Impact of artificial intelligence in vision science: A systematic review of progress, emerging trends, data domain quantification, and critical gaps.

Abstract

The prominence of artificial intelligence (AI) is growing exponentially, yet its implementation across research domains is uneven. To quantify AI trends in vision science, we evaluated over 100,000 PubMed article metadata spanning 35 years. Using Medical Subject Headings (MeSH) terms, we analyzed trends across four prominent ocular diseases: age-related macular degeneration, diabetic retinopathy, glaucoma, and cataract. Most articles utilized research techniques from at least one of the following domains: biological models, molecular profiling, image-based analysis, and clinical outcomes. Our quantification reveals that AI prominence is disproportionally concentrated in the image-based analysis domain, and, additionally, among 4 diseases evaluated, AI prevalence in cataract research is lagging. Contributing factors towards these disparities include insufficient data standardization, complex data structures, limited data availability, unresolved ethical concerns, and not gaining meaningful improvements over human-based interpretations. By mapping where AI thrives and where it lags, we offer a quantitative reference for funding agencies, clinicians, and vision scientists. Connecting various research domains with multimodal and generative AI could improve diagnostic utility; enabling earlier diagnosis, personalized therapy, reduced healthcare costs, and accelerate innovation. Future work should move AI in vision science beyond image-centric pattern recognition toward integrative, mechanistic analyses that explain - rather than merely detect - disease.