Research On Key Techniques In Ophthalmology Images Analysis

Cataract is one of the ocular fundus diseases with high incidence.Previous research shows that a majority of the patients in severe visually impairments suffered from cataract disease and that ophthalmology images is one of the most indicators in diagnosing eye disease.With the development of Computer-Aided healthcare system,ophthalmologist's burden has be eased more and more.Therefore,the analysis of ophthalmology images possesses great value in research and clinical significance.Some diseases,such as glaucoma,diabetes eye disease,kidney disease,hypertension,bacterial diseases and so on can be diagnosed by fundus examination.According to the clinical requirements and reality,the key technologies of Ophthalmology images analysis have been studied.In this paper,from the perspective of practical application that unlabeled examples are easier obtained than labeled ones,semi-supervised learning is used in analyzing ophthalmology images.The main contents of this paper are as follows:Based on the practical application,we choose more suitable methods for the diagnosis of cataract disease,including extracting G channel from origin ophthalmology images,using morphological methods to enhance the images and removing noise by means of a trilateral filter.The wavelets coefficients were extracted by the Harr wavelet transform and extract texture features from ophthalmology images,both of which act as the ground of later classification and grading.Because of the data sets in this experiment is made up of a small number of labeled samples and a large number of unlabeled samples,with the addition of comparing to labeled samples,gaining unlabeled samples are easier.So we utilize co-training to analysis ophthalmology image,which can lighten the heavy burden on ophthalmologists and improve ophthalmic healthcare quality both to a great degree.Experiments on real data sets by Tongren hospital included 476 labeled examples and 4902 unlabeled examples.Through pre-processing the medical data,extracting features of ophthalmology images and then utilizing Tri-training on the data for cataract grading four class based on clinical experience.The best performance of the experiment result up to 90%.Experiment results provide a bright future in later analysis on ophthalmology images.It also illustrates the effectiveness of the proposed approach named semi-supervised learning in cataract classification and grading.