Research And Implementation Of Deep Learning-based Accurate Quantifying Cycloplegic Refraction Model For Adolescent Myopia And Cloud-edge Collaborative Screening System

According to the report of the National Health and Medical Commission,the total incidence of myopia among children and adolescents in China is about 53.6%.Myopia can develop into a high degree of myopia,causing a series of complications,such as macular degeneration,retrosternal staphyloma,retinal detachment,etc.What does not match the health threat of myopia to adolescents is the lack of resources for ophthalmologists in the field of myopia prevention and treatment very seriously.With the development of pilot work and various research on suitable technologies for the prevention and control of myopia in eye hospitals in various provinces,municipalities,and autonomous regions,related technologies to improve screening efficiency and assessment accuracy through artificial intelligence technology are also urgently needed.Fundus photography,which has a high proportion of equipment ownership,has become an important entry point and force point for suitable technologies for myopia prevention and control.However,in most of the current public research work,there are still problems and challenges such as the lack of research on data quality improvement in real scenes,the low accuracy of youth myopia diopter evaluation,and the difficulty of deploying cloud models with large delays in screening scenarios.Given the above problems and challenges,this paper mainly studies the following contents:1)A medical image standardization and data quality control system is provided and implemented.This system provides pipeline processing for medical images:after image standardization,the quality of image data is stratified according to big data statistical algorithms,and some types of data are of different quality.The high case proposes deep learning-based augmentation and inpainting algorithms for data reuse.2)An accurate assessment model of juvenile myopia based on deep learning is proposed and implemented.Through the non-invasive fundus photographic images that can be quickly obtained,the myopia status can be accurately assessed through a neural network algorithm,to solve the problem of inaccurate myopia screening and optometry results.3)Design and implement a system implementation and deployment based on cloud-edge collaborative multi-level feedback,obtain instant feedback of optometry results by running lightweight models on edge devices,and provide more accurate delayed feedback combined with models on cloud devices to solve the problem.Screen for problems in complex network.environments in the region.