Research And Application Of OCT Image Recognition Of Fundus Retina

There are many kinds of retinal diseases in the fundus,and the causes of the disease are complicated.For example,diabetes-induced Diabetic Macular Edema?DME?,age-related macular degeneration,degenerative myopia,choroidal neovascularization?CNV?caused by fundus vascular streaks,and drusen caused by pigment epithelial dysfunction?Drusen,DRUSEN?and so on.Early screening and diagnosis of these diseases is very important.However,our country has a large population base and faces more and more patients with fundus diseases.Ophthalmologists often have to deal with huge fundus image data and make a diagnosis in a short period of time.result.In the face of massive data,manual diagnosis is extremely time consuming and labor intensive.According to the statistics of the Chinese Medical Association,the number of doctors per 1,000 people in China is only 2.1,so there is a serious gap between supply and demand of professional ophthalmologists in this area.The patient's condition cannot be treated promptly and effectively^[1].Therefore,in order to solve the above problems,the main work of this paper is to apply the artificial intelligence assisted diagnosis technology to the fundus medical imaging field to improve the work efficiency of medical institutions and doctors.The main work of this paper is as follows:?1?In the image recognition task of fundus retina optical coherence tomography,the two open source retinal?OCT?image data sets of 2014?BOE?Srinivasan and OCT2017are mainly used,and the corresponding labels in the data set are manually processed by professional ophthalmologists.Mark it.Based on these OCT image data,nine convolutional neural network models of Resnet18,Resnet34,Resnet50,Vgg16,Vgg19,Inception?v4,Inception?resnet?v2,Cnn3 and Resnet50?vgg16 are tested.?2?In order to enable doctors to better observe and verify the recognition results of Resnet50?vgg16 model for diabetic macular edema lesions,this paper further completed the segmentation task of retinal OCT image lesions.The experiment used the Unet model to implement the segmentation task of Diabetic Macular Edema?DME?on the open source2015?BOE?Chiu fundus retina dataset.?3?Based on the best convolutional neural network model for fundus retinal OCT image recognition and segmentation results,this paper developed a retinal OCT image automatic recognition system.This paper uses cutting-edge algorithms in the field of computer vision to solve OCT image recognition problems in the medical ophthalmology field.Experimental results show that Resnet50?vgg16 has a better recognition effect.The automatic retina OCT image recognition system can realize the automatic screening and diagnosis of fundus OCT images.The recognition accuracy is high,and it can assist doctors to quickly check more patients.