Detection of exudates from digital fundus images using a region-based segmentation technique

The detection and quantification of exudates can contribute to the mass screening of the diabetic retinopathy, the major cause of blindness. In this work, we outline detection techniques of the main retinal structures, namely the blood vessels, optic disk and fovea. A new method for the detection of exudates using adaptive thresholding and classification is proposed in which the retinal structures are used to remove artefacts from exudate detection results. The proposed adaptive thresholding proceeds through two stages: (1) image decomposition into a number of homogeneous sub-images using a region-based segmentation technique, and (2) edge detection using a morphological gradient technique. Classifying the exudates from non-exudates was carried out using rule-based classification. Using a clinician reference (ground truth), the proposed method was validated, in terms of pixels, with overall sensitivity of 93.1%. Speed and performance of the proposed method show that it is more reproducible than the manual method.