摘要
为构建基于眼底图像的糖网自动筛查系统,对基于阈值分割及模式分类器的硬性渗出自动检测方法进行了对比研究。首先,利用基于最小类内离散度的改进Otsu分割算法对眼底图像G通道进行粗分割以实现硬性渗出病灶候选区域的获取;然后,提取并优化候选区域特征集;最后,利用优化后的特征集以及相应人工判定结果分别建立RBF神经网络分类器以及SVM分类器,从而实现对眼底图像中硬性渗出的自动识别。对本方法及其他多种硬性渗出自动检测方法进行对比分析。结果表明,对糖尿病视网膜病变自动筛查的临床应用而言,基于阈值分割及SVM分类器的硬性渗出自动检测方法性能更优。
In order to establish an automatic screening system for diabetic retinopathy based on fundus image,a comparative study of hard exudates detection methods based on threshold segmentation and pattern classifier was carried out.Firstly,an improved Otsu segmentation algorithm based on minimum inner-cluster variance was used to roughly segment G channel of fundus images in order to obtain candidate regions of hard exudate lesions.Then,feature sets of candidate regions were extracted and optimized.Finally,RBF neural network classifier and SVM classifier were established respectively by using the optimized feature set and the corresponding artificial judgment results,so as to realize the automatic recognition of hard exudates in fundus image.This method and other hard exudates automatic detection methods were compared and analyzed.The results show that for the clinical application of automatic screening for diabetic retinopathy,the automatic detection method of hard exudates based on threshold segmentation and SVM classifier has better performance.
作者
高玮玮
左晶
GAO Weiwei;ZUO Jing(College of Mechanical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China;Department of Ophthalmology,Jiangsu Province Hospital of TCM,Nanjing Jiangsu 210029,China)
出处
《中国医疗设备》
2019年第11期74-78,共5页
China Medical Devices
基金
上海高校青年教师培养资助计划（ZZGCD15081）
上海工程技术大学校科研启动项目（E1-0501-15-0185）
