Introduction
Retinal vasculometry (RV) offers a non-invasive window into microvascular health and has emerged as a potential biomarker for neurovascular and cognitive decline. While previous studies have explored RV in relation to cognitive function, the distinct contributions of smaller (<50µm) and larger (≥50µm) width retinal vessels remain unclear. This study investigates the associations between cognitive status and artificial intelligence (AI)-derived RV metrics across the spectrum of retinal vessel calibres.
Methods
Composite cognitive scores (G4) were derived, using principal component analysis, for UK Biobank participants who had completed four cognitive tests (“pairs matching”, “reaction time,” “prospective memory,” and “fluid intelligence”). RV features (width, area, tortuosity, and width-variance) were extracted from retinal images using the Quantitative Analysis of Retinal vessel Topology and siZe (QUARTZ) Deep Learning AI algorithm. Multilevel linear regression models estimated associations between RV measures and G4, adjusting for demographic, socioeconomic, and cardiometabolic covariates, with additional exclusion of participants with self-reported cardiovascular events.
Results
Data from 69,040 (78%) participants (mean age 57 years, 54.5% female) with 124,477 retinal images were included. A one standard deviation increase in G4 was associated with wider arterioles (0.11 µm, 95%CI 0.06, 0.17μm), greater arteriolar tortuosity (0.41%, 95%CI 0.405,0.412%), and increased arteriolar area (0.03mm², 95%CI 0.029,0.039mm²). Similar positive associations were found for venular width (0.48μm, 95%CI 0.35,0.61μm) and area (0.06mm², 95%CI 0.05,0.07mm²), particularly in the 50–59 age group. Conversely, venular tortuosity (-0.577%, 95%CI -0.575, -0.579%) and vessel width-variance were inversely associated with G4. Notably, lower cognitive scores were associated with narrower venules and reduced arteriolar tortuosity, across smaller and larger sized vessels. Arteriolar width, however, showed varied results between different sized vessels. Reductions in vessel area with lower G4 were more pronounced in larger vessels, suggesting vessel size-specific patterns of association.
Conclusions
AI-derived RV features are significantly associated with cognitive performance, reinforcing their potential as non-invasive biomarkers of neurovascular health. The observed differential associations across vessel sizes highlight the importance of analysing both smaller and larger retinal vessels in cognitive and neurodegenerative research. These findings support further exploration of RV metrics as early indicators of cognitive decline and potential tools in dementia risk stratification.
扫码关注我们
求助内容:
应助结果提醒方式:
