The Research On The Retinal Fluid Segmentation Technology For Fundus Lesions

As an important basis for patient disease detection,fundus images are widely used in the medical field.Which not only can reflect related ocular lesions,but also can visually show the body illness,such as high blood pressure,cardiopathy and so on.In the diagnosis,the image,taking samples from patient’s eyeground,is the diagnostic basis of ocular changes and it can be used to analyze what’s the patient’s disease by doctors.The ocular fundus image is more visualized and clear than naked eye in drawing out ocular information,which can be good for diseases diagnosis.While enhancing efficiency of doctors’ works,it also provides more accurate diagnosis results for patients.Furthermore,the pretreatment of images through computer technology can provide doctors with some or all of the available diagnosis reports and can better help doctors and patients,which is the significance of auxiliary medical diagnosis.In this article,OCT(Optical Coherence Tomography)image macular edema segmentation is mainly divided into two parts: cluster segmentation and level set segmentation.In the work of cluster segmentation,the complete segmentation process includes three steps: image preprocessing,cluster segmentation and result optimization.The main work of the pre-processing steps is summarized as follows: background removal of the image to be segmented,reducing the pixel data that needs to be processed while ensuring image quality;explaining the application of neutrosophic theory in image segmentation,using neutrosophic domain conversion to complete image denoising;combined with the image characteristics of the neutrosophic image,the location coordinates of macular edema/effusion region were obtained.In the stage of clustering segmentation,this article first tested the performance of some improved algorithms based on Fuzzy C-means(FCM)clustering method in macular edema segmentation,and found that only relying on pixel intensity or local pixel features cannot achieve segmentation performance improvement;therefore;According to the distribution characteristics of OCT image effusion,it is proposed to add a position offset restriction item to the FCM prototype objective function,so that the position offset metric can be attached to the membership calculation;In the initial clustering parameter setting,using the effusion position obtained in the preprocess constructs the initial clustering center,then the time efficiency of the clustering algorithm can be better.In the result optimization stage,the effusion area distribution obtained by clustering segmentation is screened,and the missegmented areas that are regularly distributed in the data set are eliminated to achieve further improvement of the segmentation effect.In the work of level set segmentation,the segmentation is divided into two parts: preprocessing and level set segmentation.In the preprocessing part,the segmentation image in the preprocessing part can be done the largest connected component extraction,which is regarded as a ROI(region of interest)of subsequent segmentation.In the stage of level set segmentation,the shortcomings of C-V(Chan and Vese)model in weak edge recognition are discussed,and the region evolution segmentation is proposed by combining the characteristics of two groups of images with the joint energy driven term,which realizes the weak edge recognition of OCT images.In order to accurately converge to the desired boundary of the fluid region,the introduction of evolution control function is according to the features of OCT image’s fluid region,and finally the evolution curve stops at the boundary of the fluid region.