Analysis of Convolutional Neural Network for Fundus Image Segmentation

In this paper the study of fundus image segmentation using convolutional neural networks is carried out. A neural network architecture was made to classify four classes of images, which are made up of thick and thin blood vessels, healthy areas, and exudate areas. The CNN architecture was constructed empirically so as the required accuracy of no less than 96 % is ensured. The segmentation error was calculated on the exudates class, which is key for laser coagulation surgery. In the paper we utilized the HSL color model because it renders color characteristics of eye blood vessels and exudates most adequately. We have demonstrated the H channel to be most informative. We have investigated the robustness of technology to various noises. Experimental studies have shown the instability of the convolutional neural network to Gaussian white noise and resistance to impulse noise.