Machine Learning/Deep Learning Algorithms & Variability in Grading Improves Early Detection of DR

Abstract

Health condition which mainly influence human retina i.e. cognate by Diabetic Mellitus (DM) is a main thread of Diabetic Retinopathy (DR). As a result of the damage of the retina, it causes vision loss. In accordance with census diabetic individuals who had suffered from diabetics in a long time also have DR issues. As a result, DR has become a critical issue that needs a primary stage screening and assessment in order to prevent vision loss and blindness. Physical diagnosis of the condition is time-consuming and prone to inaccuracy. Furthermore, it is not possible to find an ophthalmologist regardless of location or time. As a result, the need for a highly advanced and computerized intelligent system arises, which can be used to diagnose DR in its early stages. Researchers have proposed a number of Machine Learning (ML) algorithms for the diagnosis of DR for decapods. For determining retinal lesion significantly and for initial stage DR diagnosis various feature extraction and analyzing approaches are recommended. Traditional Machine Learning models, on the other hand, suffer from poor generalization during feature extraction due to limited datasets. Using Deep Learning models, more datasets and high computer processing unit weak generalization problem can be reduced. This study intends to provide a DR overview as well as a brief explanation of previous efforts and current automated methods and improvements, in order to the staring exposure of DR. This paper also discusses the most up-to-date DR lesions as well as the causes and symptoms of DR and focus on how AI/ML approaches helpful in early diagnosis of DR and we have to study more on variability in grading to evaluate the best possible result for screening and improving eye disease mainly caused by diabetics.