Artificial Intelligence in the Diagnosis and Screening of Ocular Diseases: Implications for Clinical Practice and Low-Resource Settings

Abstract

Advances in artificial intelligence (AI), particularly machine learning and deep learning techniques, have introduced new possibilities for addressing these structural challenges in ophthalmology. To synthesize current evidence on the role of artificial intelligence (AI) in the diagnosis of major ocular diseases and to evaluate its clinical performance, integration with teleophthalmology, and relevance for resource-limited health systems, including Bangladesh. This article was developed through a narrative review of the published literature. A structured literature search of PubMed/MEDLINE, Scopus, and Google Scholar was conducted for studies published between January 2015 and January 2026. Reference lists of key articles and selected reports from international organizations were also screened. Original research articles, validation studies, systematic reviews, and implementation reports published in English were included if they evaluated AI-based diagnostic or screening applications for ocular diseases using ophthalmic imaging modalities. Studies lacking sufficient methodological detail or focusing solely on non-diagnostic applications were excluded. Evidence from validation and implementation studies indicates that AI-based systems achieve high diagnostic performance across several high-burden ocular diseases. For diabetic retinopathy, the most extensively studied condition, reported sensitivities typically ranged from 90% to 97%, with specificities between 85% and 95% for detecting referable disease. Promising results have also been reported for glaucoma, age-related macular degeneration, retinal vascular disorders, and selected anterior segment diseases. Increasingly, AI systems are capable of multi-disease detection from a single imaging dataset. Integration with teleophthalmology platforms supports decentralized screening and referral triage, particularly in primary care and community settings. Evidence relevant to Bangladesh suggests the feasibility of AI-assisted diabetic retinopathy screening using locally derived datasets and teleophthalmology models. However, concerns regarding data representativeness, external validity, interpretability, ethical governance, and regulatory oversight remain. AI-based diagnostic systems demonstrate substantial potential to augment ophthalmic care by improving early detection, enhancing efficiency, and supporting scalable screening models. When appropriately validated and integrated into existing referral frameworks, AI may strengthen eye care delivery in resource-limited settings. Continued emphasis on external validation, governance, and equitable implementation is essential to ensure safe and effective deployment.

Bangladesh Med J. 2025 Jan; 54(1): 28-34