The review builds on the American Heart Association’s cardiovascular–kidney–metabolic (CKM) framework, which recognizes that metabolic dysfunction, kidney injury and cardiovascular disease form an interconnected continuum rather than isolated conditions. Obesity, insulin resistance and type 2 diabetes set off a cascade of vascular damage that simultaneously attacks the heart, the kidneys and the brain. Yet the authors point out that the brain — and the central nervous system more broadly — remains insufficiently integrated into this paradigm. To close that gap, they adopt an expanded analytical framework they call cardiovascular–kidney–brain–metabolic syndrome (CKBMS). Importantly, they are not proposing a new diagnostic label; rather, CKBMS serves as an operational lens for examining how microvascular dysfunction, endothelial injury and impairment of the neurovascular unit may link damage across all three organ systems in the context of metabolic dysregulation.

Why the retina? The answer lies in its remarkable anatomy. The retina shares key structural and functional features with the heart, kidneys and brain: it hosts a dense microvascular network, relies on delicate barrier architecture (the blood–retinal barrier mirrors the blood–brain barrier), and contains neural tissue that can be imaged directly. In fact, the retina is the only place in the human body where clinicians can non-invasively visualize both microvasculature and neural tissue in vivo. Retinal vessels and cerebral vessels share similar embryological origins, size ranges and autoregulatory mechanisms, making the eye a plausible surrogate for what is happening in the brain’s smallest blood vessels — and, by extension, in the microcirculation of the kidneys and heart.

The technical toolkit for reading these signals has expanded dramatically. Color fundus photography, the workhorse of retinal imaging, captures vascular geometry: arteriovenous ratio, fractal dimension (a measure of vascular branching complexity), and calibers summarized as central retinal arteriolar and venular equivalents. Narrower arterioles, wider venules and simplified branching patterns have all been linked, in large cohort studies, to elevated blood pressure burden, incident cardiovascular events and renal function decline. Optical coherence tomography (OCT) adds a structural dimension, quantifying retinal nerve fiber layer thickness and ganglion cell–inner plexiform layer volume — neural metrics that thin when neurodegeneration strikes. Optical coherence tomography angiography (OCTA), a newer dye-free technique, maps perfusion at the capillary level, quantifying vessel density in distinct vascular plexuses and the size of the foveal avascular zone. In diabetic retinopathy research, OCTA-derived deep capillary plexus measures have emerged as particularly sensitive markers of early microvascular injury.

The evidence connecting these retinal phenotypes to systemic disease is accumulating across multiple organ fronts. Retinal vascular narrowing and venular widening correlate with blood pressure burden and predict major adverse cardiovascular events in population cohorts. In kidney disease, retinal arteriolar damage tracks with estimated glomerular filtration rate decline and albumin-to-creatinine ratio elevation, with diabetic kidney disease showing especially strong retinal–renal concordance. In the brain, retinal vessel changes associate with cerebral small vessel disease — white matter hyperintensities, lacunar infarcts and cerebral microbleeds visible on MRI — and with cognitive impairment, including vascular cognitive impairment and Alzheimer’s disease. Shared mechanisms likely underlie these correlations: chronic hyperglycemia generates advanced glycation end products that stiffen basement membranes, endothelial dysfunction promotes barrier breakdown, and renin–angiotensin system overactivity drives vasoconstriction across all affected beds.

Deep learning is the engine transforming this field from qualitative observation to quantitative risk modeling. Convolutional neural networks now extract vascular topology and subtle imaging features that escape human graders, while multimodal deep learning fuses retinal data with clinical variables to predict multi-organ outcomes. The review highlights systems such as the Singapore

The review’s authors are careful to position retinal imaging as an adjunct rather than a replacement for established clinical assessment. They emphasize that current evidence supports its role as a source of complementary risk phenotypes and an adjunctive stratification tool — not as a stand-alone modality for diagnosis or disease staging, and not as a substitute for specialist cardiovascular, nephrological, or neurological evaluation. This framing matters because retinal findings, while statistically associated with systemic outcomes in population studies, do not yet carry the predictive precision required for individual clinical decisions. Associations reported across cohorts vary in strength, and the review repeatedly stresses the need to weigh evidence quality against methodological limitations before translating research findings into practice.

Those methodological limitations form a substantial part of the critical appraisal. Many of the studies linking retinal metrics to cardiovascular, renal, and cerebral outcomes are cross-sectional or rely on retrospective cohorts, which limits causal inference. Measurement variability poses another challenge: retinal vascular calibers and perfusion parameters can be influenced by image quality, acquisition protocols, and grading software, and different quantitative tools — such as the Singapore “I” Vessel Assessment, Integrative Vessel Analysis, Vascular Assessment and Measurement Platform for Images of the Retina, and QUantitative Analysis of Retinal vessel Topology and size — do not always produce interchangeable values. Confounding is a persistent concern as well, since age, blood pressure, glycemic control, and medication use simultaneously shape both retinal appearance and systemic disease trajectories. The authors also note that much of the evidence base derives from populations of European and Asian ancestry, raising questions about generalizability to other groups.

To move the field forward, the review calls for prospective, multicenter, cross-device studies designed to validate the incremental clinical value of retinal imaging over conventional risk assessment strategies. Such studies would need to demonstrate that adding retinal phenotypes to existing risk models — which already incorporate age, blood pressure, lipid profiles, glycemic status, and renal function — meaningfully improves risk discrimination, typically measured by area under the receiver operating characteristic curve, or reclassifies patients in clinically useful ways. Standardization of image acquisition and analysis pipelines across devices and centers will be essential, since deep learning models trained on one scanner or population often degrade when deployed in new settings.

The appeal of retinal imaging as a population screening tool is nonetheless considerable. Fundus photography is fast, inexpensive, and already widely deployed in diabetic eye screening programs around the world, meaning the infrastructure for large-scale retinal image capture largely exists. OCT and OCTA are increasingly common in ophthalmology clinics, and their outputs — retinal nerve fiber layer thickness, ganglion cell–inner plexiform layer volume, vessel density in the superficial and deep capillary plexuses, and foveal avascular zone area — can be quantified automatically. If deep learning systems can reliably extract systemic risk signals from these routine scans, a single imaging session could, in principle, serve double duty: monitoring eye health while simultaneously contributing to cardiovascular, renal, and cognitive risk stratification. This would be particularly valuable in primary care settings and in regions where specialist access is limited, since metabolic diseases and their complications are rising fastest in low- and middle-income countries.

The mechanistic rationale connecting the eye to the kidney deserves particular attention. Both organs share remarkably similar microvascular architecture: the glomerular filtration barrier of the kidney and the blood–retinal barrier of the eye both depend on fenestrated endothelia, basement membranes, and supporting cells — podocytes in the kidney, Müller cells and pericytes in the retina — that are vulnerable to the same metabolic insults. Advanced glycation end products cross-link basement membrane proteins in both tissues, and hyperglycemia-driven pericyte loss is a hallmark of both diabetic retinopathy and diabetic kidney disease. This structural and biochemical parallelism helps explain why retinal arteriolar narrowing tracks with estimated glomerular filtration rate decline and albumin-to-creatinine ratio elevation, and why the concordance between retinal and renal findings is strongest in diabetic kidney disease. It also suggests that retinal imaging could serve as a repeatable, non-invasive readout of microvascular health in organs that cannot be directly visualized without biopsy or contrast-based imaging.

On the cerebral side, the neurovascular unit concept provides a unifying framework. The neurovascular unit comprises neurons, glial cells, vascular cells, and the extracellular matrix that together maintain the blood–brain barrier and couple blood flow to neural activity. Metabolic dysregulation disrupts this unit through endothelial injury, chronic inflammation — signaled by elevated interleukin-6 and tumor necrosis factor-α — and renin–angiotensin system overactivity. Because retinal vessels share embryological origins, size ranges, and autoregulatory behavior with cerebral vessels, retinal vascular geometry may reflect the state of cerebral small vessels that are themselves inaccessible to direct imaging except through MRI. The observed associations between retinal vessel changes and white matter hyperintensities, lacunar infarcts, and cerebral microbleeds are consistent with this shared-vulnerability model, although the review is careful to note that correlation does not establish that retinal imaging can predict cerebral outcomes at the individual level.

Ultimately, the review charts a measured course between enthusiasm and caution. The convergence of mature imaging infrastructure, quantitative biomarkers, and deep learning analysis tools has created a genuine opportunity to repurpose a routine ophthalmic examination as a systemic health screen. But the authors’ central message is that the science is not yet settled: the field needs rigorous, prospective validation across diverse populations and devices before retinal imaging can claim a defined role in clinical risk management. If those validation efforts succeed, the humble eye examination — long valued for its own sake — may take on an expanded role as a routine, non-invasive checkpoint for the vascular health of the heart, kidneys, and brain.

Subject of Research: The retina as a window to systemic health: non-invasive imaging biomarkers for cardiovascular–kidney–brain risk in metabolic disease

Article Title: The retina as a window to systemic health: non-invasive imaging biomarkers for cardiovascular–kidney–brain risk in metabolic disease

Article References: Ma, X., Yu, Q., Wei, L., Chen, Z., Ma, Z., Lai, T., Zheng, M., Gu, C., & Rao, H. (2026). The retina as a window to systemic health: non-invasive imaging biomarkers for cardiovascular–kidney–brain risk in metabolic disease. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-09058-1

Image Credits: AI Generated

DOI: 10.1186/s12967-026-09058-1

Keywords: retina, window, systemic, health, non-invasive, imaging, biomarkers, cardiovascular, kidney, brain, risk, metabolic