Altered neurovascular coupling in patients with human immunodeficiency virus-associated asymptomatic neurocognitive...

Human immunodeficiency virus (HIV) infection can lead to HIV-associated neurocognitive disorders (HAND), among which asymptomatic neurocognitive impairment (ANI) represents a critical stage for early intervention. However, neuroimaging biomarkers with high sensitivity and specificity for ANI are lacking. The neurovascular coupling (NVC) characteristic in ANI remains unclear. This study aimed to investigate changes in cerebral blood flow (CBF), functional connectivity strength (FCS), and their coupling in patients with ANI under both resting-state and movie-watching conditions, and to evaluate the discriminative performance of multimodal neuroimaging indicators for ANI.

This study enrolled 75 participants with HIV, including 41 with ANI and 34 who were cognitively normal (CN). All participants underwent multimodal magnetic resonance imaging (MRI), including T1-weighted imaging, arterial spin labeling (ASL), resting-state and movie-watching-state functional MRI (fMRI). CBF, FCS, and CBF-FCS coupling coefficients were calculated. Between-group differences were assessed using independent-samples t-tests, with adjustments for age and years of education, and multiple-comparison correction where applicable. Correlation analyses were conducted to explore their associations with cognitive and clinical indicators. Three machine learning (ML) models [K-Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Machine (SVM)] with leave-one-out cross-validation were constructed to evaluate the classification performance of multimodal neuroimaging metrics for ANI, and SHapley Additive exPlanations (SHAP) were applied to quantify feature importance.

The ANI group exhibited abnormal CBF in multiple brain regions and abnormal FCS in both resting-state and movie-watching-state. At the whole-brain level, the CBF-FCS coupling reversed from weakly positive in the CN participants (resting-state: r=0.0348; movie-watching-state: r=0.0364) to weakly negative in the ANI participants (resting-state: r=-0.0283; movie-watching-state: r=-0.0354), and the coupling coefficients were significantly reduced in the ANI participants compared to the CN participants (resting-state: P=0.004; movie-watching-state: P<0.001). Among the ML models, the full multimodal feature set achieved optimal classification performance [KNN: area under the curve (AUC) =0.957; accuracy =0.890; sensitivity =0.980; specificity =0.790], and the movie-based combination "CBF + movie-FCS + movie CBF-FCS coupling" showed consistently high performance across the models (AUC =0.929-0.962). SHAP indicated that the movie-watching-state NVC contributed the most prominently to the prediction of ANI.

Patients with ANI exhibit abnormal CBF, FCS, and NVC. Compared with the resting-state paradigm, the movie paradigm was more sensitive in detecting neural functional abnormalities. The integration of multimodal neuroimaging indicators showed promising discriminative performance for ANI classification. NVC decoupling may represent a candidate neuroimaging marker of early ANI-related brain alterations and warrants longitudinal validation.