基于改进的模糊C-均值聚类算法及支持向量机的眼底图像中硬性渗出检测方法

Hard exudates detection method in fundus images based on improved fuzzy C-means and support vector machine

目的提出一种基于改进的模糊C-均值(improved fuzzy C-means,IFCM)聚类算法及支持向量机(support vector machine,SVM)的检测算法,以实现对眼底图像中硬性渗出的自动识别。方法首先利用改进的FCM算法对由江苏省中医院眼科提供的120幅彩色眼底图像进行粗分割以获取硬性渗出候选区域;其次,利用Logistic回归对候选区域提取出的特征进行选择,并利用候选区域的优化特征集及相应判定结果建立SVM分类器,实现眼底图像中硬性渗出的自动检测;最后利用该方法对65幅眼底图像进行硬性渗出自动检测。结果硬性渗出自动检测得到的病灶区域水平灵敏度96.47%,阳性预测值90.13%;图像水平灵敏度100%,特异性95.00%,准确率98.46%;平均一幅图像处理时间4.56 s。结论利用改进的FCM算法与识别率较高的SVM分类器相结合的方法能够高效自动地识别出眼底图像中的硬性渗出。

Objective To detect hard exudates automatically in fundus images,an detecting method based on improved fuzzy C-means（ IFCM） and support vector machine（ SVM） is proposed. Methods Firstly,120 color fundus images gotten from Department of Ophthalmology,Jiangsu Province Hospital of TCM were segmented by IFCM, and candidate regions of hard exudates were obtained. Then, the SVM classifer was established with the optimal subset of features which were selected by logistic regression and judgments of these candidate regions. Finally, hard exudates were automatically detected in 65 fundus images. Results Average sensitivity of 96.47% and average positive predict value of 90. 13% were achieved with a lesion-based criterion. The sensitivity,specificity and accuracy were100%,95% and 98.46%,respectively,with an image-based criterion. Average time in processing an image was 4.56 s. Conclusions The method based on IFCM and SVM with higher recognition rate can efficiently detect hard exudates in fundus images.