirsch算子的边缘信息确定硬性渗出物的候选区域,再移除视盘;最后对候选区域进行形状特征、直方图统计特征以及相位特征的提取,采用SVM对候选区域进行分类,完成硬性渗出物的精确提取。对DIARETDB1和HEI-MED公共数据库中共248幅眼底图像进行实验,图像水平达到灵敏度97.3%和特异性90%,病灶水平达到灵敏度84.6%和阳性预测值94.4%。实验表明,所提出的方法能够实现眼底图像中硬性渗出物的自动检测。
Hard exudates are early signs of diabetic retinopathy (DR) and are the most prominent manifestation of diabetic macular edema. Therefore, the accurate detection of hard exudates is of important clinical significance. A method based on background estimation and SVM classifier for detection of retinal rigid exudates was proposed. Firstly, the foreground image containing the bright target is obtained through the background estimation. Then, the edge information based on Kirsch operator is used to determine the candidate region of the hard exudate, and then the disk is removed. Finally, the shape, histogram and phase Feature extraction, using SVM to classify the candidate area, to complete the precise extraction of hard exudate. A total of 248 fundus images from a common database of DIARETDB1 and HEI-MED were tested and their images reached 97.3% sensitivity and 90% specificity, with the sensitivity of 84.6% and the positive predictive value of 94.4%. Experiments show that the proposed method can achieve the automatic detection of hard exudates in fundus images.