AI and eye disease research has grown rapidly since 2015, bibliometric study finds

A new analysis of the scientific literature on artificial intelligence (AI) in retinal disease research finds that publication output has grown at an annual rate of 36% since 2010, with the sharpest acceleration occurring after 2015, coinciding with the broader adoption of deep learning in medical imaging.

The study, which reviewed 1440 peer-reviewed journal articles indexed in the Web of Science Core Collection between 2010 and 2025, used bibliometric methods to map publication trends, citation patterns, institutional collaborations, and thematic shifts in the field.

“Thematic and factorial analyses uncover a gradual shift from foundational algorithmic studies to multimodal and therapy-focused innovations, with emerging themes including explainable AI, telemedicine, and personalized diagnostics. Despite robust growth, notable gaps persist in real-world clinical integration, regulatory frameworks, and representation from low-resource regions,” study authors, led by Ruixi Zhao from the School of Optometry, The Hong Kong Polytechnic University in Hong Kong, China, stated.

Scale and scope

The dataset spans 187 journals and includes contributions from 6,335 authors, with an average of 7.86 co-authors per article — a figure the authors describe as indicative of high collaboration. International co-authorship accounted for 37.5% of publications. Average citations per document stood at 24.41.

The most prolific journals were Translational Vision Science & Technology with 125 publications, Biomedical Optics Express with 103, and Ophthalmology Science with 87. The most prolific individual authors were Schmidt-Erfurth U and Bogunovic H, with 79 and 75 publications respectively, and H-index scores of 26 each.

Geographic concentration

The United States and China led in publication volume, with both showing sharp output increases after 2020. The United Kingdom, Singapore, Austria, and Germany contributed at lower but steadily growing rates. The University of London produced the most publications among institutions, at 202, followed by the Medical University of Vienna and the National University of Singapore, each with 191.

The authors note that African, Southeast Asian, and Latin American institutions are largely absent from the collaboration networks, which they describe as a potential threat to the generalizability of AI tools across diverse patient populations.

Most-cited work

The most cited paper in the dataset, with 1764 citations, was a study by Gu et al. in IEEE Transactions on Medical Imaging on deep learning for ophthalmic image analysis. The second and third most cited were studies by Abràmoff et al. and Gargeya and Leng, focused on early clinical applications of AI in diabetic retinopathy diagnosis.

Thematic focus

Keyword analysis found that optical coherence tomography, or OCT, was the most frequently appearing term with 546 mentions, followed by deep learning at 367 and artificial intelligence at 246. Disease-specific terms — diabetic retinopathy, macular degeneration, and glaucoma — also appeared prominently.

A thematic evolution map covering 2010 to 2025 shows the field moving from foundational algorithmic work in the early years toward practical applications in image segmentation and disease detection by the mid-period, and more recently toward therapeutic monitoring and clinical integration.

Gaps identified

The authors identify several areas they say remain underrepresented in the literature. Real-world clinical validation and multi-center trials are described as lagging behind model development. Less common retinal conditions — such as retinal vein occlusion and Stargardt disease — receive comparatively little attention. Explainable AI, data privacy frameworks, and algorithmic accountability are noted as topics infrequently addressed despite their relevance to regulatory compliance.

The authors also highlight emerging areas including oculomics — the use of AI analysis of retinal imaging to identify systemic diseases — and radiomics, which involves extracting quantitative features from imaging modalities not detectable through standard clinical observation.

Limitations

The analysis was restricted to English-language peer-reviewed journal articles, excluding conference papers and preprints. The authors note that bibliometric methods assess research trends and collaboration structures but do not evaluate the methodological quality or clinical effectiveness of individual AI models. Recently published studies may also be underrepresented due to citation lag.

Reference:

1. Zhao R, Gillani S. AI in ophthalmology: A bibliometric analysis of retinal imaging innovations and global research collaboration. Photodiagnosis and Photodynamic Therapy. 2026;59:105458. https://doi.org/10.1016/j.pdpdt.2026.105458