A Dictionary Learning Based Method for Detection of Diabetic Retinopathy in Color Fundus Images

Diabetic retinopathy (DR) is a chronic eye disease characterized by degenerative changes to the retina's blood vessels. In this paper, we present a dictionary learning (DL) based method for automatic detection of DR in digital fundus images. The detection method is according to best atomic representation of fundus images based on learned dictionaries by K-SVD algorithm. However, the learned dictionaries by K-SVD should be able to discriminate the normal and diabetic classes, i.e. discriminative atoms should be designed. For this purpose, the best discriminative atoms are obtained for atomic representation of images in each class. The classification rule is based on the best sparse representation, i.e. the test image is belonged to the class with minimum number of best specific atoms. Our discriminative DL-based method was tested on 30 color fundus images which accuracies of 70% and 90% were obtained for normal and diabetic images, respectively.