Detecting Microaneurysms and Exudates in Retinal Images using Deep Neural Networks, and Comparing their Performance with Alternative Neural Network Models

Diabetes frequently results in diabetic retinopathy, which, if unchecked, can result in blindness. Around one in three diabetics, according to the World Diabetes Federation, have diabetic retinopathy in some form. In affluent nations, diabetic retinopathy is the main factor of blindness among working-age individuals. With the progression of diabetes, diabetic retinopathy is more common. Two typical diabetes-related eye abnormalities are microaneurysms and exudates, which are frequently linked to diabetic retinopathy. The severity of the condition and the presence of particular retinal abnormalities are used to grade diabetic retinopathy. Machine learning algorithms are trained to examine retinal images and find particular traits connected to diabetic retinopathy, like exudates and microaneurysms. The machine learning algorithms used here are convolutional neural network(CNN), Weighted Neural Network(WNN), Hybrid Neural Network(HNN) and Deep Neural Network(DNN). Among all the used algorithms DNN shows superior results over others having Sensitivity, specificity, precision, accuracy, and Kappa value are95.74%, 92.31%, 96.77%, 94.74%, and 0.87, respectively.