Retinal and Choroidal Vascular Metrics Predict Cerebral Atrophy and Cognitive Impairment in Noninfarcted Large Artery...

Large artery stenosis (LAS) can drive cognitive impairment and neurodegeneration even without overt infarction, yet scalable biomarkers for capturing this silent vascular brain injury are limited. We investigated whether retinal and choroidal microvascular and neurostructural measures reflect cerebral atrophy and cognitive impairment in noninfarcted LAS.

In a multicenter cohort of 1239 participants (1035 patients with noninfarcted LAS and 204 community-based controls), we integrated optical coherence tomography/angiography (OCT/OCTA), brain magnetic resonance imaging volumetry, and cognitive assessments. Machine learning models were trained internally (n = 891) and externally validated (n = 348) to evaluate the predictive value of combined retinal and choroidal features for cerebral atrophy and cognitive impairment.

LAS was associated with ganglion cell-inner plexiform layer (GCIPL) thinning and profound retinal and choroidal microvascular rarefaction, which tracked closely with reduced gray matter (GM) and white matter (WM) volumes (all P < 0.001). These retina-brain associations were strongest in bilateral subcortical and medial temporal regions (all P < 0.05). Crucially, integrating OCT/OCTA metrics into machine learning models substantially improved prediction of cerebral atrophy and cognitive impairment beyond the enhanced clinical baseline model: random forest achieved an external test R2 of 0.456 versus 0.254 for GM volume and 0.430 versus 0.251 for WM volume, while the support vector machine achieved the best classification performance for cognitive impairment (area under the curve = 0.794 vs. 0.668).

Retinal and choroidal microvascular measures are sensitive, noninvasive biomarkers of silent, vascular-driven neurodegeneration. They offer substantial incremental value for individual risk stratification in vascular contributions to cognitive impairment and dementia.