Deep Learning Based Diabetic Retinopathy Classification of Augmented Fundus Images Using Convolutional Neural Networks

Diabetic Retinopathy (DR) is the prime cause of blindness among humans with diabetes. It starves the retinal tissues and prevents blood flow since it is a microvascular disease of the ocular retina. Early detection is essential for the effective treatment of visual impairment caused by severe DR. Still, because it necessitates the expertise of ophthalmologists, DR detection is usually expensive and time-consuming. Consequently, the development of automatic detection systems aimed to expedite and lower the expense of the identification procedure worldwide. A potential method for improving the accuracy of Convolutional Neural Networks (CNNs) is augmentation. Scaling, rotation, and contrast adjustment are some of the transformations that are used in data augmentation to find valuable training samples in pre-existing training sets. This study presents a CNN model that addresses the knowledge gap in DR detection by utilizing advanced pre-processing techniques and data augmentation. The data bank was divided into subsets for testing, validation, and training, with the main goals of the research being to obtain high accuracy, [Formula: see text]-score, precision, and recall. The primary goal of the investigation, which is to enhance medical image analysis for clinical applications, is bolstered by these results, which demonstrate deep learning’s potential and data augmentation in improving DR monitoring systems. Retinal images from the APTOS 2019 Kaggle competition for blindness detection Dataset were pre-processed to assess the implementation of the Deep Learning (DL) model. The proposed enhanced method has an accuracy of 91.24% for five classes and 95.4% for binary classes.