Fundus Image Segmentation Based on Improved Generative Adversarial Network for Retinal Vessel Analysis

Abstract

The analysis of retinal vessel imaging can diagnose and monitor various diseases such as diabetes, hypertension, and so on. However, testing of retinal vessels requires professional clinicians to examine and evaluate digital colorful fundus images. This process consumes a lot of time and human labor, so automated fundus image segmentation models are necessary to provide decision-making. We propose an end-toend retinal vessel segmentation model based on a Generative Adversarial Network. First of all, a self-made amplifier was added to magnify the fundus images for generator. Next, we generated probability maps of retinal vessels. Experiments on two public fundus datasets, Digital Retinal Images for Vessel Extraction (DRIVE) and Structured Analysis of the Retina (STARE), showed that our method outperforms the state-ofthe-art methods in terms of the accuracy and the area under the receiver operating characteristic (ROC) curve.