Multi-channel Convolutions Neural Network Based Diabetic Retinopathy Detection from Fundus Images

Diabetic Retinopathy (DR) is an eye medical condition usually found in patients suffering from diabetes. The initial stages of DR can either show no symptoms or cause mild vision problems but advanced stages of the disease can cause blindness. The main cause of DR is high blood sugar in diabetic patients that affect the blood vessels supplying blood to the retina and blocks them. The body tries to grow new vessels to supply the retina, however, they usually don't develop properly and can easily leak. DR Detection is an extremely challenging task due to the variation of retina change throughout the disease stages. In this paper, a multi- channel Convolutional Neural Network (CNN) is proposed for DR detection from fundus images of the eyes. The proposed system is tested on a DR Dataset consisting of 35,126 images provided by EyePACS. Experimental results indicate that the accuracy of 97.08% is achieved through the model that outperforms those achieved through other methods in recent studies.