Research On Fast Segmentation Method Of Diabetic Macular Edema

Diabetic retinopathy(DR)is the clinical common eye disease and one of the main causes of blindness in the world.With the improvement of people’s living standards in our country,the incidence and blindness rate of diabetic retinopathy has increased significantly,which seriously affects patients’ vision function and quality of life.Diabetic macular edema(DME)is a common cause of visual impairment in diabetic patients.Diabetic macular edema(DME)in early detection and monitoring has positive significance for the treatment of diabetic retinopathy,prevention of patients with visual impairment.Optical coherence tomography(OCT)is a noninvasive and non contact imaging method that provides information on the morphology and tissue of the retina.Clinically,OCT image segmentation of diabetic macular edema is a necessary work in ophthalmic diagnosis.However,manual annotation of macular edema is affected by subjective factors and labor-intensive factors.Therefore,IT has become a new trend to segment the macular edema area with the help of computer and other IT technologies to improve the efficiency and accuracy of doctors’ diagnosis.Based on the analysis of the existing segmentation methods for macular edema,this paper studied a rapid segmentation method for diabetic macular edema using deep learning,including:(1)An improved level set segmentation method for diabetic macular edema(DMESBGFRLS)was proposed:An improved level set segmentation method for diabetic macular edema(DMESBGFRLS)was proposed according to the heterogeneity of OCT images and the diversity of DME region features and the ambiguity of DME region boundaries,including :1)The rough segmentation of macular edema in OCT images was realized based on K-means clustering algorithm,which improved the efficiency of the segmentation of macular edema;2)Taking the rough segmentation result as the initial curve,the level set method was used to achieve the fine segmentation of macular edema area.Compared with the existing level set method C-V(Chen-Vese),GAC(Geodesic Active Contour),SBGFRLS and the manual segmentation results,the results show that the DME-SBGFRLS method proposed in this paper improves the efficiency and accuracy of segmentation,and its segmentation accuracy is similar to that of manual segmentation.(2)A neural network model for the segmentation of diabetic macular edema(DMEDeep Lab)was constructed:The DME-SBGFRLS method is used to achieve high-precision segmentation of DME,which provides a sample labeling method for constructing a neural network model of diabetic macular edema segmentation.To improve the segmentation efficiency of diabetic macular edema,a neural network model for the segmentation of diabetic macular edema(DME-Deep Lab)was constructed based on Deep Lab network,including: 1)Speckle noise removal of OCT image is realized by wavelet transform;2)Rough segmentation of diabetic macular edema was achieved by using Deep Lab network;3)FCCRF was used to refine the boundary of diabetic macular edema with coarse segmentation as input.Compared with the traditional segmentation methods,end-to-end segmentation model and manual labeling results,the results show that the proposed model improves the accuracy and efficiency of the macular edema area segmentation,and the model has strong robustness,which is helpful for ophthalmologists to classify the edema area and improve the efficiency of DME diagnosis.Experiments show that the DME-SBGFRLS method proposed in this paper improves the accuracy of diabetic macular edema segmentation.The proposed DME-Deep Lab model improves the segmentation accuracy of diabetic macular edema and the efficiency of segmentation.Research results are expected to improve the efficiency and accuracy of screening for diabetic retinopathy.