Research On Diabetic Retinopathy Screening And Exudate Detection Technology

Diabetes is a chronic metabolic disease worldwide,which can cause damage to various organs and trigger all kinds of complications.Diabetic retinopathy is one of the main complications and if it is left untreated,the human eye vision will be affected and maybe at the risk of blindness.Due to the early lesions will not affect the human body,which is not easy to find for patients,but once it is found that patients will often have reached a serious stage,which is not easy to treat.So screening of diabetic retinopathy for diabetes is necessary.At present it is mainly by eye doctors to check for diabetes patients.But due to the lack of ophthalmologists,not all patients would receive effective diagnosis.So it is necessary to use the method based on image processing for detection and classification of fundus images for improving the efficiency and cost savings.A variety of exudate detection methods have been proposed by scholars,which can be roughly divided into two categories,unsupervised methods and supervised ones.Compared with supervised methods,unsupervised methods avoid tedious data labelling and can be further divided into three subcategories,including thresholding methods,morphology methods and clustering methods.These methods have obtained a certain effect on the detection of exudates.In this paper,thresholding and morphology methods to detect exudates are studied.In thresholding detection,super-pixel blocks are introduced.By clustering pixel sets into superpixel blocks,the resistance to noise is improved,and provides a new idea for supervised exudate detection methods.In morphological detection,boundary of exudate can be effectively detected through giving a edge strength value for each pixel rather than a connected area as a whole.To remove reflective in the process of exudate detection,a reflective detection method based on HSV space model is proposed.The method is evaluated at pixel level and we obtain an average sensitivity of 68.01% and an average positive predictive value of 25.53%.Comparison with other experimental results,our method has certain advantages and has the potential to be applied in the screening of diabetic retinopathy.