In 2018, researchers at Google showed that a deep learning model could infer age, blood pressure, smoking status and cardiovascular event risk from a single fundus photograph. The result was striking for what it confirmed rather than what it revealed. Ophthalmologists had long suspected that the eye is a window onto systemic vascular biology. The study suggested that we had barely begun to read it, and that most of the signal was still waiting to be extracted.

Fundus photography captures only the larger arterioles and venules. The capillary beds where microvascular disease begins stayed largely invisible without fluorescein angiography, an invasive dye-based procedure rarely justified outside eye disease. Optical coherence tomography angiography (OCTA) changed that. In seconds, without injection, it produces depth-resolved maps of retinal capillary networks at micrometre-scale resolution. That turns a structure we once glimpsed occasionally into one we can measure routinely and repeatedly.

This matters because the retina is the only place in the body where we can directly see a living microcirculation. That circulation shares its embryological origin, autoregulatory behaviour and barrier properties with the brain. This article examines how OCTA works and what its quantitative signals correlate with across diabetes, hypertension, coronary disease and cognitive decline. It also asks where the evidence currently supports its use as a systemic screening instrument, and where it does not yet.

OCTA builds on conventional OCT, which uses low-coherence near-infrared interferometry to produce cross-sectional images of retinal tissue. The angiographic step rests on a simple insight. Static tissue reflects light consistently from one moment to the next, while moving erythrocytes do not. The device acquires repeated B-scans at the same location within milliseconds and computes the variation between them. Algorithms based on amplitude decorrelation, phase variance or split-spectrum approaches such as SSADA then separate flow signal from structural signal.

Because OCT data are volumetric, the resulting flow map can be segmented by depth. Clinicians typically examine four layers:
• The superficial capillary plexus, within the ganglion cell layer
• The deep capillary plexus, at the boundary of the inner nuclear and outer plexiform layers
• The radial peripapillary capillaries around the optic disc
• The choriocapillaris beneath the retinal pigment epithelium
Each plexus has its own haemodynamics and vulnerabilities. Systemic diseases do not affect them uniformly, and that detail becomes central to interpretation.

Quantification is what turns images into biomarkers. Vessel density expresses the proportion of a region occupied by perfused vessels. Skeletonised density removes the influence of vessel calibre so that capillary count can be assessed independently. The foveal avascular zone (FAZ) area and circularity index describe the capillary-free centre of the macula. Fractal dimension captures branching complexity, and flow deficit metrics quantify voids in the choriocapillaris. Together, these provide a multidimensional readout of microvascular architecture rather than a single number.

The technology carries real constraints. OCTA detects flow only above a slowest detectable velocity, so severely slowed capillaries may appear absent rather than merely impaired. Projection artefacts from superficial vessels contaminate deeper slabs. Motion, media opacity and low signal strength systematically depress density values. Segmentation algorithms often struggle in diseased retinas, where layer boundaries are distorted. Axial length also alters image magnification and requires correction before comparing individuals.

The most consequential limitation for systemic applications is that measurements are not interchangeable across manufacturers. Scan dimensions, flow algorithms and proprietary metric definitions all differ. Normative databases also remain limited in ethnic and age diversity. A vessel density value from one platform cannot simply be compared with the same number from another. Precision medicine depends on precise measurement, and OCTA's standardisation problem remains unsolved. Any systemic interpretation has to account for which device generated the data and under what conditions.

Takeaway

OCTA does not image blood vessels; it images the movement of blood within them. That distinction explains both its sensitivity to early microvascular dysfunction and its blind spots when flow slows below detection thresholds.

Diabetes offers the strongest evidence base. Multiple studies show that people with diabetes but no clinically visible retinopathy already have reduced deep plexus density, an enlarged FAZ and lower circularity than non-diabetic controls. These preclinical changes suggest that capillary dropout comes before the microaneurysms and haemorrhages that define conventional grading. Longitudinal cohorts have also linked lower baseline density to later retinopathy progression. This positions OCTA as a potential tool for stratifying risk before structural damage becomes irreversible.

In hypertension, reduced superficial and deep vessel density correlates with blood pressure load and disease duration. Several studies link these deficits to markers of target-organ damage, including left ventricular hypertrophy and declining estimated glomerular filtration rate. This fits the underlying biology. Chronic pressure elevation drives arteriolar remodelling and capillary rarefaction across many vascular beds at once. The retina may therefore serve as an accessible proxy for microvascular injury happening in parallel in the heart and kidney.

Coronary artery disease presents a more nuanced picture. Patients with angiographically confirmed disease tend to show lower retinal vessel density than controls, and some cohorts report graded associations with severity. Coronary microvascular dysfunction is of particular interest, because conventional angiography cannot visualise it directly. Whether retinal capillary metrics track coronary flow reserve is an area of active investigation. If they do, OCTA could offer an indirect window into ischaemia with non-obstructive coronary arteries, which is notably underdiagnosed in women.

The neurological correlations draw on shared biology. The retina develops from the diencephalon, and its neurovascular unit closely mirrors that of the brain. Cross-sectional studies in Alzheimer's disease and mild cognitive impairment have reported reduced vessel density and altered FAZ geometry. Some have linked retinal findings to amyloid PET status or to cerebral small vessel disease on MRI. Results remain heterogeneous, however, with conflicting findings on whether the FAZ enlarges or contracts and which plexus is most affected.

That heterogeneity is instructive. Retinal microvascular rarefaction is a final common pathway shared by ageing, hyperglycaemia, hypertension, smoking, dyslipidaemia and chronic inflammation. As a result, any single metric has limited disease specificity. Most associations are cross-sectional and modest in effect size, and comorbidities tend to cluster in the same patients. The more defensible interpretation is that OCTA quantifies global microvascular health as a systemic phenotype. It does not reliably diagnose individual conditions in isolation, and should not be treated as if it did.

Takeaway

Retinal capillary loss is better understood as a readout of cumulative microvascular burden than as a disease-specific signature. Its clinical value lies in integration with other data, not in diagnosing any single condition alone.

OCTA's appeal for population screening rests on how easy it is to run. Acquisition takes seconds and requires no contrast agent and no ionising radiation, and pupil dilation is often unnecessary. The output is quantitative and lends itself to automated analysis. Compared with coronary calcium scoring, carotid ultrasound or cerebral MRI, it is faster, cheaper per scan and easily repeated. In principle, a routine eye examination could become a cardiometabolic and neurovascular check, delivered in optometry practices, pharmacies or primary care.

Artificial intelligence multiplies that potential. Deep learning has already shown that fundus photographs encode cardiovascular risk factors, retinal biological age and incident disease risk, a field now called oculomics. Foundation models trained on more than a million retinal images extend this to multiple systemic outcomes. OCTA adds depth-resolved capillary information that fundus imaging cannot capture. Early models combining OCTA features with clinical variables suggest some improvement in discrimination over conventional risk scores, though external validation remains sparse.

Monitoring disease progression may prove more useful in the near term than one-off screening. Because each patient serves as their own baseline, serial OCTA can detect change within an individual despite wide variability between individuals. Tracking capillary density over time could show whether glycaemic control, antihypertensive therapy or lifestyle change is preserving microvascular integrity. HbA1c and office blood pressure measure exposures. OCTA measures tissue consequences, offering a complementary end-organ readout.

Several barriers stand between promise and practice. There are not yet prospective outcome studies showing that OCTA-guided management reduces cardiovascular events or dementia incidence. Device cost and limited availability outside ophthalmology restrict reach. Variability between devices complicates pooling data across centres. Normative ranges must also account for age, sex, ethnicity, axial length and signal quality. Without these foundations, screening risks generating anxiety, incidental findings and downstream testing without clear clinical benefit.

The pragmatic path likely involves targeted rather than universal deployment. Sensible starting points include:
• People with diabetes who already undergo regular retinal screening
• Hypertensive patients who need target-organ assessment
• Individuals at intermediate cardiovascular risk, where reclassification would genuinely change management
The next steps are standardised acquisition protocols, vendor-neutral metrics and prospective trials with hard endpoints. The eye already holds the information. Healthcare systems still need to demonstrate that acting on it improves outcomes.

Takeaway

A screening test earns its place not by detecting abnormality but by changing outcomes. OCTA's measurement capability currently runs ahead of the evidence showing what clinicians should do with what it finds.

OCTA has turned the retina into a quantifiable microvascular organ. Depth-resolved vessel density, FAZ geometry and fractal metrics now give a non-invasive, repeatable readout of capillary health. That readout correlates, with varying strength, with diabetes, hypertension, coronary disease and cognitive decline.

For precision care, the most defensible approach today treats OCTA metrics as one layer within a multimodal profile. That profile would also include genetic risk, circulating biomarkers, blood pressure trajectories and imaging of other vascular beds. Serial measurement on a consistent device, judged against each patient's own baseline rather than population thresholds, gets the most signal from a technology that is not yet standardised.

The deeper shift is conceptual. Microvascular disease has long been inferred from its downstream consequences. OCTA allows it to be observed directly, early and repeatedly. Whether that visibility leads to better outcomes depends on the prospective trials still to come.