A Retinal Vessel Segmentation Algorithm with Convolutional Neural Network

Abstract

Accurate retinal vessel segmentation technology plays a critical role due to the changes in retinal blood vessels can be used to diagnose certain diseases. In this paper, we propose an application based on a new convolutional neural network for extracting retinal vessel particularly capillaries, which preserves image details as much as possible with different feature information. We evaluated this model on the DRIVE databases. Our results indicate that the network outperforms most competing approaches in term of accuracy, sensitivity, specificity, F1-score, the area under the ROC curve (AUC).