Answers are grounded in this paper and cite their sources.
Cerebral small vessel disease (CSVD) is a major cause of vascular mild cognitive impairment (MCI), but early recognition of CSVD-related MCI and individuals at risk of its development remains challenging. The retina offers a non-invasive window into cerebral microvasculature and neurodegeneration, and advances in swept-source optical coherence tomography angiography (SS-OCTA) now allow quantitative assessment of retinal vascular, structural, and morphologic alterations in vivo. This study aimed to develop and validate an interpretable machine-learning model that predicts MCI in patients with CSVD based on SS-OCTA-derived retinal biomarkers. We retrospectively enrolled 309 consecutive patients with CSVD, of whom 256 met the inclusion criteria, including 101 patients with MCI and 155 without MCI. An initial feature pool comprising clinical variables and SS-OCTA-derived retinal vascular, structural, and microvascular topological metrics was constructed. A leakage-free feature-selection workflow was implemented within the training cohort. After final refitting on the complete training cohort, the Boruta algorithm retained 22 features, which were subsequently reduced by Akaike information criterion-guided backward stepwise logistic regression to seven key retinal predictors: superficial capillary plexus vessel density in the inner ring, inner superior macular retinal nerve fibre layer thickness, inner nasal ganglion cell complex thickness, foveal ganglion cell–inner plexiform layer thickness, inner superior and inner inferior ganglion cell–inner plexiform layer thickness, and vascular bifurcation number. Logistic regression, decision tree, and extreme gradient boosting (XGBoost) models were trained on 70% of the cohort and evaluated on the remaining 30%. XGBoost achieved the highest numerical area under the receiver operating characteristic curve of 0.908 and showed favorable overall validation performance, including high specificity and precision, a favourable F1-score, good calibration, and greater clinical net benefit on decision-curve analysis. SHapley Additive exPlanations (SHAP) confirmed that inner-ring superficial microvascular rarefaction, inner macular neuroretinal thinning, and reduced vascular bifurcation complexity were strongly associated with increased MCI risk. These findings demonstrate that SS-OCTA-derived retinal biomarkers, integrated through an interpretable machine-learning model, can non-invasively identify CSVD patients at high risk of MCI and may support earlier, targeted intervention pending external validation.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: