Fully Convolved Neural Network-Based Retinal Vessel Segmentation with Entropy Loss Function

Abstract

The eye is the exclusive organ for the sense of sight in humans. Morphological changes in vascular diameter and branching pattern of retinal vessels lead to blindness. Segmentation of retinal vessels is done to analyse these morphological changes in retinal vessels. However, due to the presence of illumination, multiplex distribution of blood vessels, and low contrast between target and background, the task of segmentation of retinal blood vessels is highly challenging. In this chapter to segment retinal blood vessels, we propose a method based on fully convolutional neural networks and pixel classification with cross-entropy function to avoid the class imbalance problem. Our proposed architecture of fully convolutional neural networks combines the output of each stage to learn the hard samples. The cross-entropy loss function is performed to avoid misclassification of vessels.