DRPAN: A novel Adversarial Network Approach for Retinal Vessel Segmentation

Retinal Vessel segmentation is the key to automatic detection of retinal diseases in ophthalmoscope images. This paper proposes a method for generating accurate probability maps of retinal blood vessels using generative adversarial network. Unlike the traditional generative adversarial network (GAN) structure, the segmentation model of the GAN in this paper is represented by the encoder-decoder network. The encoder network consists of an dilated residual layer and a decoder network is composed of the pyramid pooling network and the three-layer convolution layer. Due to traditional GAN suffer from training instability, we use Wasserstein GAN-gp (WGAN-gp) in this paper to improve training. Experimental results show that the method proposed in this paper has a satisfactory effect on the segmentation of retinal vessels on the datasets of DRIVE and STARE.