Detection Of Retinal Diseases Based On OCT Image And Machine Learning

Human retina is very fragile and easy to be damaged so that vision of the patient is damaged by different degrees.Age is one of the important factors affecting the human vision,and as the aging degree of our country is increasing,the demand of the diagnosis and treatment of the ophthalmic diseases has increased dramatically.OCT is a new imaging technique widely used in the diagnosis of ophthalmic diseases in recent years.In clinic,ophthalmologists often need to process huge OCT image data in a short period of time,which not only increases the work pressure of doctors,but also increases the probability of misdiagnosis,so that patients can not get timely and effective treatment.In order to solve the problems of more patients,fewer doctors and great medical pressure,it is of great significance to develop an automatic detection system for retinal diseases based on OCT images.This paper adopts the machine learning method to study the core problem in the automatic detection system of retinal diseases based on OCT images: the classification of retinal OCT images.The main work of this paper includes:1.The LPQ feature is applied to the detection of retinal disease,and an algorithm for retinal OCT image classification based on LPQ feature and support vector machine is proposed,and is compared with several common features: LBP feature,SIFT feature and Gabor feature by experiments.The experimental results show that LPQ feature have good performance in retinal OCT image classification.2.The canonical correlation analysis(CCA)is applied to the detection of retinal diseases,and two automatic detection algorithms of retinal diseases using CCA are proposed: one is to apply CCA for the design of classifiers,and the other is to use CCA for feature fusion.The experimental results show that the first algorithm can get the recognition rate close to that of SVM,and the calculation speed is faster than SVM.The second algorithm merges the two features and classifies them with SVM.Compared with the use of a single feature,the recognition rate is improved,especially when LPQ and LBP are combined,the recognition rate is improved obviously.3.An algorithm for detecting retinal diseases based on LPQ features and sparse coding with spatial pyramid matching(Sc SPM)is proposed.All OCT images are divided4.into image patches with fixed size,and some patches are randomly selected,and the LPQ features of the selected patches are extracted and used to learn a dictionary.After that,all patches in an image are sparsely encoded with the learned dictionary,and then the feature vector of the image are obtained by spatial pyramid pooling method.Finally,the image is classified by SVM.The experimental results show that the recognition rate of this method is much better than that of the common retinal OCT image classification algorithms.