A new deep learning approach for the retinal hard exudates detection based on superpixel multi-feature extraction and patch-based CNN

Diabetic Retinopathy (DR) is a severe complication of chronic diabetes causing significant visual deterioration and may lead to blindness with delay being treated. Exudative diabetic maculopathy, form macular edema where hard exudates (HE) develop, frequent cause in DR. The detection HE comprises role the DR diagnosis. In this paper, an automatic method based on superpixel multi-feature extraction patch-based deep convolutional neural network proposed. Firstly, superpixels, regarded as candidates, are generated each resized image using segmentation algorithm called Simple Linear Iterative Clustering (SLIC). Then, 25 features extracted from images patches feature. Patches subsequently used train network, which distinguishes background. Experiments conducted three publicly available datasets (DiaretDB1, e-ophtha EX IDRiD) demonstrate that our proposed methodology achieved superior when compared current state-of-art algorithms.