Research On Fast Screening Method Of Fundus Images Based On Big Data Of Physical Examination

Todays,eye diseases have become a worldwide topic.In China,the abnormal rate of fundus examination among people who are over 60 years old is higher than 21.39% in 1.4billion.In daily life,it is very difficult for people to realize that they have an eye disease because there are rarely obvious symptoms in the early stages of the disease.In addition,the eye care and protection is not sufficient,besides the annual physical examination test report,few people may test the eye.At the same time,there are few ophthalmologists in China.According to the statistics of the National Health Commission,there are only 32,000 ophthalmologists,and the low proportion of ophthalmologists and patients leads to the lack of timely diagnosis and treatment.Therefore,it is urgent and necessary to realize rapid screening of ophthalmological diseases through the analysis of fundus images.With the development of science and technology,the invention of fundus camera for non-invasive shooting of fundus images has brought convenience to the early diagnosis of ophthalmology patients.With the development of computer technology and various image processing technologies,the application of convolutional neural network to assist the diagnosis of ophthalmic diseases has become an effective means,for example,fundus image screening technology has been successfully applied in clinical practice.But these diagnosis technology can only be used in the diagnosis of single eye disease,such as single,glaucoma,diabetic retinal pathological changes,and congenital cataract.While large-scale physical fundus image contains various disease also lack corresponding technique to distinguish and rapid screen.This study started from a large number of screening problems of fundus images in physical examinations,and took the accuracy and speed of disease screening as the goals.It uses fundus image data sets to train convolutional neural networks,and a high-precision fundus image disease prediction model was obtained.Firstly,the raw images are processed uniformly,all images are removed from redundant borders,image resolution is standardized and enhanced.Secondly,the processed images are used in different convolutional neural networks and functions to obtain different prediction results.Thirdly,the Inception V3 model was selected as the main body of the network to build the program.Then,the structural characteristics of the fundus image and the clinical symptoms of various ophthalmic diseases were deeply analyzed,and the method of optimizing the enhanced image and the composition of the enhanced data set were proposed to improve the training of the model.Finally,the medical fundus images collected in the hospital are used to verify the practicability of the network model.The fundus image prediction model divides all ophthalmic diseases into 8 categories: normal,diabetic retinopathy,glaucoma,cataract,agerelated macular,hypertensive retinopathy,myopia and others.It turns out that the accuracy of the fundus image prediction model was 93.23% in the study.After the physical examination of the fundus image,the accuracy of the model was90.73%.And,it only takes 3 minutes and 20 seconds to obtain the detection results of 400 fundus images.The prediction model can be effectively used for rapid screening of fundus images for large-scale medical examinations.Compared with the prediction results of other models,the accuracy of the overall prediction model of this research model is relatively stable,and the prediction results of a single eye disease are close to the results of clinical decision.The prediction model can make good use of the information in the fundus image,can be used in the screening of the fundus image of a large-scale physical examination,and can give the predicted probability of various diseases in each image.This research model can detect disease information in fundus images which assists doctors in diagnosing eye diseases,and help patients receive treatment in time.