Research On Glaucoma Disease Classification Method Based On UBM Image

As the irreversible eye disease with the highest blindness rate,glaucoma has a long latency period,the clinical symptoms are not obvious in the early stage of the disease,and it gradually affects the patient’s vision,mostly in middle-aged and elderly people.However,this process is time-consuming and subjective.In order to better assist doctors in providing clinical diagnosis of glaucoma,this thesis proposes a study on the classification of glaucoma diseases based on UBM images,which is mainly as follows.1、Considering that the current medical public dataset is small and does not meet the experimental requirements of this thesis,the atrial horn structure of the UBM device was collected as a dataset,and the team led by a professional ophthalmologist correctly classified the dataset type,divided the atrial horn images into two major categories and five minor categories according to the atrial horn structure and carried out appropriate pre-processing for some images with unclear atrial horn structure and noise,which provided sufficient preparation for the subsequent experiments.2、In this thesis,based on the classified dataset,we propose a classification technique for automatic recognition of atrial angle opening and closing status based on the classification model of deep learning.Since the dataset is small,the existing network model of Image Net is migrated for learning and improved for the VGG19 network by modifying the number of fully connected layers and parameters,which makes the classification accuracy higher for atrial angle opening and closing status,and the algorithm model in this thesis has better results in classification performance compared with current popular classification network models experimentally,and also provides reference experience for further classification of closed angle glaucoma types.3、After the successful realization of atrial angle opening and closing state recognition,the atrial angle closing images are further classified more precisely into five categories including iris bulging and other types of glaucoma according to the relevant glaucoma types,which include the more representative ones containing both iris atrophy and iris adhesion and mixed types,which are more difficult to classify,this thesis proposes an improved image classification method for Xception for closed angle glaucoma.The similarity between atrial angle images is high,and an improved algorithm introducing L2 regularization is proposed for model training using the existing Xception network model.By judging the relevant classification evaluation indexes,it is demonstrated that the network model adopted in this thesis has a good classification effect for atrial angle closure types.This thesis takes the atrial angle images acquired by the UBM device as the core research,and based on the standard and reliable homemade data set,a recognition and classification method about glaucoma is realized by using the relevant algorithms of deep learning,which provides an auxiliary method for doctors in the early clinical diagnosis of glaucoma patients and further realizes the AI development of intelligent medical treatment with certain clinical application prospects.