基于U-Net的眼底视网膜血管图像分割

Segmentation of Retinal Blood Vessel Image Based on U-Net

针对视网膜病变导致图像血管分割出现细小血管断裂和分割精度不高的问题,为提高眼底图像视网膜血管自动分割的准确率,提出一种基于U-Net的眼底图像视网膜血管分割方法 .首先对眼底图像预处理和图像数据扩增,然后构建U-Net模型,用训练好的网络模型对测试样本进行分割,得到视网膜血管分割结果 .利用DRIVE和STARE这2个数据库的眼底图像进行算法性能测试,与多种分割方法相比较体现了更好的分割效果.实验结果的客观评价指标与主观视觉验证该方法取得较好分割性能,能提取更多细小血管,细微血管分叉处连续,为眼科疾病的计算机辅助分析和诊断提供参考.

In order to improve the accuracy of automatic segmentation of retinal vessels in fundus images,a retinal vessel segmentation method based on U-Net is proposed to solve the problem of of small blood vessel breakage and low segmentation accuracy caused by retinopathy in this paper. Firstly,the fundus image was preprocessed and the image data was amplified. Then,the U-Net model was constructed, and the trained network model was used to segment the test samples,and the retinal blood vessel segmentation results were obtained. The algorithm performance was tested on two datasets:DRIVE and STARE. Compared with some segmentation methods,it shows better segmentation effect. The objective evaluation index and subjective vision of the experimental results demonstrated that this method had better performance,it can extract more small blood vessels,and the bifurcation of small blood vessels is continuous,which provided a reference for computer-aided analysis and diagnosis of ophthalmic diseases.