Identification of exudates using an image processing algorithm designed for early diagnosis of diabetic retinopathy

Abstract: One of the most primitive and common symptoms of diseases leading to loss of sight among individuals are exudates. Diabetic retinopathy is considered as a key root of blindness, particularly among working-age adults. The extent of retinopathy is extremely correlated with the exudates. Definite areas of the retina in eye with prevalent conditions are to be photocoagulated by laser techniques to impede the disease progress and avert blindness. To summarize the areas relies on outlining the lesions and the anatomical composition of the retina. In this dissertation work, we endow with a new biomedical feature extraction technique for blood vessels that encourages the exudates detection in funds images useful for diagnosis of retinopathy. The technique proceeds with an edge detection algorithm which yields segmented image as an input. Then the novel feature-based algorithm has been employed to detect the features of blood vessels properly. This algorithm is regarded as the feature extraction key for a retinal blood vessel and considers its width range, intensities as well orientations for the function of selective segmentation. Because of its bulb- shape and its colour resemblance with exudates, the optic disc can be extracted by means of the common Hough transform technique. The extracted blood vessel tree and optic disc could be excluded from the over segmented image to acquire an initial estimate of exudates for primary diagnosis. The eventual estimation of exudates can then be achieved by morphological reconstruction based on the appearance of exudates in retinal images.

Keywords: Diabetic retinopathy, blood vessels extraction, Hough transforms, exudates detection