The U-Net model application for retinal vessels segmentation using the machine learning library TensorFlow

Abstract

In this article the implementation of neural network architecture based on a dense U-Net network is proposed. It is noted that retinal blood vessels are the basis for clinical diagnosis of some diseases. A review of the convolutional networks use for classification tasks and generalizion retinal vessel segmentation algorithms is performed. The general process of the neural network is presented. The differences between the real and the obtained results were evaluated. Evaluation of the neural network is carried out on several parameters. The figure with the recognized blood vessels as a result of the model is presented.