Diabetic Retinopathy (DR) is a major impediment of eye that is now one of the prominent sources of impaired vision due to long-term diabetes. We can save many people by early diagnosis of DR which is a reliable test that will remind patients with DR to pursue medical treatment in time. Diagnosis is a complicated process and manually fundus images are used to detect DR stages. Various computerized approaches have been proposed also. Various deep learning model showed significant performance in this context. In this study, a novel deep convolutional neural network was developed for detecting the severity stages of DR after some preprocessing of the dataset. As the dataset was imabalanced, we followed downsampling technique. Various hyper parameters performance were also studied. The results of our experiment showed that the proposed model can detect all the grades and overhead the conventional methods.

Keywords : APTOS 2019 BLINDNESS; Diabetic Retinopthy; CNN; Deep Learning