Retinal Fundus Image Segmentation using U-Net Deep Neural Network

Abstract

Deep learning algorithms have become popular in machine learning and computer vision due to their ability to achieve state-of-the-art results. This paper proposes a method for accurately classifying retinal vessel segmentation using deep learning algorithms and state-of-the-art techniques. Specifically, the approach uses a U-Net convolutional neural network to extract vessels from retinal fundus images taken from the DRIVE dataset. Pre-processing is applied to the images to extract features such as shape, size, and arterio-venous crossing types, which can provide valuable insights into various eye diseases. The model is trained on five epochs and 300 layers, achieving an accuracy of 0.9691 and producing clear and accurate vessel segmentation. These results have the potential to aid computer-aided detection tasks in retinal images.