Low-complexity computer-aided diagnosis for diabetic retinopathy

Abstract

Diabetic retinopathy (DR) is a major complication of diabetes that may eventually result in vision loss and blindness. DR is caused by the damage to the blood vessels in the retina as high blood sugar levels block minute blood vessels that supply blood to the retina. Recently, convolutional neural networks (CNN) have been introduced for analyzing fundus images and have proven their superiority in detection and classification tasks. This chapter surveys various deep learning approaches for fundus imaging and present a newly introduced low complexity CNN to detect and classify the four stages of DR (i.e., normal retinas, NPDR, severe NPDR, and PDR) based on color fundus images without the need of prior image processing or data augmentation techniques providing the tools for ophthalmologists to better assess the disease and monitor its progression.

Keywords Diabetic retinopathy (DR), Vision loss, Prior image processing