Artificial intelligence (AI) has rapidly reshaped the landscape of ophthalmic research and clinical practice, offering unprecedented capabilities for disease detection, quantitative phenotyping, risk prediction and decision support across a broad spectrum of chronic ocular conditions 1,2. In recent years, advances in deep learning, multimodal fusion and large-scale representation learning have driven substantial progress in the diagnosis and screening of retinal diseases such as diabetic retinopathy (DR) 3,4, macular disorders 5 and high myopia-related pathology 6, as well as in glaucoma 3,7,8, cataract, anterior segment abnormalities 9, ocular surface disease Figure . 1 Logical relationships among major AI research directions in chronic ocular diseases. Disease-focused and biomarker-focused studies (Ch. 1-2) feed into a shared infrastructure of quantitative analysis and workflow integration (Ch. 3), which branches into surgical/precision medicine and public health applications (Ch. 45 and jointly informs future unified and clinically deployable AI ecosystems (Ch. 6).In the field of diabetic retinopathy (DR) and related retinal diseases, multiple studies have systematically demonstrated the potential of AI across the entire pipeline from lesion detection through microcirculation assessment to risk stratification. In the OCT angiography (OCTA) domain, Wu et al. used swept-source OCTA (SS-OCTA) and found that the choroidal vascularity index (CVI) in highly myopic patients with diabetes was significantly lower than in patients with DR alone, suggesting that CVI, as a quantitative AI-derived biomarker, may be useful for early risk stratification in high-risk populations. Li et al. examined peripapillary atrophy (PPA) subregions with SS-OCTA and found that γ-zone PPA was associated with a reduced risk of DR.They proposed that "myopia-related posterior pole thinning / microvascular depletion" may exert a structural "protective effect" against DR, providing new imaging evidence for the interaction between myopia and DR. Overall, this body of work indicates that AI in retinal and DR applications has evolved from single-disease, single-modality classifiers toward multi-disease, multimodal and multi-scenario "system-level risk assessors".In glaucoma, Huang et al. constructed a random forest model guided by Boruta feature selection using demographic characteristics, metabolic indicators and biochemical parameters to predict the risk of neovascular glaucoma (NVG) in patients with proliferative diabetic retinopathy (PDR). This study highlights the value of crosssystem modeling that integrates "internal medicine metabolic indicators + ophthalmic imaging/clinical data", and shows that traditional machine learning models remain highly interpretable and practical in scenarios with limited sample sizes.In cataract, Tang et al. built an AI model using slit-lamp retroillumination images to achieve automatic diagnosis and grading, establishing a novel quantitative assessment system that can significantly improve efficiency and consistency in primary care and large-scale screening.Anterior segment and ocular surface diseases are also key components of chronic ocular disease management. Wang et al. analyzed meibomian gland energy curves derived from upper eyelid infrared imaging to quantify structural changes associated with Demodex infestation, and proposed that the derived parameters can serve as early, noninvasive biomarkers for dry eye and chronic ocular surface inflammation, thus providing a new "structure-function" quantitative pathway. In summary, this chapter provides a panoramic view of AI in the diagnosis and screening of chronic ocular diseases, illustrating a cross-disease, multimodal technological progression from retinal diseases to glaucoma, cataract, ocular surface and orbital disorders.The eye is often described as a "window to systemic diseases". Numerous studies have focused on extracting ocular imaging biomarkers related to cardiovascular and metabolic conditions.In coronary artery disease (CAD), Zhou et OCTA, more precisely characterizing hypertension-related changes in retinal vascular morphology and perfusion, and showing that AI has considerable potential as a highthroughput screening tool for hypertension and its complications.In diabetic peripheral neuropathy (DPN), Chen et al. used corneal confocal microscopy (CCM) images to compare transformer-based and CNN-based networks for DPN binary classification, and found that a transformer-based DLA exhibited higher potential for rapid screening. Such work brings "ocular micro-neural structures" into AI pipelines for systemic diabetes management and represents an important component in constructing lifelong disease trajectories.In terms of retinal vasculature and optic nerve head morphology, Bai et al. compared populations with high versus low cerebro-cardiovascular risk and found significant differences in vascular complexity and cup-to-disc area ratio, providing ocular imaging evidence for cardiovascular risk stratification. In a cohort of patients with type 2 diabetes mellitus (T2DM), Chen et al. reported that specific retinal vascular parameters-such as mean branch segment length and vessel density-were significantly associated with mild-to-moderate non-proliferative DR (NPDR), suggesting that these features may serve as preclinical biomarkers of DR-related microvascular abnormalities. Taken together, this chapter extends the focus from local ocular phenotypes to "eyeheart-brain-metabolic" interactions, markedly enhancing the interdisciplinary depth of AI research in chronic ocular diseases. These studies indicate that AI-driven fine-grained segmentation and 3D reconstruction have become essential technical supports for quantifying structural changes in chronic ocular diseases.For image quality control, Li et al. developed the DeepMonitoring system using corneal images acquired via smartphones. The system can both determine whether an image is of low quality and localize the source of the quality issue, guiding operators to retake images when necessary and thus ensuring input quality for mobile ophthalmic AI systems. In real-world applications for chronic ocular diseases, such QA/QC is a prerequisite for stable model deployment.In another study, Li et Overall, this chapter illustrates how AI technologies have evolved from "research-grade models" to "systems that can be embedded into clinical workflows", forming an endto-end path from segmentation and parameter extraction to standardization, quality control and AI agents. Collectively, these studies demonstrate the system-level value of AI in "perioperative management" of chronic ocular diseases, forming a closed loop from precise preoperative assessment and intraoperative risk control to postoperative outcome prediction and functional monitoring. In summary, this chapter illustrates the expansion of AI from a "clinical imaging tool" to an integrated role in "surgical medicine, treatment response and mechanistic discovery", serving as an important conceptual upgrade in the context of chronic ocular disease research.Several systematic reviews and methodological studies provide a high-level perspective that underpins this narrative review.For DR, the systematic review and meta-analysis by Tahir et al. showed that AI-assisted screening achieves higher sensitivity than human graders while maintaining comparable specificity, supporting the use of AI as a reliable alternative or adjunct for DR screening. These review-oriented contributions furnish a robust evidence base for the overarching framework of "Artificial Intelligence Applications in Chronic Ocular Diseases" and empirically support the research directions summarized in this article.Based on the evidence above, future directions for AI in chronic ocular diseases can be summarized as follows. The development of multimodal models capable of simultaneously handling 2D/3D imaging, clinical text, omics data and longitudinal follow-up will be key to achieving a "vertical-horizontal integrated" panoramic view of chronic ocular disease.Current research has begun to explore GPT-like large language models (LLMs) in ). In the future, ophthalmology-focused multimodal foundation models could be developed using largescale, unlabeled ophthalmic images, electronic medical records and omics data for selfsupervised pretraining, followed by adaptation to multiple downstream tasks with limited labeled data. This would enable a new paradigm in which "one foundation model supports multiple tasks and scenarios". Artificial intelligence is rapidly transforming the landscape of chronic ocular disease research and clinical practice. Across diagnostic modeling, ocular biomarker discovery, quantitative analytics, surgical decision-making, public health applications and mechanistic multi-omics studies, the 57 accepted papers in this research topic collectively demonstrate the breadth and maturity of AI-driven innovation in ophthalmology. The evidence synthesized in this editorial illustrates how AI has evolved from single-modality diagnostic tools into integrated, multimodal ecosystems capable of supporting risk stratification, longitudinal monitoring, precision surgical planning and interdisciplinary insights linking ocular and systemic health. Despite these advances, major challenges persist-including limited generalizability across devices and populations, insufficient interpretability, heterogeneous validation standards and the absence of unified regulatory frameworks. Addressing these gaps will require the development of robust, transparent and clinically grounded AI systems, as well as largescale multimodal datasets, specialty-specific foundation models and workflowembedded decision-support pipelines. Looking ahead, the convergence of imaging, clinical data, omics and large language models promises to reshape the management of chronic ocular diseases, paving the way toward unified, equitable and clinically deployable AI platforms that can support lifelong ocular health.
Yang et al. (Tue,) studied this question.