2 - Early diagnosis of diabetic retinopathy using deep learning techniques

Abstract

A major consequence of diabetes mellitus, diabetic retinopathy (DR) affects vision by causing lesions mostly on the retina that might lead to loss of vision. It may lead to blindness if it is not diagnosed early enough. Because DR is irreversible, therapy is only able to maintain eyesight. Early diagnosis and treatment of DR may considerably reduce the risk of visual loss. The traditional analysis of DR retina fundus pictures by ophthalmologists is cost-intensive, time consuming, and liable to misdiagnosis, while machine diagnostic technologies are less susceptible to these issues. There have been several approaches suggested in the literature for identifying red lesions, all of which rely on the classification of lesion candidates into true or false positives based on handcrafted attributes. Due to the significant cost of manually labeling the lesions, the application of deep learning techniques is scarce in this field. Leveraging retinal images for training the deep neural network design and achieving high precision, deep convolutional neural networks and residual neural networks are tested in this research work for image recognition and classification. In this chapter, the issues of alternative methodologies and current procedures are also explored. The efficacy of the deep learning-based model for DR classification is tested on the APTOS 2019 blindness detection dataset available on Kaggle. The proposed deep neural networkarchitecture outperforms numerous state-of-the-art comparative models, thus ensuring the supremacy of the proposed framework.