A Review: Diagnosis of Diabetic Retinopathy by using Machine Learning & Deep Learning Techniques

Abstract

This study focuses to detect all the stages of diabetic retinopathy (DR) using end-to-end deep ensemble networks. One of the most prominent and significant micro-vascular ramifications of diabetes mellitus involves diabetic retinopathy (DR). If the treatment is not received for this degenerative condition on time, it damages the retina, visual impairment and blindness may result. It impacts on the affected persons can be lessened by having a thorough understanding of the pathophysiology, on receiving a quick diagnosis, and using efficient management techniques. The complex pathophysiology of diabetic retinopathy (DR) is reviewed in this abstract, which includes a range of biochemical, molecular, and hemodynamic mechanisms brought on by persistent hyperglycaemia. Chronic hyperglycaemia contributes to the onset and progression of diabetic disease DR by causing microvascular changes, oxidative stress, inflammation, and neurodegeneration. Many hand-on engineering and end-to-end learning-based approaches are used to detect the DR using Kaggle dataset. The detection of the mild stage is important for the early control of this fatal disease. This study focuses to detect all the stages of DR using end-to-end deep ensemble networks. The results show that the proposed approach outperforms state-of-the-art methods. To get the finest mass image dataset to train models, it takes Pre-processing steps, like data augmentation will increase the number of training examples, and data normalization will precisely predict classification. So, they could train the latest CNNs model (AlexNet, VggNet, GoogleNet and ResNet) to recognize the slight differences between the image classes for DR Detection. Transfer learning and hyper-parameter tuning methods are adopted and the experimental results have demonstrated the better accuracy than non-transferring learning methodology on DR image classification.