Retinal Blood Vessel Segmentation by Multi-channel Deep Convolutional Autoencoder

Abstract

The evaluation and diagnosis of retina pathologies are usually made by the analysis of different image modalities that allows to explore its structure. The most popular retina image method is the retinography, a technique to show the retina and other structures in the fundus of the eye. This paper deals with an important stage of the retina image processing for a diagnosis tool which aims to show the blood vessel structure. Our proposal is based on a deep convolutional neural network, that avoids any preprocessing stage such as gray scale conversion, histogram equalization, and other image transformations that determine the final result. Thus, we obtain the blood vessel segmentation directly from the original RGB color retinography image. The results obtained with our method are comparable to the state-of-the art methods but using a smaller network with less memory and computation requirements. Our approach has been assessed using the DRIVE database.