Altered neurovascular coupling in patients with human immunodeficiency virus-associated asymptomatic neurocognitive impairment: a multimodal magnetic resonance imaging study

Introduction

Human immunodeficiency virus (HIV) infection remains a global health challenge and significantly affects the central nervous system (CNS). HIV enters the CNS via infected immune cells, triggering chronic neuroinflammation and damage to neurons and glial cells (1,2). These pathological processes may lead to HIV-associated neurocognitive disorders (HAND). HAND not only affects patients’ quality of life but also reduces their treatment adherence. In recent years, with the widespread use of combined antiretroviral therapy (cART), the incidence of severe HAND has decreased; however, mild HAND remains prevalent due to factors such as persistent neuroinflammation, viral neurotoxicity, and the inability to eradicate viral reservoirs (3,4).

According to the Frascati criteria, HAND is classified into asymptomatic neurocognitive impairment (ANI), mild neurocognitive disorder, and HIV-associated dementia (5). Among these, ANI is of particular significance, as studies have shown that patients with ANI at baseline have a 2–6 times higher risk of progressing to symptomatic HAND compared to those who are cognitively normal (CN) at baseline (6). The presence of HAND is associated with higher mortality, and the ANI stage represents a critical window for reversing neurocognitive decline. Thus, the early identification of ANI is essential for timely intervention and improved prognosis (2,7).

Although patients with ANI generally have preserved activities of daily living (ADL), they often exhibit measurable abnormalities on neuropsychological testing. Currently, the ANI diagnosis relies primarily on neuropsychological testing; however, neuroimaging biomarkers that can sensitively and specifically detect ANI and complement neuropsychological assessments are lacking. Accumulating evidence suggests that people living with HIV (PLWH) may still exhibit subtle brain injury, even in the absence of clinical symptoms or neurocognitive test abnormalities (8,9). Therefore, more sensitive methods are needed to evaluate potential neurobiological alterations.

In recent years, multimodal neuroimaging techniques have provided more sensitive means. Among these, functional magnetic resonance imaging (fMRI) can monitor alterations in brain activity and functional connectivity (FC) during both task and resting-state conditions. Previous fMRI studies have reported reduced resting-state FC in PLWH/HAND patients, along with abnormal brain activity across multiple regions (10-12), and machine learning (ML) based resting-state fMRI (rs-fMRI) classifiers have also been explored for early-stage HAND identification (13). Arterial spin labeling (ASL) MRI uses endogenous arterial blood as a tracer to rapidly and non-invasively quantify cerebral blood flow (CBF). CBF is closely associated with cerebral metabolic activity and is altered in various neurocognitive disorders (14,15). Studies have reported that PLWH exhibit regional cerebral perfusion changes (16,17). However, the reported associations between CBF and HAND have been highly inconsistent (18).

Recently, neurovascular coupling (NVC) has been proposed, emphasizing the balance between neural activity and blood supply, which serves as the foundation for normal brain function (19-23). The coupling between CBF and FC strength (FCS) reflects the relationship between cerebral perfusion and functional integration. While CBF-FCS decoupling has been observed in neurodegenerative disorders such as Parkinson’s and Alzheimer’s diseases (24,25), it remains unclear whether similar abnormalities exist in PLWH. Further, rs-fMRI may not adequately capture complex neural activities associated with real-life experiences. Conversely, movie-watching state fMRI (movie-fMRI) employs naturalistic stimuli and has been shown to better simulate the dynamic nature of daily cognitive processes (26). This naturalistic paradigm has demonstrated higher biomarker sensitivity in various neuropsychiatric disorders (27-29), but it has not yet been applied to HAND.

This study integrated multimodal MRI techniques, including ASL, rs-fMRI, and movie-fMRI, to screen for and identify candidate imaging biomarkers of ANI. For the first time, it systematically compared differences in FCS between resting and naturalistic stimulation states, as well as alterations in the CBF-FCS coupling under these two states, in the ANI population. Using multimodal indicators, ML models were developed and interpreted using SHapley Additive exPlanations (SHAP) to evaluate their discriminative utility for ANI. The aim was to characterize multimodal imaging alterations associated with ANI and to assess whether a parsimonious combination of imaging features can discriminate ANI from CN, thereby informing future biomarker validation and clinically feasible protocol development. We present this article in accordance with the STROBE reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-aw-2110/rc).

Methods

Participants

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Beijing Youan Hospital (No. LL-2023-070-K), and informed consent was obtained from all individual participants. Seventy-eight PLWH were recruited from the STD/AIDS Clinic of Beijing Youan Hospital. The inclusion criteria were as follows: (I) age 20–45 years; (II) male; (III) right-handed; (IV) no MRI-related contraindications; and (V) visual and auditory functions within normal limits. These criteria aligned with the demographic profile of PLWH in China, where young and middle-aged adult PLWH are predominantly male (30-32). Further, limiting the age to 20–45 years reduced the influence of age-related cerebrovascular and neurodegenerative changes on the CBF and connectivity measures. The exclusion criteria were as follows: (I) a history of substance abuse (alcohol/drugs); (II) any neurological condition, such as stroke, traumatic brain injury, or intracranial tumors; (III) any current or prior psychiatric disorder; and/or (IV) missing MRI data or MRI scans of insufficient quality for analysis. HIV positivity was verified using enzyme-linked immunosorbent assay and Western blot. Three patients were excluded due to poor-quality MRI; thus, ultimately 75 patients were included in the study. All patients had received stable cART for at least 6 months. Demographic and clinical data were collected from participant self-reports and electronic medical records, including age, education, time since HIV diagnosis, antiretroviral treatment duration, nadir and current CD4⁺ T-cell counts, CD4⁺/CD8⁺ ratio, and plasma viral load.

Neurocognitive tests

The participants completed neurocognitive tests. All tests were administered in Mandarin using validated simplified-Chinese versions consistent with Chinese normative studies (33). Neurocognitive testing covered six domains: verbal/language (Animal Verbal Fluency), attention/working memory (Wechsler Memory Scale-III and Paced Auditory Serial Addition Test), learning and memory (Hopkins Verbal Learning Test and Brief Visuospatial Memory Test, respectively), information-processing speed (Trail Making Test Part A), fine motor skills (Grooved Pegboard Test), and abstraction/executive function (Wisconsin Card Sorting Test-64). Raw scores were converted to T-scores [mean =50, standard deviation (SD) =10] using validated Chinese normative data (33). The normative conversion procedure incorporated demographic adjustments for age, sex, years of education, and the population size of the city in which the participants grew up. If a cognitive domain included multiple cognitive tests, the average T-score was calculated.

According to the Frascati criteria (5), the diagnostic criteria for ANI are as follows: (I) scores more than one SD below the normative scores in at least two cognitive domains; (II) no decline in daily living abilities; (III) impairment not meeting criteria for delirium or dementia; and (IV) no evidence that ANI was caused by other factors. ADL were assessed using the Chinese version of the ADL scale based on the Lawton-Brody framework (34). The ADL scale was completed as a self-report questionnaire under staff supervision. The scale comprises 20 items rated on a 4-point scale (1= independent, 2= some difficulty, 3= needs help, 4= unable to perform), covering basic self-care (e.g., eating, dressing, toileting, and mobility) and instrumental activities (e.g., shopping, telephone use, medication management, and financial management). In this study, an ADL total score of 20 (i.e., all items scored as 1) was defined as no decline in ADL. Of the patients, 41 were diagnosed with ANI and 34 were assessed as CN. The participant inclusion and exclusion flowchart is shown in Figure 1.

Figure 1 Flow diagram of patient inclusion. ANI, asymptomatic neurocognitive impairment; ASL, arterial spin labeling; CN, cognitively normal; fMRI, functional magnetic resonance imaging; HIV, human immunodeficiency virus; movie-fMRI, movie-watching state fMRI; PLWH, people living with HIV; rs-fMRI, resting-state fMRI; T1WI, T1-weighted imaging.

MRI data acquisition

MRI scans were obtained on a 3.0-T system (GE SIGNA Pioneer, USA) with a 32-channel head coil. High-resolution T1-weighted structural images were collected using a magnetization-prepared rapid gradient-echo sequence with the following parameters: repetition time (TR) =7.8 ms, echo time (TE) =3.2 ms, acquisition matrix =256×256, flip angle (FA) =8°, and voxel resolution =1 mm × 1 mm × 1 mm, number of slices =188, and slice thickness =1 mm. Perfusion images were acquired using a background-suppressed three-dimensional pseudo-continuous ASL sequence: TR =7,499 ms, TE =11.7 ms, acquisition matrix =128×128, voxel size =3.75 mm × 3.75 mm × 4 mm, FA =111°, number of slices =36, slice thickness =4 mm, and post-labeling delay =2,000 ms. Functional images were acquired with a gradient-echo single-shot echo-planar imaging sequence using the following settings: TR =2,000 ms, TE =30 ms, matrix =64×64, voxel size =3.5 mm × 3.5 mm × 3.5 mm, FA =90°, 36 slices, and slice thickness =3.5 mm. The rs-fMRI was performed before the movie-watching fMRI. The resting-state and movie-watching scans each had an acquisition time of 10 minutes. During movie-fMRI scanning, the participants watched the opening family scene of the movie titled “Eat Drink Man Woman” for 10 minutes. The movie stimulus was delivered using Psychtoolbox in MATLAB 2022b (MathWorks, Natick, MA, USA) through an MRI-compatible audiovisual presentation system.

MRI data processing

ASL data analysis

After acquiring the ASL data, the CBF images were automatically generated by the GE post-processing platform. The CBF images were spatially normalized to Montreal Neurological Institute (MNI) space in SPM12 using the following procedure: (I) each CBF image was co-registered with its corresponding T1-weighted structural image; (II) all co-registered brain maps from step (I) were normalized to a standard T1-template in normalized space; (III) using the normalization parameters derived from step (II), all CBF images were normalized into MNI space and resampled to a voxel size of 3 mm × 3 mm × 3 mm. To standardize the data, a z-transformation was applied to the CBF value; (IV) the z-scored CBF (zCBF) maps were smoothed with a 6-mm full width at half maximum (FWHM) Gaussian kernel.

fMRI data processing

fMRI preprocessing

The fMRI data underwent preprocessing in SPM12. The first 10 volumes were discarded to allow for signal stabilization. Slice-timing correction was then performed using the middle slice as the reference. Next, images were motion-corrected by realignment, and participants with excessive motion (>2 mm translation or >2° rotation) were excluded. The motion-corrected images were subsequently normalized to the MNI template and resampled to 3 mm × 3 mm × 3 mm voxels. Finally, the normalized data were spatially smoothed with an isotropic Gaussian kernel (6 mm FWHM).

FCS analysis

FCS in the whole-brain gray matter (GM) was defined as the average FCS between a given voxel and all other voxels, with GM segmented from each participant’s T1 image, registered from individual space to normalized space, and constrained by a T1-template GM mask. Pearson correlation coefficients were calculated between the blood oxygen level-dependent (BOLD) time series of each voxel and those of other voxels. Only positive correlations exceeding 0.2 were retained to exclude weak associations that may reflect background noise (35), thereby obtaining a complete GM FC matrix for each participant. For standardization, the z-transformation was applied to the FCS values. The FCS maps were spatially smoothed with an isotropic Gaussian kernel (6 mm FWHM).

CBF-FCS coupling analysis

To analyze the association between CBF and FCS, a voxel-wise correlation analysis across the whole-brain GM was performed for each participant. Because spatial preprocessing (e.g., co-registration and smoothing) increases dependence between neighboring voxels, the degrees of freedom for across-voxel correlation are effectively much lower than the voxel count within the GM mask. Accordingly, we estimated the effective degrees of freedom (df_eff) for the across-voxel correlation using the following equation:

dfeff=N(FWHMx×FWHMy×FWHMz)/v−2[1]

where v represents the voxel volume (3 mm × 3 mm × 3 mm), N represents the number of voxels included in the analysis (N=67,541), and FWHM_x, FWHM_y, and FWHM_z represent the mean smoothness of the CBF and FCS maps (11.4 mm × 11.9 mm × 12.3 mm), estimated with Data Processing & Analysis for Brain Imaging (version 8.2) (36). In the present study, the df_eff for the across-voxel correlation was 1,096. Thus, a single CBF-FCS correlation coefficient was obtained for each participant, indexing the whole GM spatial concordance between CBF and FCS. For anatomical labeling, brain regions were determined using the Automated Anatomical Labeling atlas (AAL116) in MNI space.

Statistical analysis

The statistical analyses of demographic data, clinical variables, neurocognitive test results, and CBF-FCS coupling coefficients were performed using IBM SPSS 27.0 software (IBM Corp., Armonk, NY, USA). First, the normality of all continuous variables was assessed using the Shapiro-Wilk test. Group differences were analyzed using independent samples t-tests (for normally distributed data) or Mann-Whitney U tests (for non-normally distributed data). Voxel-wise between-group comparisons of the zCBF and FCS maps were performed in SPM12 using a general linear model (the independent samples t-test) with age and years of education as covariates; multiple comparisons were controlled using false discovery rate (FDR) correction for zCBF maps, and Gaussian random field (GRF) correction for FCS maps (corrected P<0.05; minimum cluster size ≥20 voxels). A Pearson correlation analysis was conducted to investigate the relationships between the above imaging features with significant intergroup differences and the clinical variables and neurocognitive test scores. The significance level for all tests was set at P<0.05 (two-tailed), with FDR correction applied for multiple comparisons. As a sensitivity analysis, potential outliers were identified using the 1.5× interquartile range (IQR) rule, and key group comparisons (e.g., coupling coefficients) were repeated after excluding outliers.

ML model construction and performance evaluation

To distinguish between the CN and ANI patients, this study constructed K-Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Machine (SVM) models, all using the leave-one-out cross-validation strategy. Features with significant intergroup differences identified in the above analyses were included in the MLs. Five feature-combination categories were designed (as detailed in Table 1).

The models were evaluated using multiple binary classification performance metrics, including the area under the curve (AUC) of the receiver operating characteristic (ROC) curve, accuracy, precision, sensitivity, specificity, and F1-score, to quantify classification performance under different feature combinations.

Feature importance analysis

SHAP was performed on the optimal feature combinations to quantify the contribution of each feature to the classification decision. Positive SHAP values indicated a positive contribution to the classification of ANI, while negative SHAP values indicated a negative influence.

Results

Demographic, clinical, and neuropsychological data

There were no significant differences between the two groups in terms of age or years of education (both P>0.05). Statistically significant differences were observed between the two groups in relation to the nadir CD4⁺ T-cell count (P=0.011) and CD4⁺/CD8⁺ ratio (P=0.047), but not in relation to the HIV infection duration, treatment duration, or current CD4⁺ count (all P>0.05). All the patients had undetectable current plasma viral load. There were no significant differences between the two groups in terms of fine motor skills, but significant differences were observed across the other five cognitive domains (all P<0.05). The results are presented in Table 2.

CBF changes in ANI

Compared to the CN group, the ANI group exhibited widespread reductions in CBF, including in the left middle frontal gyrus (MFG.L), bilateral superior temporal gyri (STG.L/R), right precuneus (PCUN.R), and right dorsolateral superior frontal gyrus (SFGdor.R). However, two brain regions showed increased CBF: the right putamen (PUT.R) and left cerebellum VIII (Cere8.L) (FDR-corrected, P<0.05, with age and years of education as covariates) (Table 3 and Figure 2).

Figure 2 Brain regions showing significant between-group differences in CBF. The color scale indicates T-values. Warm (red) colors/positive T-values correspond to ANI > CN, whereas cool (blue) colors/negative T-values correspond to CN > ANI (FDR-corrected P<0.05, and cluster size ≥20 voxels. L: left hemisphere. R: right hemisphere). ANI, asymptomatic neurocognitive impairment; CBF, cerebral blood flow; CN, cognitively normal; FDR, false discovery rate.

FCS changes in ANI

In rs-fMRI, compared to the CN group, the ANI group showed significantly decreased FCS in the left lingual gyrus (LING.L) and left opercular part of the inferior frontal gyrus (IFGoperc.L), but significantly increased FCS in the MFG.L and left cerebellar crus I (CC1.L) (GRF-corrected, P<0.05; adjusted for age and years of education) (Table 4 and Figure 3A).

Figure 3 Brain regions showing significant between-group differences in FCS. (A) rs-fMRI; (B) movie-fMRI. The color scale indicates T-values. Warm (red) colors/positive T-values correspond to ANI > CN, whereas cool (blue) colors/negative T-values correspond to CN > ANI. (GRF-corrected P<0.05, and cluster size ≥20 voxels. L: left hemisphere. R: right hemisphere). ANI, asymptomatic neurocognitive impairment; CN, cognitively normal; FCS, functional connectivity strength; movie-fMRI, movie-watching state functional magnetic resonance imaging; rs-fMRI, resting-state functional magnetic resonance imaging.

In movie-fMRI, compared to the CN group, the ANI group exhibited significantly decreased FCS in the right orbital superior frontal gyrus (ORBsup.R) and left precuneus (PCUN.L), but significantly increased FCS in the right superior parietal gyrus (SPG.R) and left cerebellar crus II (CC2.L) (GRF-corrected, P<0.05; adjusted for age and years of education) (Table 4 and Figure 3B).

Alterations in CBF-FCS coupling

CBF-FCS coupling was quantified as the whole GM spatial correlation between the CBF map and the corresponding FCS map. The CN group showed weak positive coupling (rest: r=0.0348; movie: r=0.0364), while the ANI group showed weak negative coupling (rest: r=−0.0283; movie: r=−0.0354). Given the small effect size (|r| ~0.03), Figure 4 shows representative voxel-wise scatter plots (each dot represents a GM voxel) to visualize the spatial coupling pattern rather than to suggest a strong voxel-wise linear association. The coupling coefficients were significantly reduced in the ANI group compared with the CN group (Figure 5), with a larger between-group difference under the movie condition (rest: P=0.004; movie: P<0.001). In the sensitivity analysis, the CN-ANI difference remained significant and directionally consistent after excluding two mild CN movie outliers (1.5× IQR).

Figure 4 Scatter plots of voxel-wise correlations between CBF and FCS across whole-brain gray-matter voxels. (A) Resting-state, CN; (B) resting-state, ANI; (C) movie-state, CN; (D) movie-state, ANI. Each dot represents one gray-matter voxel. ANI, asymptomatic neurocognitive impairment; CBF, cerebral blood flow; CN, cognitively normal; FCS, functional connectivity strength.

Figure 5 Comparison of mean whole-brain GM CBF-FCS coupling coefficients (r) between the two groups. **, P<0.01; ***, P<0.001. ANI, asymptomatic neurocognitive impairment; CBF, cerebral blood flow; CN, cognitively normal; FCS, functional connectivity strength; GM, gray matter.

Correlation analysis

In the CN group, after FDR correction, CBF showed a negative correlation with memory in the PUT.R (r=–0.376, P=0.029), but a negative correlation with language (r=−0.421, P=0.013) and nadir CD4⁺ count (r=−0.339, P=0.049) in the STG.R. During movie-watching, FCS was found to be positively correlated with language in the SPG.R and PCUN.L (r=0.429, P=0.011; r=0.350, P=0.043), while during the resting-state condition, FCS was found to be negatively correlated with memory in the MFG.L and IFGoperc.L (r=−0.388, P=0.024; r=−0.415, P=0.015). Whole-brain CBF–FCS coupling showed a negative correlation with current CD4⁺ count under both conditions (movie-watching: r=−0.394, P=0.021; resting-state: r=−0.352, P=0.041) (Figure 6A).

Figure 6 Heatmaps of correlation analyses. (A) CN group; (B) ANI group. Heatmap colors represent Pearson correlation coefficients (r). *, P<0.05; **, P<0.01. ANI, asymptomatic neurocognitive impairment; CC1.L, left cerebelum_crus1; CC2.L, left cerebelum_crus2; Cere8.L, left cerebellum 8; CN, cognitively normal; IFGoperc.L, left opercular inferior frontal gyrus; LING.L, left lingual gyrus; MFG.L, left middle frontal gyrus; ORBsup.R, right orbital superior frontal gyrus; PCUN.L, left precuneus; PCUN.R, right precuneus; PHG.L, left ParaHippocampal; PUT.R, right putamen; SFGdor.R, right dorsolateral superior frontal gyrus; SPG.R, right superior parietal gyrus; STG.L, left superior temporal gyrus; STG.R, right superior temporal gyrus.

In the ANI group, CBF was found to be positively correlated with fine motor skills in the PUT.R (r=0.321, P=0.041), while CBF was found to be negatively correlated with executive function and memory in the Cere8.L and SFGdor.R (r=−0.505, P=0.001; r=−0.330, P=0.035). Resting-state FCS was found to be negatively correlated with the current CD4⁺ count (r=−0.346, P=0.027; r=−0.316, P=0.044) and CD4⁺/CD8⁺ ratio (r=−0.327, P=0.037; r=−0.308, P=0.049) in the MFG.L and IFGoperc.L. Whole-brain resting-state CBF-FCS coupling showed a negative correlation with the current CD4⁺ count (r=−0.311, P=0.048) and fine motor skills (r=−0.411, P=0.008) (Figure 6B).

ML model classification performance analysis

The classification performance demonstrated that among the single indicators, CBF exhibited the optimal discriminative ability (AUC: KNN =0.922, RF =0.896, SVM =0.920). For the dual-indicator combinations, the CBF + movie-FCS pair achieved notably high performance in the SVM model (AUC =0.909), slightly surpassing the CBF + rest-FCS combination (AUC range, 0.890–0.898). After incorporating the CBF-FCS coupling coefficient, the combination of CBF + movie-FCS + movie-coupling further improved performance (AUC: KNN =0.945, RF =0.962, SVM =0.929), exceeding that of the resting-state triple-indicator combination (AUC range, 0.915–0.928). The cross-state combination (CBF + rest-FCS + movie-FCS) had an AUC range of 0.898–0.918, which was lower than that of the movie-state triple-indicator combination. The combination incorporating all indicators achieved optimal performance (AUC: KNN =0.957, RF =0.947, SVM =0.943). The ROC curves of the KNN model are shown in Figure 7, and the diagnostic performance is presented in Table 5. The results for the other models are presented in Figures S1,S2, and Tables S1,S2.

Figure 7 ROC curves of different feature combinations in the KNN model. (A) Five single indicators; (B) CBF + rest-FCS, CBF + movie-FCS; (C) CBF + rest-FCS + rest CBF-FCS coupling, CBF + movie-FCS + movie CBF-FCS coupling; (D) CBF + rest-FCS + movie-FCS; (E) combination of all indicators (CBF + rest-FCS + movie-FCS + rest CBF-FCS coupling+movie CBF-FCS coupling). AUC, area under the curve; CBF, cerebral blood flow; FCS, functional connectivity strength; KNN, K-Nearest Neighbors; ROC, receiver operating characteristic.

Key feature importance analysis

The three significant combinations in the KNN model (the optimal diagnostic model) are shown in Figure 8. For the single CBF indicator (Figure 8A), CBF in specific brain regions (e.g., SFGdor.R, MFG.L, and PUT.R) demonstrated strong importance in classifying ANI. In the combination of CBF + movie-FCS + movie CBF-FCS coupling (Figure 8B), the CBF of MFG.L and SFGdor.R, as well as the CBF-FCS coupling coefficient, contributed substantially to ANI classification. For the full feature combination (Figure 8C), the movie CBF-FCS coupling coefficient contributed the most to ANI prediction. Additionally, the movie-FCS of ORBsup.R and CBF features also played important roles in ANI classification.

Figure 8 SHAP for different feature combinations in the KNN model. (A) Single CBF indicator; (B) combination of CBF + movie-FCS + movie CBF-FCS coupling; (C) full feature combination. Red-colored features indicate a higher likelihood of classification as ANI, while blue-colored features suggest a higher probability of classification as CN. ANI, asymptomatic neurocognitive impairment; CBF, cerebral blood flow; CC1.L, left cerebelum_crus1; CC2.L, left cerebelum_crus2; Cere8.L, left cerebellum 8; CN, cognitively normal; FCS, functional connectivity strength; IFGoperc.L, left opercular inferior frontal gyrus; KNN, K-Nearest Neighbors; LING.L, left lingual gyrus; MFG.L, left middle frontal gyrus; ORBsup.R, right orbital superior frontal gyrus; PCUN.L, left precuneus; PCUN.R, right precuneus; PUT.R, right putamen; SFGdor.R, right dorsolateral superior frontal gyrus; SHAP, SHapley Additive exPlanation; STG.R, right superior temporal gyrus.

Discussion

To the best of our knowledge, this study was the first to examine CBF-FCS coupling alterations in PLWH using an integrated BOLD-ASL framework. Combining dual-modality fMRI with ML, we aimed to identify sensitive neuroimaging biomarkers for ANI. The main findings were as follows: first, ANI patients exhibit significant CBF abnormalities; second, alterations in the spatial pattern of FCS were observed under both resting-state and movie-watching conditions, with partially overlapping yet distinct anomalous brain regions across states. Most importantly, this study was the first to discover that the whole-brain CBF-FCS coupling in ANI patients was reversed (shifting from a positive correlation to a negative correlation), and this reversal was more pronounced under the movie-state. Further, integrating multimodal features enabled the optimal identification of ANI (AUC >0.94), and this model significantly outperformed single-modality indicators.

This study identified characteristic CBF alterations in ANI patients compared to the CN group, including reduced CBF in higher-order cognitive cortices such as the MFG, STG, and PCUN, as well as increased CBF in the PUT and Cere8. These CBF changes reflect aberrant neural activity, neurotransmitter alterations, and neuroinflammation-related microvascular dysregulation (37,38). Cortical hypoperfusion may be attributed to vascular endothelial dysfunction caused by HIV-induced neuroinflammation and immune activation, which in turn impairs neurovascular regulatory function (7,39,40). Studies have shown that brain regions with enhanced neural activity have higher metabolic demands, leading to increased cerebral perfusion (41). The increased perfusion in subcortical structures (e.g., PUT) may represent a compensatory mechanism, reflecting elevated metabolic demand to offset cognitive decline (18). This pattern of “cortical hypoperfusion coupled with subcortical hyperperfusion” is consistent with the coexisting features of “temporal-occipital hypoperfusion and insular hyperperfusion” reported by Narvid et al. in elderly PLWH (16), suggesting that CBF regulation is already disrupted in ANI patients. This characteristic pattern of abnormal CBF provides important imaging evidence for the early neurobiological changes in ANI.

FCS reflects the contribution of each voxel to information transmission within the whole-brain network (25). This study found that both resting-state and movie-state FCS in ANI patients exhibited coexistent decreases and increases, but the abnormal patterns were state-specific. During the resting state, ANI patients exhibited reduced FCS in the LING.L and IFGoperc.L, indicating impaired neural synchronization involved in visual processing and language functions, which reflects early disruption of the brain’s intrinsic functional architecture (42,43). Conversely, increased FCS in the MFG.L and CC1.L suggests a compensatory mechanism that helps maintain cognitive function despite underlying neural injury (44,45). During the movie-watching state, reduced FCS was observed in the ORBsup.R and PCUN.L in the ANI group. The ORBsup.R mediates social cognition and emotional regulation processes, while the PCUN.L serves as a core component of the default mode network and is involved in self-referential processing (10,46). These findings indicate vulnerabilities in higher-order cognitive pathways (e.g., social-emotional integration pathways) among ANI patients under naturalistic stimulation. Conversely, the increased FCS in the SPG.R and CC2.L may reflect adaptive recruitment of attention and sensorimotor networks to support dynamic cognitive demands. Naturalistic movie-watching paradigms also tend to capture greater inter-individual variability than resting-state measures (26,47), which may partly account for the wider within-group dispersion observed under the movie condition. This is consistent with the trend observed in studies on first-episode psychosis and depression, where natural paradigms are more likely to detect abnormalities (28,48,49).

Previous studies have shown that brain regions with higher neuronal activity typically exhibit increased cerebral perfusion and energy supply, indicating that NVC plays a critical role in maintaining brain function (50). This study was the first to identify whole-brain NVC disruption in patients with ANI: the CN group showed weak positive coupling between CBF and FCS (resting-state: r=0.0348; movie-state: r=0.0364), whereas the ANI group exhibited a reversal to weak negative coupling (resting-state: r=−0.0283; movie-state: r=−0.0354). Importantly, the coupling effect size was small (|r| ~0.03), so we interpreted these results as reflecting a modest whole-GM spatial coupling pattern. Our primary inference, therefore, relies on the between-group differences in the coupling coefficients (Figure 5) rather than the voxel-wise P values (Figure 4). This reversal indicates a loss of the normal synergistic relationship between neural activity and hemodynamic supply, suggesting that neurovascular decoupling may represent a core feature of ANI.

The structural basis of NVC is the neurovascular unit (NVU), composed of neurons, glial cells, and microvessels (20). Impairment to any component of the NVU can impair its integrated function, leading to disrupted coupling between neural activity and local blood supply (25). HIV-induced chronic neuroinflammation and glial cell damage may disrupt the integrity of the NVU (2,51,52), resulting in unsynchronized or even reduced perfusion during increased neural activity. Precise regulation of NVC is central to brain homeostasis (23), and its disruption may be more important than isolated CBF or FCS abnormalities, serving as an early sign of failed functional compensation. Notably, NVC disruption was more pronounced during the movie-watching state. This may be attributed to the fact that natural stimuli require the engagement of more complex neural circuits (e.g., those underlying narrative comprehension and attention allocation) and impose higher demands on neural functional coordination (26,28). This amplifies the underlying coupling deficits in ANI patients, confirming the advantage of naturalistic paradigms in detecting neural functional abnormalities.

Correlation analyses showed that associations between imaging metrics and cognitive domain scores were overall modest and not consistently present across domains, whereas several imaging measures were related to immune indicators. Specifically, the CBF-FCS coupling coefficient showed a significant negative correlation with current CD4⁺ T-cell counts in both the CN and ANI groups, suggesting that immune status may be associated with impaired neurovascular integrity. In the ANI group, CBF was negatively correlated with abstraction/executive function in the Cere8.L, and CBF was negatively correlated with memory in the SFGdor.R. These findings suggest that neurovascular/network alterations in ANI may reflect early brain vulnerability and compensatory processes that are not linearly captured by cross-sectional neuropsychological performance; therefore, brain-behavior associations in the present study should be interpreted as exploratory.

The ML models demonstrated that the integration of multimodal neuroimaging markers significantly improved the discrimination of ANI. Although the single CBF indicator exhibited excellent classification performance (AUC: 0.896–0.922), multimodal feature combinations further enhanced the classification efficacy (AUC: 0.943–0.957). Notably, the combination of “CBF + movie-FCS + movie CBF-FCS coupling” consistently achieved excellent discriminative ability across all three MLs (AUC: 0.929–0.962), significantly outperforming combinations based solely on resting-state indicators. This suggests that the FCS and NVC captured through the movie paradigm offer discriminative power beyond that of traditional resting-state imaging, and more accurately reflect abnormal brain function in ANI patients when processing daily dynamic information (53). Related research has also demonstrated the feasibility of rs-fMRI-based ML models for early-stage HAND classification (13).

The feature importance analysis further revealed that the movie-state CBF-FCS coupling coefficient, movie-FCS in the ORBsup.R, and CBF in the SFGdor.R contributed most significantly to classification. These neurovascular markers are associated with higher-order cognition and social information processing, and they play a key role in distinguishing ANI. The full feature-combination model, which integrates all indicators, achieved optimal classification performance (AUC: KNN =0.957, RF =0.947, SVM =0.943). These results affirm that a multimodal approach combining CBF, FCS across both resting and naturalistic conditions, and NVC represents the most effective strategy for identifying ANI.

Limitations and prospects

This study had some limitations. First, the cross-sectional design limits the inferences that can be drawn about longitudinal trajectories between brain alterations and cognitive impairment. Second, the cohort included only male participants, which may limit the generalizability of the findings to female PLWH. Third, some cognitive domains were represented by a single test, which may reduce construct coverage and measurement reliability. Fourth, the MLs were not externally validated. Fifth, the multimodal protocol increases scan time and cost, potentially limiting near-term clinical feasibility. Future multicenter studies with broader sex representation, more comprehensive domain-level neuropsychological batteries, external validation, and streamlined imaging protocols are warranted.

Conclusions

This study was the first to systematically show that patients with ANI exhibit significant CBF, FCS, and NVC abnormalities. Notably, compared with the resting-state paradigm, the movie-watching paradigm was more sensitive in capturing functional abnormalities during dynamic social cognitive processes. ML models integrating multimodal indicators showed promising discriminative performance for ANI classification, with movie-state NVC indicators contributing the most prominently. These findings provide neuroimaging evidence of altered CBF, FCS, and their coupling in ANI. Additionally, the study revealed that neurovascular decoupling may represent a core feature of pathological changes in ANI. Multimodal imaging biomarkers are expected to offer new perspectives and potential therapeutic targets for ANI, and although brain-behavior correlations were modest in this cross-sectional sample, longitudinal follow-up and external validation will help determine their clinical relevance and translational potential for the development of clinically feasible protocols.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-aw-2110/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-aw-2110/dss

Funding: This work was supported by Beijing Hospital Authority Clinical Medicine Development special funding support (No. ZLRK202333), the National Natural Science Foundation of China (Nos. 82271963 and 61936013), the Beijing Natural Science Foundation (No. L222097), and the Open Project of Henan Clinical Research Center of Infectious Diseases (AIDS) (No. KFKT202403).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-aw-2110/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Beijing Youan Hospital (No. LL-2023-070-K) and informed consent was obtained from all individual participants.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Sami Saribas A, Cicalese S, Ahooyi TM, Khalili K, Amini S, Sariyer IK. HIV-1 Nef is released in extracellular vesicles derived from astrocytes: evidence for Nef-mediated neurotoxicity. Cell Death Dis 2017;8:e2542. [Crossref] [PubMed]
2. Saylor D, Dickens AM, Sacktor N, Haughey N, Slusher B, Pletnikov M, Mankowski JL, Brown A, Volsky DJ, McArthur JC. HIV-associated neurocognitive disorder--pathogenesis and prospects for treatment. Nat Rev Neurol 2016;12:234-48. Erratum in: Nat Rev Neurol 2016;12:309. [Crossref] [PubMed]
3. Rojas-Celis V, Valiente-Echeverría F, Soto-Rifo R, Toro-Ascuy D. New Challenges of HIV-1 Infection: How HIV-1 Attacks and Resides in the Central Nervous System. Cells 2019;8:1245. [Crossref] [PubMed]
4. Wallet C, De Rovere M, Van Assche J, Daouad F, De Wit S, Gautier V, Mallon PWG, Marcello A, Van Lint C, Rohr O, Schwartz C. Microglial Cells: The Main HIV-1 Reservoir in the Brain. Front Cell Infect Microbiol 2019;9:362. [Crossref] [PubMed]
5. Antinori A, Arendt G, Becker JT, Brew BJ, Byrd DA, Cherner M, et al. Updated research nosology for HIV-associated neurocognitive disorders. Neurology 2007;69:1789-99. [Crossref] [PubMed]
6. Grant I, Franklin DR Jr, Deutsch R, Woods SP, Vaida F, Ellis RJ, et al. Asymptomatic HIV-associated neurocognitive impairment increases risk for symptomatic decline. Neurology 2014;82:2055-62. [Crossref] [PubMed]
7. Gelman BB. Neuropathology of HAND With Suppressive Antiretroviral Therapy: Encephalitis and Neurodegeneration Reconsidered. Curr HIV/AIDS Rep 2015;12:272-9. [Crossref] [PubMed]
8. Hakkers CS, Arends JE, Barth RE, Du Plessis S, Hoepelman AI, Vink M. Review of functional MRI in HIV: effects of aging and medication. J Neurovirol 2017;23:20-32. [Crossref] [PubMed]
9. Ances BM, Hammoud DA. Neuroimaging of HIV-associated neurocognitive disorders (HAND). Curr Opin HIV AIDS 2014;9:545-51. [Crossref] [PubMed]
10. Jiang X, Hou C, Ma J, Li H. Alterations in local activity and whole-brain functional connectivity in human immunodeficiency virus-associated neurocognitive disorders: a resting-state functional magnetic resonance imaging study. Quant Imaging Med Surg 2025;15:563-80. [Crossref] [PubMed]
11. Liu D, Zhao C, Wang W, Wang Y, Li R, Sun J, Liu J, Liu M, Zhang X, Liang Y, Li H. Altered Gray Matter Volume and Functional Connectivity in Human Immunodeficiency Virus-Infected Adults. Front Neurosci 2020;14:601063. [Crossref] [PubMed]
12. Hall SA, Towe SL, Nadeem MT, Hobkirk AL, Hartley BW, Li R, Huettel SA, Meade CS. Hypoactivation in the precuneus and posterior cingulate cortex during ambiguous decision making in individuals with HIV. J Neurovirol 2021;27:463-75. [Crossref] [PubMed]
13. Hou C, Zhang M, Jiang X, Li H. Early-stage diagnosis of HIV-associated neurocognitive disorders via multiple learning models based on resting-state functional magnetic resonance imaging. Quant Imaging Med Surg 2025;15:7989-8007. [Crossref] [PubMed]
14. Korte N, Nortley R, Attwell D. Cerebral blood flow decrease as an early pathological mechanism in Alzheimer's disease. Acta Neuropathol 2020;140:793-810. [Crossref] [PubMed]
15. Joshi D, Prasad S, Saini J, Ingalhalikar M. Role of Arterial Spin Labeling (ASL) Images in Parkinson's Disease (PD): A Systematic Review. Acad Radiol 2023;30:1695-708. [Crossref] [PubMed]
16. Narvid J, McCoy D, Dupont SM, Callen A, Tosun D, Hellmuth J, Valcour V. Abnormal Cerebral Perfusion Profile in Older Adults with HIV-Associated Neurocognitive Disorder: Discriminative Power of Arterial Spin-Labeling. AJNR Am J Neuroradiol 2018;39:2211-7. [Crossref] [PubMed]
17. Ances BM, Sisti D, Vaida F, Liang CL, Leontiev O, Perthen JE, Buxton RB, Benson D, Smith DM, Little SJ, Richman DD, Moore DJ, Ellis RJHNRC group. Resting cerebral blood flow: a potential biomarker of the effects of HIV in the brain. Neurology 2009;73:702-8. [Crossref] [PubMed]
18. Su T, Mutsaerts HJ, Caan MW, Wit FW, Schouten J, Geurtsen GJ, Sharp DJ, Prins M, Richard E, Portegies P, Reiss P, Majoie CBAGEhIV Cohort Study. Cerebral blood flow and cognitive function in HIV-infected men with sustained suppressed viremia on combination antiretroviral therapy. AIDS 2017;31:847-56. [Crossref] [PubMed]
19. Liang X, Zou Q, He Y, Yang Y. Coupling of functional connectivity and regional cerebral blood flow reveals a physiological basis for network hubs of the human brain. Proc Natl Acad Sci U S A 2013;110:1929-34. [Crossref] [PubMed]
20. Iadecola C. The Neurovascular Unit Coming of Age: A Journey through Neurovascular Coupling in Health and Disease. Neuron 2017;96:17-42. [Crossref] [PubMed]
21. Phillips AA, Chan FH, Zheng MM, Krassioukov AV, Ainslie PN. Neurovascular coupling in humans: Physiology, methodological advances and clinical implications. J Cereb Blood Flow Metab 2016;36:647-64. [Crossref] [PubMed]
22. Wang R, Liu X, Sun C, Hu B, Yang L, Liu Y, Geng D, Lin J, Li Y. Altered Neurovascular Coupling in Patients With Mitochondrial Myopathy, Encephalopathy, Lactic Acidosis, and Stroke-Like Episodes (MELAS): A Combined Resting-State fMRI and Arterial Spin Labeling Study. J Magn Reson Imaging 2024;60:327-36. [Crossref] [PubMed]
23. Yang L, Zhao W, Kan Y, Ren C, Ji X. From Mechanisms to Medicine: Neurovascular Coupling in the Diagnosis and Treatment of Cerebrovascular Disorders: A Narrative Review. Cells 2024;14:16. [Crossref] [PubMed]
24. Li T, Liu T, Zhang J, Ma Y, Wang G, Suo D, Yang B, Wang X, Funahashi S, Zhang K, Fang B, Yan T. Neurovascular coupling dysfunction of visual network organization in Parkinson's disease. Neurobiol Dis 2023;188:106323. [Crossref] [PubMed]
25. Li H, Huang Z, Gao Z, Zhu W, Li Y, Zhou S, Li X, Yu Y. Sex Difference in General Cognition Associated with Coupling of Whole-brain Functional Connectivity Strength to Cerebral Blood Flow Changes During Alzheimer's Disease Progression. Neuroscience 2023;509:187-200. [Crossref] [PubMed]
26. Meer JNV, Breakspear M, Chang LJ, Sonkusare S, Cocchi L. Movie viewing elicits rich and reliable brain state dynamics. Nat Commun 2020;11:5004. [Crossref] [PubMed]
27. Eickhoff SB, Milham M, Vanderwal T. Towards clinical applications of movie fMRI. Neuroimage 2020;217:116860. [Crossref] [PubMed]
28. Rikandi E, Pamilo S, Mäntylä T, Suvisaari J, Kieseppä T, Hari R, Seppä M, Raij TT. Precuneus functioning differentiates first-episode psychosis patients during the fantasy movie Alice in Wonderland. Psychol Med 2017;47:495-506. [Crossref] [PubMed]
29. Guo CC, Nguyen VT, Hyett MP, Parker GB, Breakspear MJ. Out-of-sync: disrupted neural activity in emotional circuitry during film viewing in melancholic depression. Sci Rep 2015;5:11605. [Crossref] [PubMed]
30. Hou Y, Jin Y, Cai C, Qin Q, Tang H, Lyu F. Comparative Analysis of Epidemiological Features of HIV/AIDS Cases Aged Over and Under 50 Years Old - China, 2010-2022. China CDC Wkly 2023;5:1079-83. [Crossref] [PubMed]
31. Zhao H, Liu H, Wang L, Yang X, Wang S, Han M, Li J. Epidemiological Characteristics of Newly-Reported HIV Cases Among Youth Aged 15-24 Years - China, 2010-2019. China CDC Wkly 2020;2:913-6. [Crossref] [PubMed]
32. Wei L, Jia H, Geng Y, Dou Z, Zhao D, Gan X, Zhao Y. Trends in Life Expectancy of HIV-Infected Patients Receiving Antiretroviral Therapy - China, 2013-2023. China CDC Wkly 2025;7:838-42. [Crossref] [PubMed]
33. Shi C, Kang L, Yao S, Ma Y, Li T, Liang Y, Cheng Z, Xu Y, Shi J, Xu X, Zhang C, Franklin DR, Heaton RK, Jin H, Yu X. The MATRICS Consensus Cognitive Battery (MCCB): Co-norming and standardization in China. Schizophr Res 2015;169:109-15. [Crossref] [PubMed]
34. Lawton MP, Brody EM. Assessment of older people: self-maintaining and instrumental activities of daily living. Gerontologist 1969;9:179-86.
35. Liu F, Zhu C, Wang Y, Guo W, Li M, Wang W, Long Z, Meng Y, Cui Q, Zeng L, Gong Q, Zhang W, Chen H. Disrupted cortical hubs in functional brain networks in social anxiety disorder. Clin Neurophysiol 2015;126:1711-6. [Crossref] [PubMed]
36. Yan CG, Wang XD, Zuo XN, Zang YF. DPABI: Data Processing & Analysis for (Resting-State) Brain Imaging. Neuroinformatics 2016;14:339-51. [Crossref] [PubMed]
37. Chen J, Xue K, Yang M, Wang K, Xu Y, Wen B, Cheng J, Han S, Wei Y. Altered Coupling of Cerebral Blood Flow and Functional Connectivity Strength in First-Episode Schizophrenia Patients With Auditory Verbal Hallucinations. Front Neurosci 2022;16:821078. [Crossref] [PubMed]
38. Sukumar N, Sabesan P, Anazodo U, Palaniyappan L. Neurovascular Uncoupling in Schizophrenia: A Bimodal Meta-Analysis of Brain Perfusion and Glucose Metabolism. Front Psychiatry 2020;11:754. [Crossref] [PubMed]
39. Silva JN, Polesskaya O, Wei HS, Rasheed IY, Chamberlain JM, Nishimura C, Feng C, Dewhurst S. Chronic central nervous system expression of HIV-1 Tat leads to accelerated rarefaction of neocortical capillaries and loss of red blood cell velocity heterogeneity. Microcirculation 2014;21:664-76. [Crossref] [PubMed]
40. Jia P, Zhao Z, Hulgan T, Bush WS, Samuels DC, Bloss CS, et al. Genome-wide association study of HIV-associated neurocognitive disorder (HAND): A CHARTER group study. Am J Med Genet B Neuropsychiatr Genet 2017;174:413-26. [Crossref] [PubMed]
41. Venkat P, Chopp M, Chen J. New insights into coupling and uncoupling of cerebral blood flow and metabolism in the brain. Croat Med J 2016;57:223-8. [Crossref] [PubMed]
42. Wei W, Wang T, Abulizi T, Li B, Liu J. Altered Coupling Between Resting-State Cerebral Blood Flow and Functional Connectivity Strength in Cervical Spondylotic Myelopathy Patients. Front Neurol 2021;12:713520. [Crossref] [PubMed]
43. Zhu J, Zhuo C, Xu L, Liu F, Qin W, Yu C. Altered Coupling Between Resting-State Cerebral Blood Flow and Functional Connectivity in Schizophrenia. Schizophr Bull 2017;43:1363-74. [Crossref] [PubMed]
44. Straathof M, Sinke MR, Dijkhuizen RM, Otte WM. A systematic review on the quantitative relationship between structural and functional network connectivity strength in mammalian brains. J Cereb Blood Flow Metab 2019;39:189-209. [Crossref] [PubMed]
45. Cocchi L, Harding IH, Lord A, Pantelis C, Yucel M, Zalesky A. Disruption of structure-function coupling in the schizophrenia connectome. Neuroimage Clin 2014;4:779-87. [Crossref] [PubMed]
46. Smitha KA, Akhil Raja K, Arun KM, Rajesh PG, Thomas B, Kapilamoorthy TR, Kesavadas C. Resting state fMRI: A review on methods in resting state connectivity analysis and resting state networks. Neuroradiol J 2017;30:305-17. [Crossref] [PubMed]
47. Finn ES, Glerean E, Khojandi AY, Nielson D, Molfese PJ, Handwerker DA, Bandettini PA. Idiosynchrony: From shared responses to individual differences during naturalistic neuroimaging. Neuroimage 2020;215:116828. [Crossref] [PubMed]
48. Finn ES, Bandettini PA. Movie-watching outperforms rest for functional connectivity-based prediction of behavior. Neuroimage 2021;235:117963. [Crossref] [PubMed]
49. Gruskin DC, Rosenberg MD, Holmes AJ. Relationships between depressive symptoms and brain responses during emotional movie viewing emerge in adolescence. Neuroimage 2020;216:116217. [Crossref] [PubMed]
50. Huang H, Zhao K, Zhu W, Li H, Zhu W. Abnormal Cerebral Blood Flow and Functional Connectivity Strength in Subjects With White Matter Hyperintensities. Front Neurol 2021;12:752762. [Crossref] [PubMed]
51. Ellis RJ, Marquine MJ, Kaul M, Fields JA, Schlachetzki JCM. Mechanisms underlying HIV-associated cognitive impairment and emerging therapies for its management. Nat Rev Neurol 2023;19:668-87. [Crossref] [PubMed]
52. Hong S, Banks WA. Role of the immune system in HIV-associated neuroinflammation and neurocognitive implications. Brain Behav Immun 2015;45:1-12. [Crossref] [PubMed]
53. Jääskeläinen IP, Sams M, Glerean E, Ahveninen J. Movies and narratives as naturalistic stimuli in neuroimaging. Neuroimage 2021;224:117445. [Crossref] [PubMed]