A Fundus Fluorescein Angiography-based Intelligent Diagnosis System For Neovascular And Macular Edem Of Diabetic Retinopathy

Objective: To combine fundus fluorescein angiography pictures of diabetic patients with computer deep learning technology and image registration,use a large number of fluorescence angiography pictures of neovascularization and macular edema to learn and train the intelligent diagnostic system,and set up an intelligent diagnosis system for diabetic retinopathy neovascularization and macular edema based on fundus fluorescence imaging.The intelligent diagnosis system is then tested and verified to obtain the corresponding sensitivity and specificity.Methods: A total of 30034 fundus fluorescein pictures of 682 diabetic patients from 2 different medical institutions were collected.Three fundus specialists were invited to validate and label the fluorescein pictures "back-to-back".The majority agreement principle in conjunction with the clinical diagnosis report were used to label single picture with different opinions.Using random selection approach,608 accurately labeled patients data(a total of 25695 images)were selected out of 682 patients as the training/validation set,and the labeled data of the remaining 74 patients(a total of 3253 images)as the test set for the training and testing of the intelligent diagnosis system,and finally the system was evaluated with sensitivity and specificity measures.Results: The internal and external verifications were twice used to verification and test the intelligent diagnostic system.The sensitivity and accuracy of the first internally verified neovascularization were 0.524 and 0.185 and the macular edema sensitivity and accuracy were 715 and 0.556,respectively.The sensitivity and accuracy of the first externally verified neovascularization were 0.538 and 0.151,and the sensitivity and accuracy of macular edema were 0.812 and 0.605,respectively.The sensitivity and specificity of the second internally verified neovascularization were 0.914 and 0.780,and the macular edema sensitivity and specificity were 0.928 and 0.875,respectively.The sensitivity and specificity of the second externally verified neovascularization were 0.963 and 0.786,and the sensitivity and specificity of macular edema were 0.823 and 0.857,respectively.Conclusion: Under the control of big data,deep learning computing technology and image registration technology were used to train and learn computer intelligent diagnosis system models.United medical technology team and computer technology team.The approach is expected to develop a set of intelligent diagnosis systems for diabetic retinopathy neovascularization and macular edema based on fundus fluorescein angiography,assisting ophthalmologists to quickly complete the diagnosis of neovascularization and macular edema in diabetic retinopathy,so as to reduce the workload of doctors and improve work efficiency.At the same time,it lays a solid foundation for the next development of an intelligent diagnostic system for diabetic retinopathy based on fundus fluorescence angiography.In our country with a large population,a shortage of medical resources,and extremely uneven distribution,the development and use of artificial intelligence-assisted systems is undoubtedly a good choice.