Eye fundus retina image segmentation method based on deep convolutional neural network

Abstract

The invention discloses a fundus retina image segmentation method based on a deep convolutional neural network. The method employs the deep convolutional neural network to map features of a vascular tissue, an optic disc optic cup tissue and a lesion tissue in a medical image, and employs the convolutional network to segment the image. In addition, in order to increase segmentation accuracy, a new data preprocessing method of the fundus retina image is used for enhancing the image; an end-to-end deep convolutional network is used for solving a problem of small blood vessel segmentation, and deep significant features of a lesion area are obtained and visualized; a series of problems caused by large pixels of various medical images are solved by using a method of combining multiple deep neural networks.

Description

technical field [0001] The invention belongs to the technical field of medical image processing, and relates to a method for segmenting a retinal fundus image, in particular to a method for segmenting a fundus retinal image based on a deep convolutional neural network. Background technique [0002] Retinal images have been widely used in the diagnosis, screening, and treatment of cardiovascular and ophthalmic diseases, including systemic diseases such as age-related macular degeneration, glaucoma, hypertension, arteriosclerosis, and cardiovascular disease. Vessel, optic disc, and optic cup segmentation are fundamental steps required for quantitative analysis of retinal images. First, the division of the optic disc and optic cup is usually used for the detection of glaucoma. The optic nerve head assessment method is a convenient and widely accepted method used by clinicians. This method calculates the risk of glaucoma by using the ratio of the vertical cup-to-disk ratio, so ...

AI Technical Summary

Problems solved by technology

However, these methods have certain problems: 1) simple image preprocessing methods cannot effectively utilize the structural information of the feature map, which may hinder the segmentation performance; 2) use downsampling to improve the receptive field, but jointly segment the optic disc and cup , because the optic disc area on the label map is relatively small, an excessively large downsampling factor will cause the loss of these feature information; 3) For an excessively large segmented area, the receptive field in these methods is not large enough to fully understand the global 4) These methods use deconvolution for upsampling, which will increase the parameters and calculation of the model and affect the performance of the model

Images