一种浅层非对称结构的视网膜血管分割网络

针对传统视网膜血管分割网络随着网络深度加深导致微小特征信息丢失，网络分割灵敏度低的问题，提出了一种有别于传统对称编码-解码模块的非对称视网膜血管分割结构。网络权重参数量为7.2MB，以残差注意力模块和多尺度空洞卷积模块作为基础特征提取模块，特征图的最大通道层数只有64层，特征图尺寸减半和反卷积操作都只有两次，能够减少特征图尺寸变化带来的信息丢失现象。本文所提方法在DRIVE和CHASE-DB1数据集上进行测试的准确性分别为96.85%和97.39%，灵敏度分别为84.03%和86.50%，特异性分别为98.08%和98.12%，AUC分别为98.63%和98.99%。

A Retinal Vascular Segmentation Network with Shallow Asymmetric Structure

Aiming at the problem that the traditional retinal blood vessel segmentation network leads to the loss of small feature information and the low sensitivity of network segmentation with the deepening of the network depth, An asymmetric retinal vessel segmentation structure is proposed, which is different from the traditional symmetric encoder-decoder module. The amount of network weight parameters is 7.2 MB, and the residual attention module and the multi-scale dilated convolution module are used as the basic feature extraction modules. The maximum number of channel layers of the feature map is only 64 layers, and the size of the feature map is halved and the deconvolution operation is only twice, which can reduce the information loss phenomenon caused by the change of feature map size. The test accuracy of the proposed method on the DRIVE and CHASE-DB1 datasets is 96.85% and 97.39%, respectively, the sensitivity is 84.03% and 86.50%, the specificity is 98.08% and 98.12%, and the AUC is 98.63%, respectively and 98.99%.