Research On Diagnosis Of Fundus Diseases Based On Deep Learning

Fundus disease is the main factor that endangers human vision,and it is particularly important to be able to detect fundus disease in a timely manner.With the rapid development of fundus imaging technology,various imaging methods can show the health status of the fundus in an all-round way,helping doctors troubleshoot hidden diseases.Color fundus images are the most common method.Applying the deep learning method to the fundus disease diagnosis system can improve the efficiency of doctors’ diagnosis and reduce the occurrence of misdiagnosis.Macular disease is a comprehensive disease,which is a fundus disease that occurs in the macular area of the fundus.Because the macular area is the sensitive part of the eye,macular lesions are extremely harmful to vision.Diabetic retinopathy is a symptom of diabetic patients in the eye.Diabetic retinopathy can produce a variety of lesions in the fundus and affect the patient’s eye health.Therefore,timely detection of macular disease and diabetic retinopathy of the fundus is of great significance for protecting vision.The main work of this article is as follows:First,a method for macular localization and diagnosis of macular lesions was proposed.The method is divided into two parts: localization of macular center and disease diagnosis.Firstly,a method for localization of macular center based on deep clustering was proposed.The macular candidate area is divided by a certain rule,and the sub-regions are clustered in combination with the deep embedded clustering network to find the coordinates of the macular center and realize the location of the macula.Then,a method for diagnosing macular degeneration based on convolutional neural network is proposed.The two networks,Res Net50 and Dense Net121,are jointly trained to realize the determination of the abnormality of the macular region.The experimental results show that the method can be used for the diagnosis of macular degeneration well.Second,a method for diabetic retinopathy lesion segmentation based on U-net is proposed.This method uses U-net as the basic framework and adds dense connection modules to effectively enhance the transmission of feature information.In addition,this method also uses an attention mechanism,and an attention gate module is added to the network.During feature transfer,the attention gate module suppresses feature responses in irrelevant or low-correlation regions.The experimental results show that this method can accurately segment the three lesions of Hard Exudation,Hemorrhage and Soft Exudation,and the segmentation effect of Microaneurysm needs to be improved.