Retinal Vessel Segmentation Using Densely Connected Convolution Neural Network with Colorful Fundus Images

Retinal vessel segmentation is a key step for diabetic retinopathy diagnosis with colorful fundus images. For the retinal vessel segmentation, a densely connected deep convolution neural network model is constructed and built to improve the performance of retinal segmentation. In this paper, we utilize 17-layer Convolution neural network (CNN) to distinguish the vessel and background in fundus images. Moreover, a dense connection method, that is, the input of layer L is the output of the front L-1 layer in series is also used in the paper. In this way, the back layer of the network can be used as the features of the front layer, and the gradient can be effectively prevented from disappearing. Finally, the binary vessel segmentation result is output with our network. The model is tested with the DRIVE dataset and the accuracy of the experiment exceeded 95%. These experimental results showed that the proposed CNN framework is a state-of-the-art vessel segmentation algorithm.