Detection and Classification of Diabetic Retinopathy Using 3D Squeeze Net Architecture

Large number of people is affected by the diabetic retinopathy disease. It results in vision impairment. Deep learning (DL) techniques are widely used nowadays for the screening and research of diabetic retinopathy (DR). Researchers have been utilizing the convolutional neural network (CNN) technique to detect the fundus in retinal images. The application of CNN helps to improving the accuracy to make things automated. Accurate classification of the affected part is a little tedious compared to the existing methods since accurate annotations are expensive. A notable success has been observed when using the squeeze net architecture to detect and classify diabetic retinopathy. A novel 3D squeezenet architecture has been proposed in this paper. Gaussian blurring and random shifting have been used to support the proposed 3D squeezenet method in classifying the affected part. A trustworthy classification can be obtained by reducing the excess fitting in the beginning itself by using data augmentation. This technique has the advantage of having less complexity and cost without compromising the results' accuracy. The findings demonstrate the high efficacy of the proposed methodology in the automated identification of diabetic retinopathy. Consequently, this contributes to a higher degree of prevention of visual impairment.