Clinical bottom line
Artificial intelligence (AI) has emerged as a transformative tool in diabetic retinopathy (DR) screening, offering high accuracy, scalability, and potential to improve early detection and reduce vision loss. Multiple validated AI algorithms demonstrate sensitivity and specificity comparable to expert graders, enabling efficient triage in diverse clinical settings. Real-world implementation studies confirm feasibility and patient acceptance, though integration challenges remain. Cost-effectiveness analyses support AI-enabled screening as a sustainable approach to address growing screening demands globally (Huang et al., 2022; Ahmed et al., 2026).
What the evidence shows
Huang et al. (2022) provide a comprehensive overview of AI applications in DR diagnosis and screening, highlighting convolutional neural networks’ ability to detect referable DR with high sensitivity and specificity. They emphasize AI’s role in augmenting screening capacity, particularly in underserved areas.
Zafar et al. (2022) review current AI algorithms, noting that many have received regulatory approval and demonstrate robust performance across ethnicities and imaging modalities. They discuss challenges including image quality variability and the need for continuous algorithm refinement.
Grauslund et al. (2022) discuss the evolving landscape of DR screening in the AI era, underscoring that AI can reduce workload for ophthalmologists by automating initial image grading and prioritizing patients needing specialist evaluation.
Ahmed et al. (2026) systematically review real-world AI screening performance, confirming consistent diagnostic accuracy and reporting high patient and provider satisfaction. They also highlight implementation barriers such as infrastructure and data privacy concerns.
Piatti et al. (2022) report preliminary experience from the Italian healthcare system, demonstrating successful AI integration with improved screening coverage and timely referrals.
Amble et al. (2025) provide a systematic review of cost-effectiveness studies, concluding that AI-enabled DR screening reduces costs per case detected and is economically favorable compared to traditional screening models.
Wang et al. (2021) summarize progress in AI for DR screening, emphasizing ongoing advances in multimodal imaging and explainability to enhance clinician trust.
Grzybowski et al. (2022) offer a critical appraisal of AI tools, recommending standardized validation protocols and integration with clinical workflows to maximize benefit.
Caveats and uncertainty
Most AI validation studies focus on referable DR; performance for other diabetic eye diseases is less established.
Variability in image acquisition and patient populations can affect algorithm accuracy.
Ethical and legal considerations around data security and liability remain unresolved.
Long-term impact on clinical outcomes and health system workflows requires further study.
Potential disparities in access to AI technology may exacerbate health inequities if not addressed.
How this may change practice
Ophthalmologists and diabetes care providers should consider incorporating validated AI screening tools to expand capacity and improve early detection of DR.
AI can streamline referral pathways, allowing specialists to focus on patients with sight-threatening disease.
Training and infrastructure development are essential to support AI integration, including image acquisition quality control.
Health systems should evaluate cost-effectiveness and equity implications when adopting AI-enabled screening.
Collaboration between clinicians, AI developers, and regulators will be critical to ensure safe, effective deployment.