Muwei Li

Functional-connectivity analyses often treat signals sampled from nearby cortical positions as interchangeable, despite anatomical and hemodynamic transitions at the gray-white boundary. We tested whether sampling position constitutes a state-sensitive dimension of macroscale functional organization. In 176 Human Connectome Project young adults with four 7 T resting-state and four movie-watching runs, BOLD time series were sampled from subject-specific midthickness and gray-white-boundary surfaces, summarized across 400 cortical parcels, and embedded in a common functional-gradient space. Movie viewing reduced principal-gradient separation between surfaces, but this effect was spatially heterogeneous: Visual, Limbic and Default networks converged, whereas Salience/ventral-attention cortex differentiated. Shapley decomposition and counterfactual analyses showed that the state change reflected coordinated reconfiguration of both surface representations rather than one surface approaching the other. Changes in the original connectivity profiles tracked gradient changes, indicating that the result was not solely introduced by embedding. Both surfaces also exhibited stimulus-locked intersubject functional connectivity during movies. However, stimulus-shared separation captured only part of movie-state organization and did not explain the movie-minus-rest change. Finally, in 172 matched participants, resting-state separation maps generalized from 7 T to 3 T, whereas absolute magnitude and cross-field individual ranking were less stable. These findings establish cortical sampling position as an interpretable measurement dimension of functional-connectivity geometry. Its spatial organization is reproducible across acquisitions, but its magnitude and state dependence vary, emphasizing that where BOLD signals are sampled relative to the cortical boundary can shape conclusions about macroscale brain organization.

Natural audiovisual perception may not be fully captured by decomposing movies into auditory and visual streams. I introduce a computational-counterfactual framework that keeps movie viewing intact while varying only AI-derived descriptions of the same clips. Using 7 Tesla movie fMRI imaging data from 176 participants, I tested whether cortical responses were better predicted by native audiovisual semantics than by a dimension-matched additive reconstruction from audio-only and video-only descriptions. The native model outperformed the matched additive baseline under content-aware purged cross-validation, with strongest gains in auditory, visual, and dorsal attention systems. Representational-similarity, feature-replacement, and content-gating analyses showed that the advantage reflected feature- and network-specific routing linked to coherent audiovisual semantic emergence rather than raw auditory-visual discrepancy. The effect survived stronger temporal purging and repeat-content exclusion, suggesting that intact movie viewing evokes cortical structure aligned with native audiovisual meaning beyond additive unimodal semantics.

Functional MRI studies have traditionally focused on gray matter, whereas white-matter BOLD signals have often been treated as weak or artifactual. Recent work suggests that white-matter BOLD fluctuations contain reproducible functional information, but most gray-to-white matter analyses rely on marginal functional connectivity, which cannot separate pairwise coupling from shared variance among distributed cortical systems. Here, we used a multivariate cortical encoding framework to test whether spontaneous white-matter BOLD activity can be predicted from distributed cortical gray-matter activity and whether this predictive structure reveals organization beyond marginal connectivity. Resting-state fMRI data from 81 Human Connectome Project young adult participants were analyzed using a strict white-matter mask with no overlap with cortical predictors. For each white-matter voxel, time series from 400 Schaefer cortical parcels were used to predict held-out white-matter BOLD signals with nested leave-one-run-out ridge regression. Cortical activity modestly but reliably predicted white-matter BOLD dynamics, demonstrating consistent cross-validated prediction accuracy across a broad spatial extent of the white matter. Ridge beta fingerprints strongly recapitulated marginal functional connectivity fingerprints, indicating a shared functional backbone, but their first gradients diverged reproducibly. This beta-FC divergence axis organized FC-adjusted prediction residuals and remained robust after controlling for gray-matter proximity, mask-boundary distance, white-matter prevalence, temporal signal variability, spatial coordinates, and spatial autocorrelation. The high-divergence end showed relatively low marginal FC but high FC-adjusted prediction residuals and was enriched for posterior thalamic/optic-radiation and posterior corona-radiata anatomy. These findings suggest that multivariate cortical encoding reveals a tract-organized dimension of white-matter functional coupling not captured by pairwise connectivity alone.

Naturalistic paradigms offer a powerful tool to investigate human brain function, but it remains difficult to link rich, continuous movie content to distributed brain activity in an interpretable way. In this study, I use a multimodal large language model (Gemini) as an automated “semantic annotator” to bridge naturalistic movie stimuli, brain responses, and cognitive performance. Using the Human Connectome Project movie-watching dataset, I segmented the film into 293 overlapping clips, prompting Gemini to rate each clip on 11 psychologically interpretable dimensions. Simultaneously, I extracted clip-wise BOLD activation patterns from the fMR images in 360 cortical ROIs. In this way, the AI and the brain effectively “watch” the same movies in parallel. For each brain ROI, I then fit linear regression models to predict clip-to-clip variation in movie-evoked responses from these features. Gemini-derived features robustly predicted movie-evoked responses in temporal, medial parietal, and lateral frontal association cortex, but explained little variance in unimodal somatosensory, dorsal parietal, insular, and piriform regions. Feature-weight maps reflected known functional specializations, and features with the largest global influence overlapped with the most explainable ROIs. Partial least squares analysis revealed that individual differences in resting-state connectivity strength and semantic explainability covaried along an asymmetric intrinsic axis: strongly integrated sensory-opercular systems at rest were associated with poorer AI predictability, whereas a smaller set of dorsal and medial association regions showed enhanced alignment. Finally, regional AI explainability in medial parietal and left perisylvian association areas was positively related to specific cognitive abilities. Together, these findings demonstrate that interpretable features from AI models provide a simple and scalable framework for quantifying AI-derived semantic predictability in naturalistic settings, offering a practical framework for utilizing artificial models as semantic references to probe human neural processing and individual differences.

Hemodynamic lag in white matter (WM) remains poorly understood despite its relevance to neurovascular health. We developed Local Propagation Mapping to quantify the spatiotemporal architecture of relative hemodynamic timing within WM. In young adults, WM lag architecture was reliable, aligned with venous anatomy, and reconfigured during working-memory task engagement; the magnitude of macroscopic lag modulation was associated with cognitive performance. Across aging, resting-state WM lag showed tract-specific bidirectional shifts and increased spatial fragmentation, with partial overlap with task-sensitive systems observed in young adults. Static baseline markers did not significantly mediate age-related cognitive differences, suggesting that aging effects may not operate through a simple resting-state pathway. Together, these findings establish relative WM lag as an informative dimension of neurovascular regulation and motivate a baseline-load hypothesis for future studies of cognitive aging.

Macroeconomic conditions shape daily constraints and perceived environment, yet it remains unclear whether real-world economic dynamics are reflected in intrinsic brain organization at the population level. Here, I leveraged the Human Connectome Project (HCP) acquisition timeline as a naturalistic sampling frame to test whether macroeconomic time series vary with large-scale resting-state networks. Resting-state fMRI from 726 healthy young adults was aligned to quarterly macroeconomic indicators using the HCP "Quarter" variable. I quantified intrinsic organization using network functional connectivity (FC) and network-level amplitude of low-frequency fluctuations (ALFF). I implemented (1) a quasi-natural experiment contrasting a pre- vs post-gasoline-price shock (collapse) cohort, and (2) quarter-level partial correlations between network measures and four macro indicators, including gasoline price, consumer sentiment, unemployment, and stock market return. Post-shock participants showed significantly higher within-network FC across all seven networks and selective increases in between-network coupling concentrated among sensory and attention systems, as well as limbic-default interactions. ALFF exhibited bidirectional shifts: Visual ALFF increased post-shock, whereas Limbic and Frontoparietal ALFF were higher pre-shock. Quarter-level analyses mirrored this differential pattern, with gas price positively associated with Limbic and Frontoparietal ALFF and negatively with Visual and Dorsal Attention ALFF. Across macro indicators, gasoline price and consumer sentiment showed the most widespread FC associations, with largely opposing correlation signatures, while unemployment and equity returns were comparatively weak after correction. These findings suggest that salient, behaviorally proximal macroeconomic dynamics, particularly energy price variation, track population-level differences in intrinsic brain network architecture, motivating future work with finer-grained timing and individual-level exposure measures to strengthen causal inference.

Functional connectivity (FC) measurements in the brain using BOLD fMRI have largely focused on correlations between gray-matter (GM) BOLD signals, ignoring the possible relevance of white-matter (WM) in delineating neural networks. We first employed a tract-informed mediation framework to test whether WM pathways contribute indirect, bundle-specific effects to cortico-cortical coupling. Using resting-state fMRI data from the Human Connectome Project, we quantified pathway-level mediation effects across GM-WM-GM units and found that WM BOLD signals significantly mediated GM-GM associations, with clear tract-dependent heterogeneity. Building upon this, we examined how dynamic features of WM BOLD signals relate to time-varying intra-cortical FC. We found that temporal variability of WM signals, rather than their mean amplitudes, more closely tracked fluctuations in GM-GM dynamic connectivity, with correlations reaching approximately 0.4 across specific tracts. Together, these results indicate that WM BOLD signals are related to how cortical regions interact, underscoring their integral role in large-scale brain network organization.

Naturalistic paradigms offer a powerful window into human cognition, but it remains difficult to link rich, continuous movie content to distributed brain activity in an interpretable way. In this study, I use a multimodal large language model (Gemini) as an automated "semantic annotator" to bridge naturalistic movie stimuli, brain responses, and cognitive performance. Using the Human Connectome Project movie-watching dataset, I segmented the film into 293 overlapping clips, prompted Gemini to rate each clip on 11 psychologically interpretable dimensions, and simultaneously extracted clip-wise BOLD activation patterns from the fMRI images in 360 cortical parcels. In this way, the AI and the brain effectively "watch" the same movies in parallel. For each parcel, I then fit linear regression models to predict clip-to-clip variation in movie-evoked responses from these features. Gemini-derived features robustly predicted movie-evoked responses in temporal, medial parietal, and lateral frontal association cortex, but explained little variance in unimodal somatosensory, dorsal parietal, insular, and piriform regions. Feature-weight maps recapitulated known functional specializations, and features with the largest global influence overlapped with the most explainable parcels. Partial least squares analysis revealed that individual differences in resting-state connectivity strength and semantic explainability covaried along an asymmetric intrinsic axis: strongly integrated sensory-opercular systems at rest were associated with poorer AI predictability, whereas a smaller set of dorsal and medial association regions showed enhanced alignment. Finally, regional AI explainability in medial parietal and left perisylvian association areas was positively related to fluid and crystallized cognitive abilities. Together, these findings demonstrate that prompt-defined, interpretable features from foundation models provide a simple and scalable framework for quantifying brain-AI alignment in naturalistic settings, offering a practical bridge between biological and artificial semantic representations.

Functional magnetic resonance imaging studies have traditionally focused on gray matter, overlooking white matter despite growing evidence that functional blood oxygenation-level dependent effects also occur there. In particular, functional coupling across the gray-white matter boundary, an interface between local and global processing, remains poorly understood. This study introduces two metrics: gray-white matter functional connectivity, which captures temporal synchrony across the boundary, and gray-white blood oxygenation-level dependent power ratio, which reflects differences in signal amplitude. Gray-white matter functional connectivity aligns with patterns of myelination, long-range connectivity, and sensorimotor organization, suggesting efficient signal transmission. In contrast, the power ratio shows an inverse pattern, with higher values in higher-order regions, possibly reflecting increased metabolic demands in white matter. It also increases with age (8 to 21 years), suggesting developmental shifts in energetic demands. Together, these metrics highlight distinct yet complementary roles of signal fidelity and energy modulation at the gray-white matter boundary.