Accurate segmentation of retinal vessels using U-Net and spatial convolution neural network

Early diagnosis of diseases such as diabetes and high blood pressure, which have a direct effect on retinal blood vessels, is very important. In this research, we propose a complex convolutional network with a spatial U-Net, which is used without the need for a large number of training data in a data augmentation manner for optimal use of samples. U-Net is a spatial module that expands the attention map along the spatial dimension and multiplies it to the input feature map for adaptive feature refinement. The input of the complex convolutional network is the structured output blocks from the previous step and does not use the initial spatial U-Net convolution, which avoids additional processing and increases accuracy. To evaluate the proposed method, you use two retina data sets. Two sets of retinal vessel extraction data (DRIVE) and data (CHASE_DB۱) show that the proposed method performs well in both data sets.

کلیدواژه ها:

U-Net ، retinal vessel ، image segmentation ، convolution network ، denoising