Diagnostic Accuracy and Feasibility of Artificial Intelligence Systems for Diabetic Retinopathy Screening: A Systematic Review

Abstract

Background

Diabetic retinopathy (DR) remains a leading cause of preventable blindness worldwide, with screening coverage often hindered by resource limitations and access barriers. Artificial intelligence (AI)-driven screening systems have emerged as promising tools to enhance early DR detection, yet questions remain regarding their diagnostic accuracy, feasibility, and broader health system implications.

Objective

To systematically evaluate the diagnostic performance, real-world feasibility, and stakeholder perspectives surrounding the implementation of AI-based screening systems for diabetic retinopathy.

Methods

A systematic review was conducted according to PRISMA 2020 guidelines. Seven databases (PubMed, MEDLINE, Embase, Scopus, Cochrane Library, IEEE Xplore, and Web of Science) were searched through April 2025 for studies evaluating the diagnostic accuracy and implementation of AI-based DR screening systems. Study selection, data extraction, and risk of bias assessment were independently conducted by the author. Narrative synthesis and thematic analysis were used to report outcomes.

Results

Seventeen studies met inclusion criteria. Most AI systems demonstrated high sensitivity (range: 87–100%) and specificity (range: 76–96%) for referable and vision-threatening DR, with performance comparable to expert human graders. Feasibility studies confirmed successful integration in primary care, teleophthalmology, and national screening programs. Patients generally expressed high satisfaction with AI screening, and providers viewed it as a tool to improve efficiency, though some raised concerns about algorithm transparency. Economic analyses suggested potential cost-effectiveness, particularly in underserved settings. However, successful implementation requires addressing regulatory, infrastructural, and ethical considerations.

Conclusions

AI-based DR screening systems are diagnostically reliable and operationally feasible in diverse healthcare settings. They offer substantial potential to enhance screening coverage and reduce preventable vision loss, especially in resource-constrained regions. Continued evaluation of long-term clinical outcomes, cost-effectiveness, and ethical deployment is essential to guide scalable and equitable integration.