Research And Implementation Of Auxiliary Diagnosis System For Retinal Vessel Segmentation Based On Deep Learning

In ophthalmology clinical medicine,the morphological structure of retinal blood vessels is an important basis for doctors to diagnose ophthalmological diseases and other related diseases.The structure of blood vessels in the retina is complex.In large-scale fundus examination,doctors spend a lot of energy to detect by the naked eye and may miss or misdiagnose.The use of computer segmentation of retinal blood vessels to assist doctors in diagnosis can improve diagnosis efficiency and reduce the rate of misdiagnosis.With the rapid development of deep learning technology and the wide application of convolutional neural networks in image recognition,more and more image processing technologies based on convolutional neural networks are used in medical image segmentation tasks.Compared with the traditional image segmentation technology,the segmentation algorithm using deep learning technology is more accurate.At present,medical image segmentation based on convolutional neural networks has become a research hotspot.This article applies deep learning technology to the research of retinal vascular segmentation,and develops the retinal vascular segmentation auxiliary diagnosis system,which can provide doctors with accurate retinal vascular segmentation service and ophthalmology remote auxiliary diagnosis service.The main contents of this study are as follows:(1)Design an optimization algorithm for retinal vessel segmentation based on deep learning.The preprocessing and image amplification methods of retinal fundus image data sets are proposed to prepare for the subsequent network model training.Select a suitable reference network model U-Net,combined with the characteristics of the retinal vascular segmentation task,the improved model is designed to make full use of context information by designing multi-path jump connections to improve network segmentation performance.Replace the standard convolution in the network with deformable convolution to improve the ability of the network to adapt to the geometric deformation of blood vessels,introduce batch normalization layers to avoid gradient dispersion or gradient explosion in the network,and add an improved residual module to deepen the network to further improve the network to extract feature information Ability.The improved network model is trained and evaluated on the DRIVE database.The experimental results show that the improved U-Net model has achieved a good segmentation effect on the retinal blood vessel segmentation task.(2)Adopt a test-driven development strategy to develop an auxiliary diagnostic system for automatic segmentation of retinal vessels.The system development adopts the B / S architecture,and separately developed a fundus image acquisition subsystem,a retinal vessel automatic segmentation subsystem,and a remote auxiliary diagnosis subsystem.Fundus image acquisition subsystem realizes basic patient information acquisition and fundus picture acquisition functions;automatic retinal vessel segmentation subsystem uses this improved U-Net model to segment the collected fundus images to achieve automatic retinal vessel segmentation service;remote assisted diagnosis subsystem By displaying the basic information of the patient,the fundus image,and the segmented fundus image to realize the reading function,the doctor gives diagnosis opinions after reading the film to realize the diagnosis function,and then forms a diagnosis report after diagnosis.In addition,the system displays patient information in the form of a list and adds a search function to implement patient information management functions.Finally,through system testing,it is concluded that the developed system has certain practicality in ophthalmology clinical auxiliary diagnosis.Figure [34] table [12] reference [76]...