Human Brain Mapping

Foundation models pre-trained on large neuroimaging datasets offer a promising approach to overcome the limited sample sizes typical of mental health imaging studies, yet their generalization across diverse clinical populations remains unclear. We present the first systematic benchmark of four publicly available structural MRI foundation models -- AnatCL, BrainIAC, 3D-Neuro-SimCLR, and SwinBrain -- on tasks relevant to mental health research. Using T1-weighted MRI from Parkin-sons Progression Markers Initiative (PPMI), Healthy Brain Network (HBN), and Nathan Kline Institute (NKI), we evaluate these models on sex classification, brain age prediction, and Parkinsons disease (PD) classification, benchmarking against models trained from FreeSurfer-derived cortical thickness and cortical surface area features, as well as an un-trained CNN baseline. Although some individual foundation models out-performed FreeSurfer on particular tasks and datasets, 3D-Neuro-SimCLR demonstrated the most consistent performance overall, with the notable exception of HBN sex classification, and all models failed to classify early-stage Parkinsons disease above chance. Notably, untrained CNNs achieved performance comparable to or exceeding FreeSurfer in multiple instances, establishing them as computationally efficient reference models. The cross-model feature correlation analysis reveals that foundation model representations correlate differently with traditional cortical measurements. These findings position structural MRI foundation models, particularly 3D-Neuro-SimCLR and AnatCL, as promising avenues to boost the performance of neuroimaging predictive models in mental health.

Big neuroimaging data enable researchers to study subtle structural and functional brain changes and relationships between brain characteristics and genetics, lifestyle, and disease factors. However, substantial effort is needed to minimise technical, non-biological differences between data batches to avoid incorrect inferences. In this study, we address a previously identified bias in UK Biobank FreeSurfer IDPs derived from only the T1 image compared to those using both T1 and T2-FLAIR by treating the bias as a batch effect and using harmonisation approaches. We investigate and characterise this bias through direct within-participant comparison at the image and IDP level, comparing the results with those seen in the wider UKB sample. We then assess different methods of addressing the effect of missing T2-FLAIR, starting from simple linear regression before moving to ComBat, a widely used harmonisation method, testing different approaches for applying ComBat and showing its similarity to simple linear regression. Finally, we examine how ComBat estimates vary with batch and sample size. Our results show clear benefits in using both T1 and T2-FLAIR data in FreeSurfer, as opposed to just the T1, which is more common, with the pial surface fitting being less likely to fail and showing greater biologically plausible inter-subject variability. This is particularly important for cortical thickness IDPs, where T2-FLAIR omission leads to reduced true variability and systematic underestimation, as shown through within-participant repeat testing. We demonstrate that ComBat can address this bias, with its standard use (i.e., applied separately on different IDP categories) showing the best improvement in cortical thickness measures where the bias is strongest, and we find that it is important not to pool ComBat priors across different classes of IDPs. Our proposed version of ComBat with a reference batch (i.e., estimating mean and variance only from data with T2-FLAIR available) performed best in recovering both mean and variance differences between batches across different IDP classes and offers a promising approach for cases where a reference batch is clearly identifiable. While ComBat reliably corrects mean (additive) batch effects with relatively small sample sizes ({approx}30 subjects per batch), we show that its variance (multiplicative) correction is substantially less stable, requiring much larger sample sizes and becoming unreliable when batches are small or imbalanced, or when there is a large variance difference between them.

BrainAge models hold promise as a clinical biomarker for developmental brain health, especially in childhood when there is the potential for early intervention. To distinguish between normative developmental variance and pathological divergence, BrainAge models should reflect the dynamic and diverse neurodevelopmental processes that occur in distinct developmental windows across childhood. We utilized multi-modal neuroimaging data from three pediatric cohorts covering ages 4 to 13 years (n = 1005, 2126 scans), split into Train and Test datasets. Twelve sex-stratified BrainAge models were built stratified by type and different combinations of neuroimaging features. Model types were "Full-Span" models covering the full age range, and "Phase-Specific" models split into early- and late-childhood. We first compared BrainAge estimates in the Test dataset amongst our candidate models, then benchmarked the best-performing model against published pre-trained models and DNA-based biological age measures. Our findings show that a BrainAge model that was phase-specific and consisted of both structural and functional features (cortical thickness, subcortical volumes, and functional network integration measures) showed good prediction of age and best distinguished between healthy and symptomatic subgroups. We present a proof-of-concept for developmental models supporting building BrainAge models of higher temporal resolution that align to different childhood developmental phases.

Normative modeling quantifies individual deviation from population norms by estimating the conditional mean and variance of brain-derived measures as functions of clinically relevant parameters such as age. The rapid growth of multicenter consortia has created an urgent need for normative models that incorporate batch harmonization. Several harmonization methods based on linear mixed models--ComBat, GAMLSS, HBR, and Generalized Normative Modeling (GNM)--offer explicit formulations of the mean and variance, making them natural candidates for batch-harmonized normative modeling; yet the absence of a unified theoretical framework leaves it unclear whether and how these methods support the computation of batch-harmonized z-scores. We bridge this gap by writing existing harmonization methods as special cases of a single location-scale equation, y = m(x, {Theta})+{sigma}(x, {Theta}){varepsilon} , which we term the unified form of batch harmonization equation for normative modeling. The methods differ only in the functional forms of m and{sigma} , how batch parameters enter{Theta} , and how{Theta} is estimated. This unified form yields both harmonized data y* and site-invariant z-scores from the same model, providing a common theoretical language for harmonized normative modeling. Building on this framework, we evaluate the underlying regression engines (parametric, spline, Gaussian process, kernel, deep learning), sensitivity to outliers, computational scalability, and federated decomposability for privacy-preserving multi-center computation. By clarifying what each method assumes, what it delivers, and where the boundaries of current methodology lie, the unified equation establishes a principled foundation for method selection and charts a path toward reliable, scalable, and privacy-aware normative modeling across multi-center neuroimaging.

IntroductionWhile global topological properties of brain networks reach relative maturity early in development, functional reconfigurations at the regional level continue throughout adolescence to support cognitive maturation. However, regional age and sex-specific developmental patterns of functional reconfiguration remain incompletely understood. MethodsWe analyzed resting-state fMRI data from 528 participants aged 5-21 years from the Human Connectome Project in Development. Three regional graph-theory metrics (betweenness centrality, hub score, and local efficiency) were computed for each individuals functional network. Cognition was measured using NIH toolbox. Parallel factor analysis was employed to decompose an individual x region x metric array into factors representing distinct developmental properties in the full sample and separately for males and females. Brain-cognition associations were examined in developmental subgroups (<13, 13-18, >18 years). ResultsThree factors emerged, characterizing visual, multimodal integration, and higher-order factors. Across development, metrics capturing network integration (betweenness centrality and hubness) showed general stability, while metrics capturing segregation (local efficiency) presented distinct peaks, particularly in the visual factor. Females showed earlier peaks and declines in higher-order factor, while males exhibited greater variability and protracted maturation in multimodal and higher-order factors. Brain-cognition associations were modest with early childhood and crystallized cognition composites showed small negative correlations with hub score in entire sample (r=-0.212) and local efficiency in males aged <13 years (r=-0.215). ConclusionFindings highlight nonlinear, sex-specific functional reconfiguration at region-level during childhood and adolescence, underscoring the importance of sex-stratified analyses in developmental and providing a crucial foundation for future investigations of developmental disorders.

Functional magnetic resonance imaging (fMRI) data are inherently complex, characterized by high dimensionality, intricate inter-regional dependencies, and substantial individual variability across experimental paradigms. Traditional linear mixed models (LMMs) provide a principled framework that models population-level fixed effects while estimating variance components arising from subject-level random effects; however, they often fail to adequately capture nonlinear relationships inherent in neuroimaging data. To address these limitations, we introduce the nonlinear mixed model (NMM) approach, an innovative extension of the LMM framework that integrates neural networks to flexibly model complex fixed-effect relationships while preserving the random-effects structure to account for individual differences. NMM advances fMRI analysis by: (1) identifying robust functional connectivity (FC) patterns consistently observed across multiple paradigms; (2) leveraging SHapley Additive exPlanations (SHAP) analysis to provide post-hoc interpretability of the nonlinear fixed effects, quantifying how age, sex, and paradigm contribute to predicted FC and how these effects are distributed across large-scale brain networks; and (3) using subject-specific random effects as neural fingerprints that not only show systematic variability across attention and default mode systems but also predict standardized cognitive scores, demonstrating biological relevance. Applied to the Philadelphia Neurodevelopmental Cohort (PNC) across emotion, n-back, and resting-state paradigms, NMM achieved superior model fit relative to classical LMMs, as evidenced by lower mean squared error (MSE) in predicting FC. This framework offers a statistically rigorous and practically explainable approach for modeling large-scale FC from modest covariates while explicitly separating population-level effects from stable individual variability in functional brain organization.

Clinical adoption of new biomedical techniques depends on establishing reference values against which individual patients can be compared. In resting-state functional MRI (rsfMRI), most biomarker research has relied on the case-control paradigm, whose underlying assumptions are often invalid as diseases are frequently heterogeneous, limiting biomarker generalizability. Normative modeling offers a complementary alternative by characterizing individual deviations against a reference population. However, in rsfMRI, normative modeling has been applied almost exclusively to functional connectivity, with limited attention to age trajectories and sex effects. We address these gaps by developing a spatial normative model of four rsfMRI metrics that capture complementary features of the blood-oxygen-level-dependent (BOLD) signal across age and sex. Five publicly available datasets were aggregated to form a sample of 1,978 participants aged 10-30 years. Four metrics were computed for each of 110 grey matter regions: amplitude of low-frequency fluctuations (ALFF), fractional amplitude of low-frequency fluctuations (fALFF), regional homogeneity (ReHo), and Hurst exponent. A machine-learning model based on hierarchical Bayesian regression with a non-Gaussian likelihood was fitted per metric, modeling non-linear age effects, sex, and multi-site acquisition. Models were well calibrated across all four metrics, with fALFF showing the strongest predictive performance and Hurst exponent the weakest. Normative trajectories varied across brain regions for each metric, but on average, the median of each distribution remained bounded across regions, while the spread was more regionally variable. All four metrics showed predominantly negative slopes with age, indicating a decrease in each metric over the age window. This work provides a normative reference across four rsfMRI metrics that capture distinct features of the BOLD signal, complementing the case-control paradigm and supporting individual-level inference.

BackgroundWhile resting-state fMRI demonstrated that brain networks are spatially dynamic (expanding, shrinking, and changing complexity over time), understanding the transient spatial network interactions that remain poorly characterized is critical for revealing the mechanisms underlying brain disorders. MethodsWe introduce DESINE (Dynamic Estimation of Spatially Interactive Networks), a novel framework using joint density distributions (2D histograms) of voxel-wise activity to quantify 4D spatial network interactions across sliding windows. We analysed transient deviations from the average functional state using root-mean-square error (RMSE) and mean absolute deviation (MAD), and characterized recurring interaction patterns using k-means clustering. We applied DESINE to 91 network pairs (14 networks) in a cohort of 508 subjects (315 healthy controls; 193 patients with schizophrenia, SZ). ResultsSZ is characterized by a significantly "constrained dynamic repertoire" of network interactions. SZ patients showed markedly lower means and standard deviations for both RMSE and MAD metrics across network pairs, particularly in regions of high activity, indicating systematic rigidity. Cluster analysis revealed significant alterations in state affinity metrics, suggesting a global breakdown in the brains capacity to preserve diverse, high-fidelity spatial configurations. Critically, these interaction metrics were associated with cognitive performance, symptom scores on the positive and negative syndrome scale, and chlorpromazine equivalent drug scores. ConclusionsThis work introduces DESINE as a global, voxel-agnostic framework for characterizing time-varying spatial interactions. Our findings highlight spatial rigidity as a fundamental feature of psychopathology, suggesting that the inability to express a diverse range of spatial interactions is a factor underlying cognitive deficits in schizophrenia.

The relationship between cortical morphology and intelligence during adolescence has been widely studied, with existing literature reporting varying degrees of association across different modeling approaches. This study provides a comprehensive comparison of model performance in investigating the association between crystallized intelligence and cortical surface area using data from 11,351 subjects in the Adolescent Brain Cognitive Development (ABCD) study. We evaluate ten widely used models ranging from linear regression to graph convolutional networks across three covariate adjustment formulations: full (no adjustment), partial (age and sex adjusted), and total surface area (TSA) partial (age, sex, and TSA adjusted). Using bootstrap resampling with 50 iterations, we estimate the fraction of variance explained (FVE) for each model. Our results suggest that more complex models do not lead to higher FVE, with LASSO having the highest FVE of 15.9% (full formulation), Ridge at 10.5% (partial formulation), and Principal Component Regression (PCR) with 102 PCs at 2.5% (TSA partial formulation). Our results also reveal that the relationship between cortical surface area and crystallized intelligence is predominantly driven by global factors age, sex, and TSA, rather than by localized cortical surface area.

Background Children in low- and middle-income countries (LMICs) face an elevated risk of developmental delay, yet scalable neuroimaging tools to study early brain development in these contexts remain limited. Children who are HIV-exposed but uninfected (CHEU) represent a growing population with evidence of language and motor delays and altered brain development compared with children who are HIV-unexposed (CHU). Ultra-low-field (ULF) MRI offers a more affordable alternative to conventional high-field (HF) MRI, but its application in early childhood remains underexplored. Methods We compared brain volumes derived from ULF (64mT) and HF (3T) MRI in South African CHEU and CHU as part of the DolPHIN-2 PLUS study. Volumetric segmentation was performed using FreeSurfer v7.4.1 and SynthSeg on the Flywheel platform. Agreement between modalities was assessed using Pearsons and Lins concordance correlation coefficients across global and subcortical regions. Associations between ULF-derived brain volumes and developmental outcomes, measured by the Bayley Scales of Infant Development, Third Edition, were evaluated using partial correlations adjusted for sex and age. Results Forty-five children (9 CHEU, 36 CHU; mean age 45.6 months) had paired ULF and HF scans of usable quality. Strong correlations were observed between ULF and HF volumes for global white and grey matter regions (r > 0.92) and larger subcortical grey matter structures such as the thalamus, caudate, and putamen (r = 0.86-0.89). Moderate-to-weak correlations were evident in smaller structures (hippocampus, pallidum, amygdala). ULF underestimated most grey matter volumes, and overestimated total white matter volume relative to HF. ULF-derived global and subcortical volumes were associated with receptive and expressive communication (r = 0.34-0.59, all p < 0.05). Conclusions ULF MRI produces brain volume estimates comparable to HF MRI and captures meaningful associations with early language development. These findings support ULF MRI as a feasible and scalable tool for studying neurodevelopment in vulnerable paediatric populations in LMICs.

The proliferation of gambling advertising has intensified concerns regarding its influence on vulnerable populations, yet the neural mechanisms underlying cue-reactivity to these stimuli remain underexplored in ecologically valid settings. This study protocol proposes a novel methodological framework to investigate prefrontal cortical responses to gambling advertisements in individuals with varying degrees of gambling experience. Materials and methods: This cross-sectional study will recruit 44 participants, divided into a clinical group (individuals with high-frequency gambling or gambling disorder) and a matched control group. Neural activity will be recorded using fNIRS while participants view gambling-related, neutral, violent, and sexual stimuli. Secondary measures include validated scales for gambling severity (SOGS), impulsivity, sensation seeking, and alexithymia. Data analysis will primarily utilize inter-subject correlation (ISC) to quantify neural synchronization and multiband frequency decomposition to capture dynamic affective processing. Advanced preprocessing, including short-channel regression, will be applied to ensure signal robustness. Discussion: By combining portable neuroimaging with a data-driven ISC approach, this study aims to identify objective neural markers of gambling vulnerability. The findings will provide novel insights into the idiosyncratic processing of commercial stimuli, potentially informing public health policies and the development of more effective evidence-based regulations for gambling marketing.

Normative modeling expresses individual brain phenotypes as z-scores relative to a population norm, but in multi-site studies batch effects contaminate these z-scores and undermine their biomarker value. Existing approaches either harmonize data before fitting a normative model (ComBat+Normative), letting residual site effects leak into z-scores, or use parametric one-step methods (GAMLSS, HBR) that cannot flexibly model multivariate covariate interactions. We propose Generalized Normative Modeling (GNM), a onestep hierarchical framework that jointly estimates the global trajectory and site-specific effects via NUFFT-accelerated kernel regression with GCV bandwidth selection. Because the z-score is the ratio of batch-corrected residual to batch-corrected scale, residual site variance cancels algebraically -- a property we term self-correction. On ABIDE I cortical thickness (387 HC, 11 sites, 68 ROIs) and HarMNqEEG log-power spectra (1,564 subjects, 14 sites, 18 channels x 235 frequency bins), GNM produced the most site-invariant z-scores and best age-signal preservation among four methods. This work provides an open-source MATLAB toolbox with a declarative formula interface (https://github.com/LMNonlinear/Generalized-Normative-Modeling), enabling reliable individual-level inference in pooled multi-site cohorts and advancing the use of normative deviations as clinical biomarkers in precision psychiatry and neurology.

Machine-learning models are increasingly used to predict cognitive and clinical outcomes from neuroimaging data, yet challenges in fairness and generalizability remain. Large-scale datasets are often racially and ethnically imbalanced, leading to systematic performance disparities, with models typically achieving higher accuracy for majority populations represented in the training data. In this study, we evaluated whether supervised domain adaptation methods--including balanced weighting, two-stage TrAdaBoost, feature augmentation with SrcOnly prediction, and linear interpolation--can mitigate these biases. Using the ABCD dataset, we assessed whether models trained on 80 MRI measures from White American participants could generalize more effectively to African American participants. All domain adaptation methods reduced prediction error for African American participants, particularly for MRI modalities with large baseline disparities (e.g., structural MRI), while offering limited improvements where initial gaps were smaller (e.g., functional connectivity). Among the approaches, balanced weighting performed best and remained stable and beneficial even when only 10 African American participants were used to adapt the original model trained exclusively on White American participants. These findings suggest that simple, low-cost strategies can effectively reduce cross-ethnic performance gaps and improve equity in predictive neuroimaging, offering a practical path forward for future neuroimaging predictive biomarkers. Significant StatementLarge-scale neuroimaging datasets increasingly enable machine-learning models to predict cognitive and clinical outcomes; however, these datasets are often ethnically/racially imbalanced. As a result, predictive models tend to generalize poorly to underrepresented populations. We demonstrate that, across 80 MRI phenotypes, a class of machine-learning approaches collectively known as supervised domain adaptation can substantially reduce cross-ethnicity disparities in neuroimaging-based cognitive prediction, even when only limited data from underrepresented groups are available. Among the methods evaluated, balanced weighting achieved the best performance while maintaining low computational cost. Together, these findings provide a practical and scalable framework for improving fairness and generalizability in neuroimaging-based machine learning under realistic conditions of ethnic/racial imbalance.

Generalizable neuroimaging biomarkers that detect cerebral cortical changes after traumatic brain injury (TBI) and predict patient outcomes are needed to improve care and to develop targeted therapies. We used morphometric inverse divergence (MIND) analysis of structural MRI to investigate cortical gray matter morphological networks cross-sectionally and longitudinally after TBI and correlate these with symptoms, disability and cognition six months after injury. Our findings support the Triple Network Model from functional MRI of post-traumatic alterations in the relationship between task-positive, default mode and salience networks. However, the strongest associations between early cortical similarity metrics and long-term patient outcomes involved the dorsal attention network and the limbic network as well as similarity metrics across Mesulam's hierarchy of laminar differentiation. Since MIND mapping of cortical gray matter networks only requires data that is a routine part of standard clinical MRI protocols and does not need image harmonization across different scanners, this work reports a promising new tool that is immediately available for advancing research and clinical care in TBI.

Background: Persistent neurological and cognitive symptoms following SARS-CoV-2 infection point to long-term alterations in brain structure and function. The thalamus, orbitofrontal cortex, and limbic networks are particularly susceptible to inflammatory and neurovascular stressors. However, the relationship between cortical, white-matter, and thalamocortical alterations in post-COVID syndrome remains unclear. Methods: 76 COVID-19 recovered participants (CRPs) and 51 healthy controls (HCs) underwent multimodal MRI comprising T1-weighted structural, diffusion, and resting-state functional acquisitions. Grey-matter morphology was assessed using voxel-based morphometry (VBM), white-matter microstructure using tract-based spatial statistics (TBSS), and thalamocortical functional connectivity (TC-FC) using seed-based analyses from major thalamic nuclei. Results were evaluated both across the groups (HC vs. CRP) and after stratifying CRPs by hospitalisation status (HC vs. Non-hospitalized patients (NHPs) vs. Hospitalized patients (HPs)). Results: No group-level grey-matter differences were observed between HCs and CRPs; however, HPs showed localized volume loss in the orbitofrontal and frontal-pole cortices (pFWE < 0.05). TBSS revealed widespread microstructural abnormalities, including reduced fractional anisotropy and mean diffusivity across association and commissural tracts (pcorr < 0.05), with regional increases in mode of anisotropy indicating selective loss of crossing fibres (pcorr < 0.05). Resting-state analyses revealed increased TC-FC from the mediodorsal thalamic nucleus to anterior cingulate, parietal, and occipital cortices (pcorr < 0.05), while differences in pulvinar and ventrolateral nuclei were not significant (pcorr > 0.05). Conclusions: Our findings indicate that COVID-19 recovery is associated with enduring alterations in fronto-limbic and thalamo-cortical circuits, most prominently in individuals with severe infection. Convergent structural and functional changes involving the orbitofrontal cortex and mediodorsal thalamus suggest network-specific reorganisation that may underpin persistent cognitive and affective symptoms of post-COVID syndrome.

Regions within ventral occipito-temporal cortex exhibit category-selective BOLD responses during episodic encoding and retrieval of visual information. How these regions interact with other brain areas during successful encoding and retrieval, and whether these interactions relate to memory performance, remains unclear. The present study examined category-selective functional connectivity using psychophysiological interaction (PPI) analyses in younger and older adults during the encoding and retrieval of word-image associations. Seed regions comprised three scene-selective regions - the parahippocampal place area, medial place area, and occipital place area - and one object-selective region, the lateral occipital complex (LOC). During encoding, scene-selective regions exhibited greater connectivity with posterior occipital and occipitotemporal regions during scene relative to object encoding, whereas the LOC exhibited less extensive connectivity with similar posterior regions during object encoding. During retrieval, both scene- and object-selective regions demonstrated increased connectivity with left lateral prefrontal and parietal cortices during the retrieval of their preferred category. Age differences in scene-selective connectivity were evident at both phases. Moreover, associations between source memory performance and scene-selective connectivity were significant only in younger adults. These findings suggest that scene- and object-selective regions exhibit convergent patterns of functional connectivity during encoding and retrieval which, for scenes, vary with age.

Brain similarity networks (BSNs), extracted from structural magnetic resonance imaging, provide a validated framework for studying brain network organization and encode neurodevelopmental information relevant for psychiatric disorders. Recently, a neurodevelopmental hypothesis has been proposed for bipolar disorder (BD), where evidence demonstrates neuroprogression phenotypes differing from controls. BSNs offer a promising framework for investigating BD's neural correlates but remain largely underexplored. Parallelly, graph neural networks (GNNs) have emerged as suitable deep learning models for exploiting network-level information. This study aimed to investigate BSNs for discriminating subjects with BD from controls within a GNN framework using the multi-site StratiBip network, composed of 605 controls and 501 subjects with BD. Leveraging advanced analysis tools, we developed a multi-site classification framework including: i) the state-of-the-art MIND algorithm for computing morphometric similarity (MS) networks based on gray matter volumes (GMV), ii) MS integration with age, sex, and GMV, iii) a leave-one-site-out cross-validation for multi-site model generalizability evaluation. The best model achieved a mean multi-site accuracy of 68%. Explainability analyses revealed meaningful MS patterns in the basal ganglia, frontal and temporal lobes, and a particularly relevant integration with age. This study provides interpretable insights into the role of MS in BD and unveils evidence supporting ageing-related processes as a significant component of BD pathophysiology.

Inter-individual variability poses a significant challenge in decoding brain activity across subjects. While standard anatomical registration procedures reduce morphological differences, they fail to capture functional variability between subjects. Functional alignment methods address this issue by establishing voxel-to-voxel correspondences between pairs of individuals, thereby constructing a shared functional space. This shared space can be extended at the group level by generating a functional template. However, despite the availability of toolboxes, functional templates remain underused in fMRI analysis. Adopting this approach is currently difficult due to the diversity of existing methods and the lack of clear guidelines. Comprehensive evaluations of functional templates are limited to movie-watching paradigms. Here, we extensively compare functional alignment methods (Optimal Transport, Procrustes, Ridge regression, and Shared Response Model) and template construction strategies (in-sample, out-of-sample, pairwise) within the more general framework of task decoding. In this framework, decoding accuracy measures how well individual activation patterns align. Across multiple tasks and datasets, we demonstrate that population templates built using Optimal Transport (a) yield the highest decoding accuracy, (b) are not significantly biased by individual subjects, which facilitates generalization to new subjects, and (c) preserve the cortical signal topography.

Conventional connectome edge weights, such as number of streamlines (NOS) or diffusion tensor imaging (DTI) metrics, lack specificity to microstructural details which may hold relevance for macroscale brain organisation. Since biophysical diffusion modelling offers greater specificity to microstructure, we investigated whether parameters from the Standard Model of diffusion in white matter provide informative alternatives for connectome weights - namely the intra-axonal signal fraction (f) and perpendicular extra-axonal diffusivity [Formula], as proxies of axonal density and myelination, respectively. Using diffusion MRI data from healthy adults, we constructed structural networks at four parcellation scales, weighted by f, [Formula], NOS, fractional anisotropy (FA) and radial diffusivity (RD). While all weights reproduced expected small-world properties, only [Formula] and normalised NOS captured non-random properties of local organisation across all spatial scales. We then correlated each weighted connectome with resting-state fMRI functional connectivity and intracranial measurements of conduction velocity. At the whole-brain level, although NOS gave strongest coupling with fMRI functional connectivity, only [Formula] exhibited significant structure-function coupling across all spatial scales and modalities. At the regional level, [Formula] and RD gave highest consistency in structure-function coupling across spatial scales. Thus, connectome weights derived from [Formula] capture meaningful aspects of brain network organisation with functional relevance.

A practical barrier in multimodal neuroimaging is that structural MRI, fMRI, and EEG are often analyzed in modality-specific spaces or reduced to atlas- and sensor-based summaries, limiting the construction of common, interpretable subject-level brain signatures. We evaluate cortical Laplace-Beltrami eigenmode coordinates as a shared geometry-aligned language for structural MRI (sMRI), resting-state fMRI (rs-fMRI), and EEG. In this framework, sMRI morphometric fields are represented by cortical eigenmode coefficients, rs-fMRI by covariance among eigenmode time-series coefficients, and EEG by mode-frequency-condition summaries. Using the Max Planck Institute Leipzig Mind-Brain-Body dataset (MPI-LEMON), we compared unimodal eigenmode-coordinate summaries, multimodal cortical eigenmode-coordinate PCA, conventional atlas/sensor-based PCA and ridge representations, and geometric eigenmode multiview factorization (GEMF). GEMF is a structured decomposition that preserves the modality-native organization of the data objects while separating shared from modality-specific variation. Eigenmode-coordinate representations yielded compact subject-level signatures with strong external validity for chronological age and a secondary cognitive outcome. Multimodal eigenmode-coordinate PCA was among the strongest-performing approaches, reached high age-prediction performance at moderate dimension, and consistently outperformed conventional low-dimensional PCA. GEMF selected an even lower-dimensional shared representation and remained competitive with the benefit of providing interpretable shared and modality-specific factors. These findings support cortical eigenmode coordinates as a practical foundation for compact, interpretable, and multimodally aligned subject-level brain signatures.