Artificial intelligence in oculomics: A comprehensive review of ocular biomarkers for ophthalmic and systemic health

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Oculomics leverages high-dimensional ocular imaging as a non-invasive window into systemic, microvascular, and central nervous system health. The integration of artificial intelligence (AI) and deep learning has revolutionized this field, expanding applications from localized ophthalmic diagnostics to systemic risk stratification and biological aging prediction. This narrative literature review synthesizes current clinical and computational evidence regarding AI-driven oculomics, evaluating diagnostic performance, self-supervised foundation models, and translational hurdles. Evidence demonstrates that supervised deep learning systems achieve specialist-level accuracy across major posterior-segment pathologies—including diabetic retinopathy, glaucoma, age-related macular degeneration, and papilledema (AUCs > 0.93–0.99)—culminating in the first FDA-authorized autonomous diagnostic system in primary care. In neuro-oculomics, AI algorithms decode subclinical retinal signatures to identify Alzheimer’s disease and predict incident Parkinson’s disease through macular inner layer thinning years prior to clinical presentation. Furthermore, the shift toward self-supervised foundation models, such as RETFound, resolves expert label scarcity while offering exceptional label efficiency and cross-population generalizability across cardiovascular and neurodegenerative disease predictions. Despite these advancements, prospective clinical deployment faces core challenges surrounding real-world algorithmic degradation, black-box opacity, data standardization, and legal liability. Overcoming these hurdles through privacy-preserving multimodal omics integration and rigorous prospective trials will be essential to realize the potential of oculomic AI in preventive medicine and narrow the health span–lifespan gap.