Research On Auxiliary Diagnosis System Of Fundus Diseases Based On Deep Learning

In recent years,with the development of science and technology and the progress of productivity,the possibility of human suffering from eye diseases is also higher and higher,and more and more people are difficult to live and survive because of eye diseases.Myopia,retinal detachment,retinal vein occlusion,diabetic retinopathy and other diseases not only need to be diagnosed in time,but more importantly,patients hope to detect diseases early in the disease and treat them as early as possible.As more and more people suffer from different degrees of eye diseases,the eye department in the medical system is gradually unable to do what it wants.Therefore,it is of great significance to use computer technology to assist doctors in diagnosis for all kinds of photos taken by ophthalmology,which can greatly increase the efficiency of doctors and reduce the waiting time of patients.A lesion detection system based on ultra wide-angle fundus photography is designed to detect normal eyes,pathological myopia,retinal detachment,retinal vein occlusion and diabetic retinopathy.Based on efficientnet network,aiming at the characteristics of ultra wide-angle fundus photography,we add attention module and regularization parameters to the original network,and achieved excellent results.Finally,our network achieved a good result of 92.57% of the total accuracy,and the results of each lesion classification are also satisfactory.At the same time,we also use the same data set to compare with resnet50 network and resnet101 network,which proves that our improved network can more accurately diagnose diseases in ultra wide angle fundus photography.At the same time,we also designed a Mask-RCNN network based on the two stage image segmentation in view of bleeding and diabetic lesions in diabetic retina.The network extracts the features in the input image,accurately divides the candidate boxes of each target,and then judges the possible lesion areas in the candidate boxes pixel by pixel.We introduce an additional attention module into the network to further optimize the detection effect of the network.Finally,our network achieved good results in the detection of hemorrhagic lesions and cotton velvet lesions by 90.01% and 87.64%respectively.It is proved that the location of specific lesions and lesion areas in diabetic retina can be extracted by using deep learning technology,which can help doctors to diagnose results and increase work efficiency in the process of clinical diagnosis.A complete set of auxiliary diagnosis system for fundus lesions can not only greatly reduce the workload of doctors and accelerate the speed and accuracy of disease diagnosis,but also combine with remote diagnosis and treatment and remote medical technology.It can also provide benefits for patients in remote areas who cannot obtain superior medical resources,so as to truly realize the great vision of intelligent medical treatment.