Research And Implementation Of Computer Aid Diagnosis System For Ophthalmic Diseases Based On Deep Learning

Deep learning technology has been the first mock exam in ophthalmology in recent years.It has attracted extensive research in the field of intelligent prediction of ophthalmic diseases.However,it still faces the following problems:(1)most of the current studies only use single modality image as a model to assist in the diagnosis of disease input,which does not accord with the actual process of clinical diagnosis of most eye diseases.In addition,there are many kinds of ophthalmic diseases and the incidence rate is seriously unbalanced,and there are many rare ophthalmic diseases.However,most of the image data used in the existing studies have a balanced distribution of disease types and a small number of disease types,which limits the application of deep learning technology in real clinical scenes.(2)Secondly,in the field of deep learning in traditional scenes,data annotation and acquisition is relatively simple,while in ophthalmic images with high requirements for professional level,data acquisition and annotation are relatively difficult,resulting in a large number of valuable ophthalmic image data can not be used by artificial intelligence,which hinders the development of deep learning technology in the field of ophthalmic images.(3)Finally,the current research lacks a model deployment platform that can support a variety of intelligent auxiliary diagnosis tasks of ophthalmic diseases.To solve the above problems,the main research contents and achievements of this paper are as follows:(1)a multi-modal ophthalmic image convolution neural network M2LC-Net is proposed,which can effectively use multi-modal ophthalmic images to improve the accuracy of ophthalmic disease classification,and solve the problem of poor classification effect of long tail caused by uneven distribution of disease types in clinical scenes,At the same time,Grad-CAM module is used to visualize the contribution of M2LC-Net to disease prediction,so as to provide interpretability for the prediction results of the model.M2LC-Net has been effectively verified in 34396 images of the third class a hospital.Compared with the latest technology,various performance indicators have been significantly improved.The Cohen’s Kappa coefficient increased by 3.21%.(2)Research and implement an ophthalmic data acquisition and annotation platform based on deep learning,complete ophthalmic data acquisition and annotation path through architecture design,and provide model pre annotation ability by enabling deep learning ability for annotation system,so as to effectively improve the annotation efficiency in the field of ophthalmic image annotation.The ophthalmic image annotation system has been implemented in the ophthalmic hospitals of the two laboratory cooperation units,with a total of more than 60000 ophthalmic images annotated,supporting the research and implementation of laboratory related ophthalmic image artificial intelligence algorithm.(3)This paper proposes and implements a multimodal intelligent disease assistant diagnosis system for ophthalmic images,which provides model deployment of intelligent recognition and subdivision tasks of various ophthalmic diseases,as well as visual prediction ability.The implementation of the multi-modal auxiliary diagnosis system for ophthalmic images has integrated the deployment of artificial intelligence models for various types of ophthalmic diseases.