Comprehensive Review and Meta-Analysis of Machine Learning Applications in Screening for Diabetic Retinopathy Analysis

The leading cause: diabetic retinopathy global blindness, affects 10% to 24% of individuals with type 1 or type 2 diabetes in primary care. Early detection using deep learning methods is critical for timely intervention and preserving vision. Our comprehensive review, including a meta-analysis, evaluates the effectiveness of these algorithms in DR detection. This study extensively assessed machine learning’s diagnostic accuracy in identifying diabetic retinopathy across diverse cases using color fundus images, aiming to pinpoint the most advanced ML strategy available. We extensively reviewed relevant literature from January 2015 to December 2022, utilizing EMBASE, PubMed, Google Scholar, and Scopus. Following PRISMA guidelines, we focused on machine learning-based study designs. Two authors independently assessed articles for inclusion based on predefined criteria, and data were collected using a standardized form. The meta-analysis reveals strong performance of machine learning in detecting diabetic retinopathy in color fundus photos, indicating readiness for clinical use. However, caution is advised due to methodological limitations in some earlier studies, such as lack of external validation and potential biases in participant selection