Automatic Detection of Blood Vessels in Retinal Images Using FC-DenseNet Neural Networks

Abstract

Retinal blood vessel detection and analysis through medical imaging is very important for a support system in diagnosing and establishing a treatment for retinal diseases. In this sense, the use of high-performance neural networks for the semantic segmentation of blood vessels in retinal images is a trend in such applications. The paper proposes two neural segmentation networks, called DensegNet1 and DensegNet3, respectively, based on the combination of elements from classical networks: SegNet and DenseNet. The detection and segmentation performances based on the CHASE-DB1 database are analyzed, both for large vessels and for thin vessels. The results were promising and the aim is to integrate the analyzed networks into even more efficient structures of complex GAN type.