A Critique on Deep Learning Methodologies Employed for the Identification of Diabetic Retinopathy Using Fundus Images

In humans, diabetic retinopathy (DR) is a severe problem that can lead to prevent blindness. In many Computer-Aided Diagnosis (CAD) applications, early DR diagnosis is needed to assess the relevance of fundus image features for detection and classification. Based on the intricacy of the lesion characteristics, it becomes challenging to categorize the extent of DR. The test process requires an efficient identification strategy to find the retina's minor problems. Innovative algorithms for deep learning algorithm used for self-identification could potentially stop negative effects like eyesight loss. A very efficient and reliable system may be developed to help doctors autonomously diagnose DR at the beginning, despite the requirement for expert healthcare resources as a result of many advancements in AI methodologies. In order to categorize non-invasive DR, exudates, hemorrhaging, and microaneurysms, several contemporary structures constructed using DL networks are in-depthly examined in this study. This studyoffers a detailed overview of the use of DL algorithms at different phases of automated DR categorization using fundus pictures. On a number of datasets, effectiveness metrics are compared using DL techniques. This study concludes by summarizing the work that has already been completed to improve the accuracy of DL techniques.