Improved Classification of Stages in Diabetic Retinopathy Disease using Deep Learning Algorithm

Abstract

Diabetic Retinopathy (DR) is an ophthalmic condition in which the retinal blood vessels of the eye are repaired. The presence of a large amount of glucose in the blood vessels causes DR, which alters the microvasculature of the retina. The early warning signs of DR aid in the detection of visual loss. In order to anticipate DR, there are numerous processes to go through. Normal, Mild, Moderate, Severe, and Proliferative are the phases. The DR phases are determined by the type of retinal lesions that occur. To detect this deadly condition, the ophthalmologist examines the patient's fundus images. To detect DR phases, computer vision algorithms are presented. These techniques, on the other hand, are unable to encode the complicated Macular Edema characteristic and categorize DR stages with a lower level of accuracy. To encode the macular edema feature and improve classification in all five stages of DR, a ResNet 101 model with a hundred and one deep Convolutional Neural Network (CNN) is given in this study. The training set for analysis is 413 (80%) while the training set for analysis is 103 (20%). The suggested experimental automated approach for DR detection is critical for early identification of DR. The suggested deep learning method outperforms existing algorithms in terms of accuracy. The investigation was carried out using the publicly accessible fundus Indian DR Datasets. The findings demonstrate that the proposed method accurately detects different phases of DR and outperforms existing strategies. ResNet 101 deep CNN is implemented tested, and the accuracy of the method is compared to that of the ResNet 50 algorithm.