NeuroImage

Population-graph models aggregate a query subject with an unlabeled cohort, so their reported performance can depend on the evaluation regime rather than on predictive information alone. Evaluation context has two separable axes: the visibility of the cohort during training, and the context available at inference. We study the training-time axis under a protocol fixed before analysis, on an ABIDE-I cohort of 871 subjects (20 acquisition sites). Two population-graph implementations are compared: a learned/adaptive graph (EV-GCN) and a static weighted graph (Parisot-GCN). Within each implementation the cohort, connectivity features, outer splits and training and selection rules are held fixed, and only the information available to the graph varies. The primary estimand is the implementation-by-training-context interaction: the difference between the P0-to-P1 AUC changes of the two implementations ({Delta}{Delta}AUC). Across three preselected outer splits (666, 1024, 2024) the direction is consistent and positive, with observed interaction range 0.06285-0.13090; uncertainty is assessed separately within each split by paired subject and site-cluster bootstrap resampling and is reported per split. Splits are replication units: they are not pooled and no cross-split average is reported, so the result is not a claim of invariance to arbitrary splits. These findings tie reported population-graph performance to the evaluation information regime and motivate explicit reporting of cohort visibility. The contribution is methodological: a two-axis taxonomy of information regimes, preselected cross-split evidence that the sensitivity of reported performance to cohort context depends on the implementation, and a reusable reporting checklist.

The superficial white matter (SWM), a thin layer of white matter beneath the cortex, contains a complex arrangement of short- and long-range fiber systems that support communication between cortical regions. These fiber populations include subcortical U-shaped fibers (U-fibers) that predominantly follow a tangential trajectory along the gray matter-white matter interface and long-range fibers that extend more radially through the deeper white matter to connect distant cortical regions. Here, we developed a surface-based framework to dissociate radial and tangential fiber components within the human SWM and investigated their microstructural properties and functional relevance using in-vivo high-field MRI. These components were quantified using apparent fiber density (AFD), fractional anisotropy (FA), and mean diffusivity (MD). We further contextualized SWM fiber architecture by examining depth-dependent microstructural profiles and their relationship with normative patterns of cortical organization and hierarchical specialization. The anatomical assessment of the proposed fiber orientation separation was supported by tractography-based bundle analyses, while the anatomical accuracy of tractography reconstructions was confirmed through comparison with post-mortem white matter dissection data. To assess their functional relevance, the fiber predominance map derived from radial and tangential AFD was related to spatial patterns of functional variability across the cortex. Regions with greater radially oriented fiber contribution exhibited more gradual changes in functional similarity across geodesic distance, whereas tangentially oriented fiber-enriched regions showed more rapid local functional differentiation. These findings reveal distinct structural and functional signatures of radial and tangential SWM fiber components and provide a framework for investigating their contributions to cortical organization and functional specialization. SIGNIFICANCE STATEMENTThe human cerebral cortex is organized into specialized regions that support diverse functions, but the structural principles underlying this organization remain incompletely understood. The superficial white matter (SWM), a thin layer of white matter beneath the cortex, contains intermingled short- and long-range fiber systems that support communication between cortical regions. However, these fiber populations have remained difficult to characterize because they are closely intertwined within the SWM. By developing a reproducible framework for mapping and quantifying distinct SWM fiber systems, this work provides a foundation for systematic investigation of this complex compartment. Integration of multimodal datasets, together with open sharing of analytical tools and standardized features, promotes transparency, reproducibility, enabling future studies of how superficial connectivity supports brain organization.

Subanaesthetic ketamine alters the content of consciousness while leaving responsiveness intact. We asked whether this state can be decoded from single eyes-closed EEG epochs, and how spectral power and phase-based connectivity compare when used as features. Re-analysing openly available 62-channel EEG from ten participants (awake versus subanaesthetic ketamine), we trained classifiers under leave-one-subject-out cross-validation to discriminate the two conditions from log band power, weighted phase-lag index (wPLI) connectivity, and their concatenation. Band power decoded the ketamine state above chance (balanced accuracy 0.71), whereas wPLI connectivity computed on the same epochs was at chance (0.47), and combining the feature sets did not improve on power alone. The dissociation held across three classifier families and across spatial montages, and was not explained by the dimensionality of the connectivity feature space. To identify its source, we decomposed the per-feature drug effect into components shared across subjects and subject-specific. The ketamine effect on connectivity was large within individuals but largely subject-specific, and therefore not transferable to held-out subjects (shared fraction 0.05), whereas the spectral effect was substantially shared across subjects (0.53); both feature classes carried comparable individual identity, so the asymmetry reflects transferability rather than fingerprint-likeness. The spectral signature was also recoverable from a sparse electrode subset, as a five-channel lateral montage performed as well as the full array. Undirected phase connectivity thus fails to decode subanaesthetic ketamine not through insufficient spatial sampling, but because the connectivity drug effect is subject-specific in form.

Neural fingerprinting seeks to identify individuals based on measurements of brain activity, exploiting the fact that aspects of brain function unique to an individual remain stable across repeated scans. Magnetoencephalography (MEG) is a powerful technique for fingerprinting. However, most MEG studies have used conventional (SQUID-based) MEG technology and typically rely on data aggregated over time, overlooking the rich temporal dynamics available in MEG. Here, using SQUID-MEG and the more recently introduced OPM-MEG, we showed fingerprinting was possible within and between modalities using static (time-aggregated) features; this is consistent with previous work. We further asked whether fingerprinting was possible based on network dynamics, estimated using a canonical hidden Markov model (CHMM). Results showed that the CHMM-derived networks provide a better fit to SQUID data than to OPM data; however, this effect was small and to be expected given the canonical networks were trained on SQUID data. We further showed that fingerprinting was possible within and between modalities using CHMM-derived state activation time courses and state power spectral densities. However, state summary statistics and transition probabilities only supported within modality fingerprinting. Our findings suggest that subject-specific information is preserved within CHMM states and across MEG technologies. This represents an important step towards advancing our understanding of brain dynamics, and particularly the brain networks delineated by the CHMM. The study also supports the future use of the CHMM for processing and interpretation of OPM-MEG data.

The posterior parietal cortex integrates somatosensory information with signals supporting action, spatial cognition, and multisensory processing, yet the fine-scale topographic relationship between the intraparietal sulcus (IPS) and primary somatosensory cortex (S1) remains poorly understood. Here, we used resting-state fMRI in 10 awake rhesus macaques to characterize functional connectivity between four S1 subregions (areas 3a, 3b, 1, and 2) and 223 single-voxel seeds densely sampling an IPS-centered parietal territory. Across S1 subregions and hemispheres, IPS-S1 connectivity exhibited a highly structured spatial organization characterized by systematic anteroposterior and mediolateral gradients and a prominent semi-oval topology following the geometry of the intraparietal sulcus. Connectivity was generally stronger toward the cortical banks and convexities and weaker near the sulcal fundus, a pattern that could not be explained by reduced local temporal signal-to-noise ratio. Unsupervised hierarchical clustering independently recovered this spatial organization and identified connectivity-defined subdivisions that only partially corresponded to classical cytoarchitectonic boundaries. Along the S1 mediolateral axis, prominent troughs in cluster-averaged connectivity profiles corresponded to several previously described somatotopic transitions, allowing to consistently map putative boundaries between tongue and face representations and between hand and trunk-leg representations. Based on these boundaries, individual IPS clusters exhibited connectivity with multiple S1 body-part sectors, with the pattern of connectivity across these sectors varying across IPS clusters and S1 subregions. Together, our findings reveal that IPS-S1 functional connectivity is organized by a large-scale sulcal topology while preserving fine-grained features of S1 somatotopic organization. These findings suggest that the functional organization of the IPS is shaped by multiple spatial dimensions, including large-scale sulcal topology and the organization of S1 subregions and body-part representations, which may together contribute to the functional specialization of individual IPS regions.

Functional MRI provides a powerful way to characterize the architecture of the human brain. Yet individual brains often vary in the exact anatomic locations of functionally specific regions; as a result of this variability, traditional random-effects group analyses (RFX) often require unrealistically large sample sizes to converge onto stable results. Here, we explore the scaling patterns for an alternative fMRI analysis framework that accounts for inter-individual topographic variability while still capturing spatial similarities--group-constrained, subject-specific (GcSS) analysis. Using sixteen contrasts from four Human Connectome Project tasks (spanning language, social cognition, motor, and working memory), we first show that the GcSS approach yields much larger effect sizes than traditional RFX analyses. We then investigate the impact of sample size (N=10-450) on the estimation of three key GcSS outputs: (1) group probabilistic maps for the contrast of interest, (2) group-level parcels that serve as anatomical constraints for subsequent functional region of interest (fROI) definitions in individual subjects, and (3) effect sizes of fROI responses to task conditions. For most contrasts, the probabilistic maps show good reliability at N=100 (intraclass correlation, ICC=.75) and excellent reliability at N=200 (ICC=.90). Most group parcels also achieve substantial agreement at N=100 (Dice coefficient, DC=.80). Critically, once robust group parcels are established, the subject-specific portion of the analysis can proceed with much smaller sample sizes: even samples of N=10 participants yield accurate effect size estimates within subject-specific fROIs, detecting 80% of practically meaningful effects (effect size [≥] 0.2); with N=20, this increases to 90%. The scaling patterns we observed held not only in the cortex, but also in the cerebellum. Our results challenge the view that fMRI research requires large samples: once the broad region of interest is established (with N=100 or higher), fROI-based analyses can yield generalizable results with small sample sizes (N=10 or N=20).

Submillimeter functional magnetic resonance imaging (fMRI) based on blood-oxygenation-level-dependent (BOLD) signal enables the study of brain function at the submillimeter level, uncovering insights into fine-scale organisations like cortical layers and columns. However, its inherently low contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR) often limit its reliability and applicability. Noise Reduction with Distribution Corrected Principal Components Analysis (NORDIC PCA) is a locally low-rank denoising algorithm that reduces thermal noise levels in BOLD fMRI in a local patch manner. However, local patches often contain a mixture of signals from multiple tissues that negatively affects the low-rank structure of the patches, which limits the denoising capabilities of the algorithm. We propose an alternative approach for patch formation by gathering similar non-local voxels, dubbed voxel-matching (VM) NORDIC. The results on submillimeter-resolution BOLD fMRI data indicate that VM-NORDIC effectively promotes the low rank of the patches by boosting signal redundancy, allowing for more efficient noise attenuation. Moreover, the method barely affects spatial smoothness due to the non-local voxel selection based on time-series similarity. In particular, VM-NORDIC outperforms standard NORDIC with default local patching (Standard-NORDIC) in terms of temporal SNR (tSNR) (~9-90% larger than Standard-NORDIC; ~23-250% larger than the original) and spatial smoothness estimates (~20% of the smoothness induced by Standard-NORDIC). These improvements are fundamental to improving the validity and precision of fMRI studies at submillimeter resolutions.

Internal and external noise produce uncertainty in the neural representations of sensory stimuli. Uncertainty about basic stimulus features can be decoded from responses in sensory brain regions by estimating full probability distributions over stimulus features, rather than point estimates. Such decoded uncertainty correlates with subjective uncertainty reports, providing insight into the neural basis of metacognitive judgments. However, in humans, probabilistic decoding has only been applied to functional magnetic resonance imaging (fMRI) data, which has low temporal resolution and thus can give only limited insight into the dynamics of uncertainty in the brain. Here, we assessed whether probabilistic decoding of uncertainty about stimulus features could be extended to electroencephalography (EEG) data, which has higher temporal resolution but lower spatial resolution and different noise properties compared to fMRI. Participants performed a spatial location estimation task and provided subjective uncertainty reports. We found that time-resolved probabilistic decoding in EEG was feasible, as decoders produced accurate predictions of stimulus location following stimulus onset, and decoding error correlated trial-by-trial with decoded uncertainty. The choice of noise covariance structure critically impacted these metrics. Further, decoded uncertainty was a more reliable trial-by-trial indicator of stimulus information than decoding error, illustrating the advantages of probabilistic decoding over standard decoding approaches. However, uncertainty decoded from EEG had no significant trial-by-trial correlation with subjective uncertainty at any time point. Based on these results, uncertainty decoded from EEG provides a time-resolved estimate of stimulus information available from the brain signal on a single trial but may not relate to metacognitive reports.

Subcortical structures are central to cognition, emotion, and social processing, and their dysfunction underlies a wide range of neurological and psychiatric conditions. Yet subcortical fine-scale functional organization has remained largely inaccessible to group-level fMRI analysis: nuclei are small and low-SNR, and their functional topography varies idiosyncratically across individuals in ways that anatomical alignment cannot resolve. This idiosyncrasy carries real stakes. For example, deep brain stimulation targets with more variable functional connectivity across individuals show correspondingly less consistent clinical outcomes. Here we show that this seemingly idiosyncratic subcortical organization in fact reflects shared, fine-grained structure that can be recovered by modeling each individual's connectivity in a common functional space (connectivity hyperalignment, CHA), rather than relying on anatomical correspondence. Applying CHA to fMRI data collected during naturalistic movie viewing, we found that aligning subcortical voxels by their cortical connectivity, rather than their anatomical location, sharply increased cross-subject correspondence in both stimulus-evoked response and connectivity profiles. Critically, this alignment sharpened rather than blurred local structure: neighboring voxels became more functionally distinct from one another, and cortical connectivity targets became substantially more decodable across subjects. These results indicate that human subcortical nuclei possess a common fine-scale functional architecture that is masked by anatomical variability but recoverable at the individual level. This reframes a central methodological constraint in subcortical neuroimaging: rather than trading spatial resolution for signal quality via smoothing, group-level analyses can achieve voxel-level precision by aligning function rather than anatomy, opening a path to studying cortical-subcortical circuit organization in greater detail.

Magnetoencephalography (MEG) measures human neural activity non-invasively with spatio-temporal precision, and has been foundational in enabling impactful discoveries in cognitive neuroscience. New on-scalp MEG sensor technologies, such as OPM-MEG, offer the opportunity to capture more information about the neuronal magnetic fields with higher sensitivity to more complex, higher order spatial components, leading to improved source localization. The accuracy of MEG and OPM-MEG source localization relies on data preprocessing techniques to isolate the neuronal fields from other magnetic and biomagnetic sources through signal space separation, rejection, and suppression methods. Current preprocessing methods risk rejecting brain signals of interest, or can spread sensor noise artifacts unknowingly. Here we propose a novel preprocessing method for MEG, and test the extent to which it overcomes limitations of prior methods. Specifically, we derive, apply, and assess a novel signal space separation (SSS) method with Fosters inverse, a weighted matrix inversion protocol that can utilize information about the MEG sensor noise and artifacts to reconstruct neuronal activity. With simulations, phantom head cryogenic MEG recordings, and subject recordings with two OPM-MEG systems, we show that Fosters inverse with SSS offers a more robust and stable reconstruction of the neuronal magnetic fields, especially in the face of sensor noise and artifacts. As the field of cognitive neuroscience continues to embrace MEG and OPM-MEG, Fosters inverse with SSS offers a robust and powerful data preprocessing technique for reducing noise and improving source localization of the underlying neuronal currents.

The eigenvalue structure of resting-state fMRI (RS-fMRI) signals provides a compact representation of its variance-covariance structure, yet the information it encodes remains unclear. In this study, we introduce an eigenvalue-based framework to characterize RS-fMRI data using two features derived from the eigenspectrum: the log10-transformed first eigenvalue (log10({lambda}1)) and the slope of the log10-transformed spectrum ({beta}). We systematically evaluated these features across 14 denoising strategies using two independent datasets (China167 (83 males, 84 females; mean age +/- SD: 41 +/- 14 years old) and HCP1200 (425 males, 501 females; mean age +/- SD: 29 +/- 4 years old)). Specifically, we used singular value decomposition to obtain the eigenspectrum of the cortical BOLD signals after preprocessing and nuisance regression, and a linear model to parameterize it. We then examined the relationships between the eigenvalue parameters, head motion, and functional connectivity metrics across subjects and denoising strategies. Our key findings include: (1) log10({lambda}1) and {beta} strongly covaried with one another across subjects, denoising strategies, and datasets, indicating that they capture highly coherent aspects of the eigenspectrum; (2) both features were systematically influenced by denoising strategies; (3) within each denoising strategy, participants with greater log10({lambda}1) had greater mean framewise displacements (mFD), demonstrating sensitivity to residual motion effects, for all strategies in HCP1200 and 12/14 in China167; (4) across denoising strategies, mean log10({lambda}1) reflected the proportion of functional connectivity edges significantly associated with motion, indicating that higher log10({lambda}1) reflects more widespread motion-related contamination across large-scale functional networks; and (5) global signal regression consistently reduced log10({lambda}1), whereas spike regression had dataset-dependent effects. Together, these results establish eigenvalue signatures as robust and sensitive metrics for quantifying residual motion effects in RS-fMRI and provide a framework for evaluating denoising performance based on eigenvalue structure. The full procedure is implemented in R, and the corresponding script is available at: https://github.com/lejianhuang/EigenvalueSignature.

Diffusion MRI tractography enables noninvasive mapping of white matter fiber tracts. Atlas-based white matter parcellation supports automated tract identification by assigning individual streamlines to atlas-defined clusters and anatomical tract labels. Because clustering-based white matter atlases are constructed from cohort-specific tractography data, they capture common white matter organization represented in the atlas-construction population. Although major white matter anatomy is shared across populations, subtle population-related anatomical variability may influence atlas representation and test-retest correspondence. Therefore, the reproducibility and cross-population generalizability of tractography-based white matter atlases are important considerations for quantitative neuroimaging studies. In this study, we evaluated whether incorporating cross-population anatomical variability during atlas construction improves test-retest reproducibility. To do so, we compared a single-population ORG atlas constructed from a Western cohort with the cross-population East-West White Matter Atlas constructed from both Eastern and Western cohorts. Test-retest diffusion MRI scans from the Human Connectome Project Young Adult (HCP-YA) dataset and the Connectivity-based Brain Imaging Research Database (C-BIRD) were analyzed as independent Western and Eastern test-retest cohorts, respectively. Whole-brain tractography was reconstructed for each dMRI scan and parcellated using both atlases. Reproducibility was assessed using tract detection rate at both cluster level and anatomical tract level, weighted Dice coefficient, and the relative difference of mean fractional anisotropy (FA). Both atlases showed stable tract detection across test-retest scans in both cohorts. Compared with the ORG atlas, the East-West White Matter Atlas achieved higher overall spatial overlap and lower test-retest variability in mean FA, although atlas performance varied across individual tracts. These findings suggest that integrating cross-population information during atlas construction improves the reproducibility and generalizability of white matter atlas mapping across independent populations and imaging protocols.

Background: High-density electroencephalography (EEG) improves spatial sampling but entails greater acquisition cost, preparation time, participant burden, and computational demand. However, the extent to which electrode reduction affects successive levels of EEG analysis from global signal properties to cortical network inference remains insufficiently understood. Methods: We systematically subsampled two independent resting-state EEG datasets comprising 120 children recorded with 128 channels and 98 adults recorded with 256 channels. Density-dependent effects were assessed across 128-, 256-, 64-, 32-, and 19-channel configurations using a multilevel framework encompassing signal representation and quality, global and regional spectral power, EEG microstates, scalp-level effective connectivity, traveling waves, and source-level cortical hub topology. Results: Sensitivity to electrode reduction increased with the spatial specificity of the outcome. Global relative spectral power and explained variance were comparatively stable across configurations. In contrast, signal-to-noise ratio, entropy, reconstruction error, regional spectral distributions, traveling-wave estimates, effective connectivity, and microstate dynamics became progressively more density-sensitive. Although canonical microstate classes remained broadly identifiable at lower densities, their global explained variance, temporal parameters, and transition structure were altered. Source-level hub identity and centrality ranking showed the greatest vulnerability, with markedly reduced cross-density agreement at 32 and 19 channels. Across both datasets, several spatially dependent measures showed a transition around 64 channels, whereas estimates obtained with 128 and 256 channels were generally more similar. Conclusions: Electrode density affects EEG outcomes according to the spatial scale of the intended inference. Sparse montages may preserve broad global summaries but are less reliable for regional, spatiotemporal, and source-network analyses. Under the resting-state conditions examined here, approximately 64 channels represented a pragmatic transition for several spatially dependent measures. This value should be interpreted as context-specific guidance rather than a universal threshold.

Abstract Optically pumped magnetometer (OPM) sensors can be placed closer to the scalp than conventional superconducting quantum interference devices (SQUID), resulting in larger magnetoencephalography (MEG) signals from neuronal activity. For event-related sensor data, such as epileptogenic activity or sensory and motor evoked responses, however, OPMs and SQUIDs often differ less in signal-to-noise ratio (SNR) than in signal magnitude. We examined two factors contributing to the relative SNR: the dependence of the signal magnitude on source depth and the effect of scalp-to-sensor distance on the noise level. Simulated MEG data for a current dipole in a spherical head model confirmed that on-scalp sensor placement delivers the largest SNR gain for superficial sources. Depending on the relative overall noise level, there may be a crossover source depth at which SNR is equal for on-scalp and off-scalp sensors and beyond which off-scalp sensors achieve higher SNR. Analysis of the equal-SNR source depth in different-sized spherical head models indicated that, for a given relative noise level, the proportion of the brain where SNR is higher in OPM than in SQUID was larger in small head models, supporting the benefits of OPMs in pediatric studies. To experimentally evaluate noise contributions of brain and non-brain origin to the SNR, we recorded somatosensory evoked fields (SEFs) at varying scalp-to-sensor distances. Generally, both the evoked response magnitude and the noise level were lower when the sensors were further away from the scalp; consequently, the SNR depended less than the signal magnitude on the scalp-to-sensor distance. Comparison of power spectral densities (PSDs) at different sensor-to-scalp distances allowed us to identify whether the dominant noise source was of brain or non-brain origin at different frequency bands. Overall, the results highlight complementary properties of OPMs vs. SQUIDs in terms of SNR, which is of interest when optimizing MEG experiments for specific subject populations and brain regions.

Non-invasive, whole-brain neuroimaging methods such as functional magnetic resonance imaging, electroencephalography (EEG), and magnetoencephalography (MEG) are essential tools for studying the basis of human cognition in health and disease. MEG offers the opportunity to study neural activity at its intrinsic timescale, by recording the magnetic fields generated by electrical currents within the brain from outside the skull. Moreover, recently developed optically pumped magnetometers (OPMs) allow these recordings to take place in new settings, for example during naturalistic behaviour and in previously inaccessible populations. These breakthroughs have led to a shift in the neuroimaging landscape, with a global increase in the adoption of MEG. Crucially, however, the extent to which MEG recordings can measure different features of neural activity remains unclear. To address this issue, we leveraged a unique and rare dataset of concurrent MEG and intracranial EEG recordings from a cohort of epileptic patients. We found that group-level inferences of spontaneous oscillatory dynamics made with source-localised MEG, i.e. estimates of power and bursts, accurately reflected the underlying neural activity. As expected, the agreement was strongest for lower-frequency activity (delta, theta, and alpha) and superficial sources, and weakest in the gamma range. Crucially, however, MEG was also sensitive to deep structures: it captured oscillatory power and burst dynamics in the hippocampus, most robustly in the theta band. These findings demonstrate that MEG is sensitive to physiologically meaningful activity in cortical and subcortical regions and establish a foundation for the interpretation of future MEG studies across a wide range of research domains.

Amygdala responses are widely reported in BOLD fMRI studies of human emotions and mood disorders. Although modern acquisition methods have improved image quality, the hemodynamic nature of the BOLD signal continues to complicate interpretation of these responses. Across an array of both visual and auditory tasks, we here show that apparent amygdala BOLD signal is substantially confounded by signal from nearby deep veins, including the basal vein of Rosenthal. In contrast to prior work, we show that the amygdala, when engaged, is the leading but not exclusive contributor to this venous signal. Across the tested conditions, this signal was sensitive to, but not specific for, amygdala engagement. This confound mislocalizes apparent activation and prevents reliable attribution of signal to amygdala subnuclei with conventional BOLD fMRI. We further show that spatial smoothing exacerbates this mislocalization by merging peri-amygdalar venous signal with the anatomical amygdala. We reproduce our findings from precision imaging in a larger sample obtained from the Human Connectome Project. While broad conclusions about amygdala engagement remain largely valid when based on unsmoothed signal within the anatomical amygdala, further conclusions about subnuclear function drawn from conventional fMRI alone warrant considerable caution.

Estimating population-level dependence between brain regions is challenging when the number of regions greatly exceeds the number of individuals. We introduce an information-borrowing generalized Bayesian method for high-dimensional covariance estimation from region-level brain measurements. A divide-and-conquer parameterization separates marginal variances from correlation calibrations: an empirical \emph{pilot} correlation matrix--by default, obtained by standardizing a sample covariance matrix (SCM)--provides a reference structure that is adaptively calibrated through region-specific parameters shared across pairwise correlations, reducing the number of free parameters from $\mathcal{O}(d^2)$ to $\mathcal{O}(d)$ ($d$ is the number of regions) while guaranteeing a positive-definite covariance through the inclusion of a ridge term. Horseshoe priors on the calibration parameters enable selective borrowing and shrinkage of pilot information. Under suitable conditions on pilot accuracy, we establish theoretical guarantees for covariance estimation while retaining scalability in ultra-high-dimensional settings via a low-rank likelihood representation and a Woodbury-accelerated Hamiltonian Monte Carlo algorithm. Simulations showed lower estimation error than the SCM, bagging and iteratively refined principal orthogonal complement thresholding (POETRY). Applications to a semi-synthetic cortical-thickness and ADHD resting-state fMRI data revealed significant group differences in the default-mode--visual network and somatomotor network--limbic network blocks after Benjamini--Hochberg correction.

Iron is essential for neurophysiological health. Prior work has shown that subcortical tissue iron levels are the highest among all brain regions but also change over the lifespan. Recent functional MRI (fMRI) studies have provided evidence that an fMRI-derived iron-sensitive signal may index relative iron concentration in subcortical regions. In this context, we aimed to characterize the lifespan trajectory of this fMRI-derived iron-sensitive signal (relative iron quantification) in 10 subcortical regions and 14 thalamic nuclei from 881 healthy participants aged 12 to 88 years, using resting-state fMRI. Regressions were run with age as a predictor, separately for males (N = 429) and females (N = 452). The main results show non-linear effects of age, with an overall increase in iron levels across most subcortical regions within each sex. Six of the 14 thalamic nuclei in male participants had similar slopes of their lifespan iron levels, showing some homogeneity. The thalamic nuclei of female participants showed less agreement in slopes. This study provides new insight into the typical trajectory of iron levels across all major subcortical regions throughout the lifespan.

Structural similarity networks quantify the similarity of structural properties across cortical regions, providing a macroscopic window onto the organisation of cortical architecture. Morphometric inverse divergence (MIND) is a multivariate metric of similarity between cortical areas, based on the Kullback-Leibler (KL) divergence between areal distributions of multiple MRI features or morphometric variables locally measured at voxel or vertex resolution. MIND has demonstrated technical robustness and biological validity and is increasingly widely used as a measure of cortico-cortical similarity in clinical and developmental network neuroscience. Here we provide in-depth methodological background on KL divergence and MIND, highlighting possible sources of bias, critical user decision points in the design of a MIND processing pipeline, and recommendations for technical risk mitigation in using MIND as a metric of cortical similarity. We use simulated data and observational MRI datasets from adults (UK Biobank, N = 500 T1-weighted and diffusion scans) and neonates (Developing Human Connectome Project, N = 752 T2-weighted scans), to show how the estimator of KL divergence implemented in MIND is potentially influenced or biased by five properties of input MRI feature maps: (i) their smoothness; (ii) the proportion of identical values; (iii) analysis in native or common space and the choice of vertex mesh resolution; (iv) parcellation choice; and (v) covariance between input features. We offer principled and practical guidance for investigators wanting to specify and implement the MIND processing pipeline that is best suited to the constraints and opportunities of the MRI data available to them. These recommendations outline which pipeline steps should be used sparingly, such as vertex map smoothing; which should be used with informed caution, such as parcellation choice or vertex mesh resampling; and which could be newly implemented for more robust estimation of MIND, such as the use of principal component analysis to preprocess multivariate MRI features. To support further development of structural MRI similarity network analysis, and wider adoption of robust MIND methods, we also publish the code used to generate the results in this paper as an open resource.

Using a multimodal dataset including fMRI, EEG-fMRI and concurrent physiological recordings, we investigated the effect of denoising systemic low frequency oscillations (sLFOs) on the characterization of dynamic signatures of central autonomic regulation and their relation to ongoing physiological states. We demonstrated that the time-frequency profiles of couplings between BOLD time series and cardiac and respiratory processes were statistically comparable between regions of the brain associated with autonomic function and non-autonomic motor regions, suggesting that these couplings largely do not reflect neuronal activation related to autonomic activity. We further showed that model-based (via physiological response functions) and data-driven (via CompCor nuisance regressors extracted from cerebrospinal fluid) methods of sLFO denoising had a statistically similar effect on these frequency profiles. We novelly applied co-activation pattern analysis to assess state dynamics of autonomic-associated regions of the brain, finding that such brain states interrelate decreases in vigilance and increases in heart rate, respiratory flow, and head motion, thus providing evidence for global arousal processes affecting BOLD signal in autonomic-associated regions. Lastly, we modelled sliding-window dynamic functional connectivity within the central autonomic network (CAN) as modulated by heart rate variability, showing significant differences in model outcomes linked to each denoising pipeline. With these findings, we provide a comprehensive discussion of the implications for the application of denoising techniques to the CAN and comment on the origins of dynamic components of the BOLD signal linked to autonomic regulation, highlighting the global role played by arousal.