Research On Retinal Status Automatic Analysis Method From Fundus OCT Images

Optical coherence tomography(OCT)can quickly acquire the microscopic image of retina.It is an objective index for the treatment of diseases and follow-up observation,such as finding small lesions in the fundus of the eye and judging the pathological changes of diseases.Owing to the unique tissue complexity and fuzziness of retinal OCT images,it is clinically dependent on experienced physicians to make artificial judgments.With the development of information science,computer image processing and analysis technology has been applied to OCT images,which brings convenience to the diagnosis of retinal diseases and promotes the development of medical industry.In this paper,the OCT image of the fundus is taken as the research object and the analysis process of doctors is used for reference.Finally,a theoretical framework of automatic analysis of retinal status based on OCT image of the fundus is proposed,which includes definition and establishment of normal retinal reference model,quantitative feature extraction and abnormality diagnosis.It lays a solid foundation for the practical application of computer on-line and remote diagnosis technology.The main research works in this paper are as follows:1)The definition and establishment method of normal retina reference model based on retinal stratification are proposed.Ophthalmologists use OCT images to observe changes in retinal characteristics and morphology.The morphological changes of the retina caused by pathological changes will directly affect the morphology and position of the corresponding tissue layer of the retina,and the boundary of the tissue layer of the retina can show its morphological changes.In order to solve the problem that single structure element can't extract many kinds of morphological objects,and the traditional morphological structure element is monotonous and fixed,a new immune genetic morphological operator(IGM)is proposed in this paper.This method uses the immune genetic algorithm to construct the adaptive morphological structure elements,which can better express the feature information of the image itself.Normal retinal morphology is the reference for medical diagnosis.In this paper,we construct a universal reference model of normal retina to represent normal retinal morphology through the methods of retinal boundary extraction and statistical modeling.By comparing the characteristic value of OCT instrument with the quantized value of reference model,the validity of reference model is proved.2)Retinal quantitative feature extraction method based on normal retinal reference model is proposed.In view of the limitation that OCT instrument can only provide a small number of numerical features and the existing computer analysis is limited to several specific fundus diseases,which does not have universality,and the problem of texture feature selection and extraction,a method of retinal feature tuning and extraction is proposed.Firstly,by comparing the accuracy of the combination of multiple texture features and machine learning methods in retinal image classification,this method validates and extracts efficient texture features,which can be used in the construction of integrated classifier in subsequent research.Secondly,according to the shape features of retina,based on the normal retina reference model and the morphological changes of retina that doctors pay attention to in the process of clinical diagnosis,this method extracts and generates a series of retinal shape features that are suitable for computer quantitative representation.The experimental results show that the quantitative features of the reference model can not only provide doctors with the reference value of normal retinal shape features,but also compare and analyze the location and severity of the lesions in the abnormal images,providing the basis for the subsequent abnormal judgment.3)A computer-aided diagnosis method based on clinical observation and analysis flow of doctors is proposed.In view of the fact that the results of manual analysis are highly dependent on the doctors' personal knowledge and experience,a kind of auxiliary diagnostic method of fundus retinas is proposed based on the realization of retinal stratification and the quantification of computer features and the process of observing and analyzing the retinal state in OCT images by referring to the doctors' observation and analysis of OCT images.Firstly,an integrated classifier was built based on the verification results of setting retinal texture features to realize rapid screening of retinal images and classification of abnormal retinal images.Secondly,in view of the rare research problems about the location and severity of lesions,based on the geometric and morphological characteristics of the retina,the classification decision of different levels of abnormalities was designed,and the judgment of the abnormal position and degree of the image was realized.Finally,through the design of computer-aided diagnosis system,integrated image processing,feature quantification,automatic diagnosis,report generation of multiple modules,to achieve the automatic analysis of retinal OCT images,for the diagnosis of fundus diseases to provide reference.