Simultaneous Detection of the Optic Disc and Exudates in Color Retinal Images

The presence of Exudates (EX) is one of the earliest clinical symptoms of Diabetic Retinopathy (DR). Early detection and accurate segmentation of these lesions are crucial for diagnosing and treating DR. However, automatic detection of EX is challenging due to its wide range of shapes, sizes, and brightness levels, which can easily be confused with other lesions. In this paper, a novel method has been proposed for detecting EX in retinal images. The proposed method is based on a three-stage approach that includes pre-processing images, automatic coarse segmentation using a color histogram extraction, and refining inaccurately segmented EX using a mathematical morphology dilatation if the obtained coarse segmentation is suboptimal. The proposed method is evaluated on six publicly available datasets (2,067 images). The evaluation results demonstrated that the proposed method with LUV color space achieved an average sensitivity of 0.942%, specificity of 0.940%, and accuracy of 0.941 %. The proposed method can also diagnose EX in
$\sim 6$
seconds per image.