基于多路径输入和多尺度特征融合的视网膜血管分割

Retinal vascular segmentation based on multi-scale input and multi-scale feature fusion

视网膜血管的形态变化,如分叉角度、扩张程度等,可为眼底疾病的诊断提供依据。使用深度学习技术对视网膜病变程度进行评估成为目前研究的重点。提出了一种基于多路径输入和多尺度特征融合的视网膜血管分割方法来解决视网膜血管分割问题。采用了多路径输入和多特征融合的方式改进了U-Net模型,使本文的网络能够有效的解决眼底视网膜图像的分割效果差的问题。实验结果表明,算法在DRIVE和CHASED_B1数据集上,敏感性分别取得0.8148和0.8137,特异性分别取得0.9848和0.9866,在分割准确率指标上分别取得0.9699和0.9757,所提方法相较于其他方法较优。

The morphological changes of retinal vessels,such as bifurcation angle and dilation degree,can provide the basis for the diagnosis of fundus diseases.Using deep learning technology to evaluate the degree of retinopathy has become the focus of current research.This paper proposes a retinal vessel segmentation method based on multi-path input and multi-scale feature fusion to solve the problem of retinal vessel segmentation.We use multi-path input and multi-scale feature fusion to improve the U-Net model,so that our network can effectively solve the problem of poor segmentation effect of fundus retinal image.The experimental results show that the sensitivity of the proposed algorithm is 0.8148 and 0.8137 on DRIVE and CHASED_B1 datasets,the specificity is 0.9848 and 0.9866,and the segmentation accuracyis 0.9699 and 0.9757.