Hybrid convolutional neural network for detection of microaneurysms and exudates in retinal images

Abstract

The most common reason for blindness in working-age adults is diabetic retinopathy. The World Health Organization (WHO) estimates that diabetic retinopathy affects about one-third of adults with diabetes. According to the American Diabetes Association (ADA), 4.4 million Americans and 7.7 million Americans, respectively, have diabetic retinopathy that poses a threat to their vision. The development of vision loss brought on by diabetic retinopathy can be stopped or delayed with early detection and treatment. In order to increase performance for the early identification of microaneurysms and exudates, a hybrid neural network (HNN) is a sort of deep learning model that combines convolutional neural networks application on a complete image and an image segmented into 64 parts.In this instance, it is utilized to find exudates and microaneurysms in retinal pictures, both of which are symptoms of diabetic retinopathy. An accurate and effective diagnosis methodfor diabetic retinopathy (DR) detectionis created by training a hybrid convolutional neural network on massive datasets of retinal pictures. Grading diabetic retinopathy is crucial since it aids in figuring out how serious the condition is, informs treatment choices, and tracks the disease's development.The severity of the condition is often determined by the grade of the diabetic retinopathy, which ranges from very mild to severe. HNNs outperformed weighted neural networks (WNN) and convolutional neural networks (CNN) in terms of performance sensitivity, specificity, precision, and accuracy, increasing to 91.91, 87.69, 94.74, and 90.68, respectively.