Extraction And Recognition Of Lesions In Fundus Images

In recent years,due to the wide application scenarios of retinal images,retinal image segmentation has attracted more and more scholars’ attention and research.The retina itself is very stable and is not prone to wear and degradation.Clinically,doctors can diagnose and treat a variety of ophthalmic diseases,including diabetic retinopathy,macular degeneration,glaucoma,and cataract by collecting the fundus image of the patient.At the same time,with the popularity of computers,data processing technology has developed rapidly and gradually applied to all aspects of the medical field.Among them,medical image processing technology has made great contributions to medical development and human health.The eyeball includes a variety of structures.In the retinal images taken by the fundus camera,lesions in the fundus,macula,and optic disc are common.Based on image processing and machine learning methods,this paper extracts from local lesions and identifies global lesions.And the method of doing related methods in the judgment:1.Firstly pre-processing the fundus image,including the selection of the color space and the enhancement and equalization of the image;2.In the extraction of local lesions,the fundus vessels are first segmented,and the microaneurysms in vascular lesions are extracted,finally grade the micro-aneurysm lesions points;3.The hard exudate of the fundus is extracted.The evaluation of the algorithm uses the relevant classifier in machine learning to train and test the segmented lesion pixels and calculate various indicators;4.In the global lesion recognition,the convolutional neural network is used to identify and judge the fundus image lesions and normal,and the network is used to identify the single lesions of microaneurysms and hard exudates.In the course of the research,the main focus of the clinical manifestations of diabetic retinopathy,the use of relevant tag data sets,to achieve the extraction and identification of lesions,in the future work,will pay more attention to the use of unsupervised learning for clinical data Set the lesion detection work to improve the applicability of the algorithm.