Study On The Segmentation Of Optic Disc And Exudates In Fundus Images

The diabetic retinopathy (DR) is a microvascular complication of diabetes, which affects blood vessels and causes abnormalities in the retina. Diabetic macular edema (DME) is a complication of DR and is the most common cause of vision loss and blindness. Automatic detection of early signs of DR and treatment is essential to prevent vision loss. So it is important to study the automatic detection of the lesions of DR. After analyzing the characteristics of different lesions of DR and the algorithms of detection of features in retina and lesions of DR proposed by many experts, this paper presented the works have done using the fundus image with DR form the follow parts:1. A method based on the Image Relative Subtraction technique and the information of optic disc for macular detection was proposed in this paper. Firstly, considering the application of the vessel detection, a fast morphology method was applied to extract the main structure of vessel for later use. Then, with regard to optic disc localization, we located the OD with a morphological approach based on the combination of vessel convergence and intensity. Finally, we detected the macula using the Image Relative Subtraction technique and the distance relationship between macular and optic disc.2. The improved variational level set model applied in narrowband had been demonstrated in this paper. A new, fast and robust methodology for fully automatic segmentation of the optic disc based on the improved narrowband level set model had been proposed in this paper. Circular Hough Transform (CHT) was applied to obtain a circular OD boundary approximation which used as the initial active contour to reduce the number of iteration and got a precise boundary quickly. Finally, the boundary of the OD was extracted by using the improved level set deformable model. This algorithm had been validated on three public databases DRIVE DIARETDB1 and MESSIDOR. Experiments showed that the localization and segmentation methods were accurate and robust even in challenging cases. We have also compared the performance of our proposed method with Otsu, watershed transform, gradient vector flow, snake method, CHT and other methods in literatures. Overall, our results showed superiority of the proposed method3. In this paper, we proposed a coarse-to-fine exudates detection algorithm with the morphological operation based on the intensity characteristic and tested in on the DIARETDB1 database. All these works above facilitate the validity of the automatic system for grading of DME that will assist the ophthalmologists in early detection of the disease.