Research And Implementation Of An Intelligent Auxiliary Diagnosis System For Fundus Diseases Based On Machine Learning

The intelligent auxiliary diagnosis system for fundus diseases can provide ophthalmologists with fast,accurate and reliable assistance.It not only reduces the diagnostic workload of ophthalmologists,but also enables patients to obtain more timely and convenient diagnosis and treatment services,reduces the cost of patient to a certain extent,and alleviates the imbalance between patient needs and limited ophthalmologists.Among the common fundus diseases,age-related macular degeneration(AMD)is a fundus disease that occurs mostly in the elderly and has a serious impact on visual function.Due to the aging population,the number of AMD patients in China is increasing.At the same time,according to clinical manifestations and pathological changes,AMD is divided into dryAMD and wet-AMD.Wet-AMD has international medically recognized treatment standards,and it has better diagnosis and treatment significance than dry-AMD,therefore,this paper intends to take wet-AMD as the representative of fundus diseases,and carry out the research and realization of the intelligent auxiliary diagnosis system for wet-AMD.In recent years,the outstanding performance of machine learning in computer vision tasks has promoted its development in medical field.In particular,deep learning technology has become an important development direction in medical image recognition and processing.In the field of assisted diagnosis of wet-AMD,many researchers have used deep learning models to achieve intelligent assisted diagnosis.However,most researches still have the following problems:1)Only a single modal medical image is considered,which does not meet the standard process of clinical diagnosis;2)No more fine-grained classification of wet AMD,which limits the assistance of clinicians;3)There has not been a practical auxiliary diagnosis system for wet AMD.Research only exists in the field of scientific research and cannot really benefit clinicians.In response to the above problems and challenges,this paper uses deep learning technology to design and implement an intelligent auxiliary diagnosis system for fundus diseases.The system realizes the subclassification of the two subtypes of wet-AMD,improves the diagnostic efficiency and accuracy of ophthalmologists,and provides patients with more timely and convenient disease diagnosis and treatment services.The main research contents of this article are as follows:1)A dual-modality wet-AMD auxiliary diagnosis model Wet-AMDNet is proposed.This model comprehensively analyzes color fundus images and optical coherence tomography(OCT)images,and uses 3 feature fusion strategies for auxiliary diagnosis of wet-AMD subtypes.The area under the receiver operating characteristic curve(AUROC),recall,and precision of this model reached 0.9881,0.9792,and 0.9821,respectively,exceeding the average level of 4 ophthalmologists with many years of work experience.At the same time,we cooperated with Hebei Eye Hospital to construct a bimodal wet-AMD dataset,which contains a total of 469 sets of bimodal image data,and annotated with 3 disease labels.2)A knowledge-driven fine-grained classification model for wet-AMD,KFWC-Model,is proposed.This model improves the image feature processing and feature location capabilities by artificially introducing the prior knowledge of the disease signs marked by medical experts,enhancing the model’s classification accuracy for subcategorized diseases in the presence of insufficient data.This model achieves an AUROC of 0.9971 and an accuracy of 0.9373,which is an improvement of 0.0669 and 0.0736,respectively,compared to the data-driven approach.At the same time,we cooperated with Hebei Eye Hospital to construct a bimodal wet-AMD dataset containing prior knowledge of lesion signs,including 5261 sets of bimodal image data,and annotated with 10 lesion sign labels and 3 disease labels.3)A dual-mode intelligent auxiliary diagnosis system for wet AMD is designed and implemented.Based on the above two research contents,the system encapsulates Wet-AMD-Net and KFWC-Model in the back-end,and provides clinicians with convenient and fast auxiliary diagnosis of wetAMD with a friendly front-end interactive interface.In addition to the core auxiliary diagnostic functions,the system also includes auxiliary functions such as user management,data management,and diagnostic report export,which can provide assistance for the whole process of clinical diagnosis and treatment of wet-AMD.Finally,the system was applied in the diagnosis and treatment process of Hebei Eye Hospital.The efficiency,accuracy and convenience of the system have been preliminarily verified in real diagnosis and treatment scenarios,laying a solid foundation for more research on intelligent auxiliary diagnosis of fundus diseases and further community promotion to benefit thousands of families.