一种视网膜血管的分割方法

目的探讨一种自动的视网膜血管分割方法。方法在2个尺度下增强血管图像，由基于直方图的加权模糊C均值聚类算法和形态学细化算子提取血管的近似中心线；通过多尺度形态学重构得到主要血管，再由较小尺度的匹配滤波增强中小血管；合并上述两部分得到最终的血管分割结果。结果对DRIVE数据库测试图像分割的灵敏性为0.7017，特异性为0．9816，准确率为0．9457。结论较同类无监督的血管分割算法具有更高的准确率，但对于小血管的灵敏性不高。

Segmentation of Retinal Vessels

Objective This paper presents an automated method for segmentation of retinal vessels. Methods This algorithm first enhances the retinal vessel on two scales, and then extracts approximate centreline based on Weighting Fuzzy C-Means （WFCM） algorithm and morphological operator. To segment the vascular networks, multi-scale binary morphological reconstruction is employed for main vessels, and the matched filter is utilized to obtain continuous small vessels. The final segmentation is obtained thru combining the above two parts. Results The proposed approach produced an average performance of 0. 701 7 sensitivity,0. 981 6 specificity,and 0. 945 7 accuracy for test imaging on the DRIVE database. Conclusion The evaluation demonstrates our method can improve the segmentation accuracy, as compared with the other widely used unsupervised methods. However, detecting small vessels with the proposed method can not achieve the desired result in sensitivity.