Automated detection and grading of hard exudates from retinal fundus images

Diabetic retinopathy is the major cause of blindness and the appearance of hard exudates is one of its earliest signs. In this study, an automated algorithm to detect and grade the severity of hard exudates is proposed. The detection process is based on top-down image segmentation and local thresholding by a combination of edge detection and region growing. Using features of the fovea and their geometric relations with other retinal structures, a method for the fovea localisation is proposed. Grading of hard exudates was performed using a polar coordinate system centred at the fovea. The results of hard exudate detection process were validated based on clinician hand-labelled data (ground truth) with an overall sensitivity of 93.2%. The superior performance of this technique suggests that it could be used for a computer-aided mass screening of retinal diseases.