Aligning neural activity across subjects offers the promise of discovering shared computational principles and generalizable decoders. However, traditional alignment methods require shared stimuli across subjects, a constraint that limits applicability to naturalistic paradigms with limited or non-overlapping data. We introduce a Multi-Encoder-Decoder Variational Autoencoder (MED-VAE) that achieves cross-subject alignment without shared stimuli by anchoring representations to a common scaffold provided by a pretrained ANN. Using the Natural Scenes Dataset, we show that MED-VAE creates common latent spaces with superior semantic organisation, achieving higher cross-subject alignment than common methods while maintaining robust generalisation to held-out stimuli where traditional methods degrade. Reconstructing from these common spaces back to each subject's original neural space, MED-VAE preserves equal stimulus-driven signal in its cross-subject latent space. Finally, we show that this superior alignment directly enables cross-subject neural prediction, as demonstrated via cross-subject image decoding. In summary, we introduce a framework to identify generalisable common subspaces for cross-subject predictions and downstream tasks, demonstrated here for visual cortex responses to static images.

Emerging reports suggest that sample sizes commonly used in functional neuroimaging studies may be too small to detect many brain-behavior relationships, posing a major barrier to brain and mental health research. A central challenge is that planning robust studies requires researchers to know what effect sizes to expect, yet this essential information is surprisingly difficult to estimate in practice and thus often omitted from study planning. Critically, standard "mass univariate" procedures for estimating effects across multiple brain areas give an inflated picture of how large effect sizes are. Here, we introduce a method to correct this inflation bias and perform an unprecedented analysis of 63 studies in seven large datasets (n = 100-40,000; 52,979 total participants) to establish effect size benchmarks in functional neuroimaging. We find that between-subjects effects are exceedingly small at the majority of brain areas (Cohen's ∣d∣ < 0.2), requiring consortium-level sample sizes to detect even some of the strongest focal brain effects (n > 500 at 80% statistical power with FDR correction). However, multivariate analyses and within-subject task designs yield substantially larger effect sizes that can be detected at sample sizes within reach of individual labs (n < 50). By establishing data-driven effect size benchmarks, these findings lay the groundwork for more informed study planning in neuroscience while highlighting shared challenges (and the potential for shared solutions) across biomedicine.

OBJECTIVES:To investigate the relationship between alcohol consumption and dementia. DESIGN:Prospective cohort and case-control analyses combined with linear and non-linear Mendelian randomisation. SETTING:Two large-scale population-based cohorts: the US Million Veteran Programme and the UK Biobank. Genetic analyses used summary statistics from genome-wide association studies (GWAS). PARTICIPANTS:559 559 adults aged 56-72 years at baseline were included in observational analyses (mean follow-up: 4 years in the US cohort; 12 years in the UK cohort). Genetic analyses used summary data from multiple large GWAS consortia (2.4 million participants). MAIN OUTCOME MEASURES:Incident all-cause dementia, determined through health record linkage, and genetic proxies. RESULTS:During follow-up, 14 540 participants developed dementia and 48 034 died. Observational phenotype-only analyses revealed U-shaped associations between alcohol and dementia risk: higher risk was observed among non-drinkers, heavy drinkers (>40 drinks per week; HR 1.41, 95% CI 1.15 to 1.74), and those with alcohol use disorder (AUD) (HR 1.51, 95% CI 1.42 to 1.60) compared with light drinkers. In contrast, Mendelian randomisation genetic analysis identified a monotonic increase in dementia risk with greater alcohol consumption. A 1 SD increase in log-transformed drinks per week was associated with a 15% dementia increase (inverse-variance weighted (IVW) OR 1.15, 95% CI 1.03 to 1.27). A twofold increase in AUD prevalence was associated with a 16% increase in dementia risk (IVW OR 1.16, 95% CI 1.03 to 1.30). Alcohol intake increased dementia, but individuals who developed dementia also experienced a decline in alcohol intake over time, suggesting reverse causation-where early cognitive decline leads to reduced alcohol consumption-underlies the supposed protective alcohol effects in observational studies. CONCLUSIONS:These findings provide evidence for a relationship between all types of alcohol use and increased dementia risk. While correlational observational data suggested a protective effect of light drinking, this could be in part attributable to reduced drinking seen in early dementia; genetic analyses did not support any protective effect, suggesting that any level of alcohol consumption may contribute to dementia risk. Public health strategies that reduce the prevalence of alcohol use disorder could potentially lower the incidence of dementia by up to 16%.

Threshold-free cluster enhancement (TFCE) is widely used for cluster-based inference in neuroimaging, but existing implementations typically rely on discretized approximations that may introduce numerical variability. We present eTFCE, an efficient framework that provides a numerically exact evaluation of the TFCE integral using an optimized cluster retrieval algorithm. Across multiple datasets, eTFCE and the standard implementation produce highly consistent inference results. Voxel-wise comparisons reveal a systematic asymmetry: the standard method yields smaller p-values for more voxels, while eTFCE concentrates stronger statistical evidence within a smaller subset. These differences are primarily confined to voxels near the inference boundary and have minimal impact on overall inference. This pattern is consistent with discretization effects in standard implementations, where the TFCE integral is approximated using a finite set of threshold levels, introducing subtle biases in statistical evidence accumulation across thresholds. Furthermore, eTFCE improves computational efficiency (71.3

As neuroimaging analysis shifts toward large-scale, multi-site studies, managing the unwanted variability introduced by combining heterogeneous datasets has become a critical challenge. Although tools such as ComBat and its neuroimaging extensions are widely used to address this variability, they only permit the modeling of categorical site effects and cannot account for continuous sources of confounding, such as image quality, head motion, and acquisition parameters. We introduce ComCat, an extension of the ComBat framework that preserves biologically relevant covariates while removing the effects of categorical site indicators and continuous nuisance variables. The latter are modeled as smooth nonlinear functions via B-spline basis expansion. ComCat is applicable to a broad range of brain analysis tasks, including voxel- and surface-based morphometry, normative modeling, and machine learning-based prediction. To demonstrate its capabilities, we evaluated ComCat on brain age prediction across five datasets covering complementary multi-site harmonization scenarios: ON-Harmony (10 subjects x 6 scanners; n = 80); the Buchert traveling-phantom dataset (1 subject x 116 scanners; n = 531); the Tohoku single-scanner, varying-acquisition dataset (n = 121); MR-ART (148 subjects with varying motion levels); and an ABIDE subset comprising 229 control subjects and 208 individuals with autism spectrum disorder across 14 scanners. Using image quality measures derived from CAT12 as continuous nuisance variables, ComCat reduced the mean absolute error (MAE) in brain age prediction relative to ComBat-GAM in all five datasets, including the two scenarios where site information was unavailable or uninformative. In the ABIDE dataset, ComCat improved harmonization while preserving the difference between the control and ASD groups, demonstrating that scanner-related variance can be removed without affecting biologically meaningful signals. ComCat can operate with or without site labels and is agnostic to the source of image quality metrics.

Functional neuroimaging aims to uncover brain processes underlying behavior and disease, yet studies are often underpowered to detect these effects. How this literature has shaped our understanding of brain function remains unknown, and little guidance exists for planning better powered studies. An underappreciated barrier is that commonly reported effect sizes across the brain are inflated, biasing study planning. Here, we introduce a correction for this inflation bias and show how more accurate studies can be planned using corrected effect size benchmarks from a mega-analysis of 63 typical studies across seven large datasets (52,979 participants). We find that common methods of planning studies based on uncorrected effects lead to roughly half the expected detections at typical sample sizes, with limited spatial overlap with original findings. These missed effects collectively explain meaningful additional variance in the desired outcome. We show how to recover missed effects by planning not only for power but also for a target number of detections via corrected benchmarks, or by taking a whole-brain approach with multivariate effects that individual research groups can detect (n < 50 compared to n > 1,000 for a typical univariate effect). These findings lay the groundwork for more informed study planning and a richer understanding of the widespread nature of brain effects, with implications for shared challenges (and solutions) across biomedicine.

Introduction Alcohol consumption is an increasingly recognised modifiable risk factor for dementia, yet whether it has differential impacts on dementia subtypes and its role in disease progression remains unclear. This study aims to: (1) quantify the association between alcohol intake and incidence of dementia subtypes and (2) examine whether individuals who drink heavily and develop dementia referred to hereafter as ‘alcohol-related’—have poorer post-diagnosis outcomes compared with other dementia cases. Clarifying these relationships will determine whether alcohol selectively increases risk for specific dementia phenotypes or broadly heightens neurodegenerative vulnerability, with implications for prevention, clinical counselling and therapeutic targeting.Methods and analysis This population-based cohort study of alcohol and dementia will use linked UK electronic health records from Clinical Practice Research Datalink, Hospital Episode Statistics and Office for National Statistics (ONS). Participants will be eligible if they have available linked data from January 1998, when ONS death registrations became available, until the end of follow-up. Alcohol exposure will be defined through self-reported recorded weekly alcohol units and diagnostic codes for harmful or dependent alcohol use. Primary outcomes including incident all-cause and subtype-specific dementia (eg, Alzheimer’s, vascular, Lewy body, Parkinson’s, frontotemporal) as well as secondary outcomes (ie, mortality, care-home entry and neuropsychiatric symptoms). Key covariates encompassing socio-demographic factors, smoking and relevant comorbidities will be adjusted for. Multivariable Cox proportional hazards and Fine-Gray competing risk models will estimate associations with dementia incidence. Post-diagnosis prognosis will be compared for dementia in individuals with a history of heavy alcohol use (‘alcohol-related’) and dementia in individuals with minimal alcohol exposure (‘non-alcohol-related’) cases using survival and logistic regression models. Multiple testing correction will be applied across dementia subtype comparisons. Alcohol exposure will be modelled continuously and non-linearly using restricted cubic splines and categorically using binary indicators of harmful/dependent use. Missing covariate data will be assessed and addressed using appropriate methods, including multiple imputation and complete-case analysis. Data extraction and analysis are scheduled from October 2025 to October 2026.Ethics and dissemination Use of de-identified routine data will proceed under existing Research Ethics Committee and data governance approvals. Findings will be disseminated via open-access peer-reviewed journals, academic conferences and summaries targeted at patient, public and policy audiences. The results of this study will be reported according to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) and The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) guidelines.

Bayesian Image-on-Scalar Regression (ISR) provides flexible, uncertainty-aware neuroimaging analysis. However, applying ISR to large-scale datasets such as the UK Biobank is challenging due to intensive computational demands and the need to handle subject-specific brain masks rather than a common mask. We propose a novel Bayesian ISR model that scales efficiently while accommodating these inconsistent masks. Our method leverages Gaussian process priors with salience area indicators and introduces a scalable posterior computation algorithm using stochastic gradient Langevin dynamics combined with memory mapping. This approach achieves linear scaling with subsample size and constrains memory usage to the batch size, facilitating direct spatial posterior inferences on brain activation regions. Simulation studies and analysis of UK Biobank task fMRI data (38,639 subjects; over 120,000 voxels per image) demonstrate a 4- to 11-fold speed increase and an 8-18% enhancement in statistical power compared to traditional Gibbs sampling with zero-imputation. Our analysis reveals a subregion of the amygdala where emotion-related brain activation decreases by approximately 58% between ages 50 and 60. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work.

Nociplastic pain is defined by altered nociceptive processing in the absence of clear peripheral damage or somatosensory lesions. The Fibromyalgia Index (FMI), derived from the 2016 diagnostic criteria, is increasingly used as a marker of nociplastic pain severity in clinical studies, yet its neurobiological validity remains untested at scale. Using multimodal neuroimaging data from over 40 000 participants in UK Biobank, we examined whether FMI scores were associated with altered functional and structural connectivity within the descending pain modulatory system (DPMS), a brain network involved in endogenous pain control and implicated in nociplastic pain conditions. Functional connectivity was assessed using resting-state functional MRI (rfMRI), and structural connectivity using diffusion-weighted MRI (dMRI) tractography. Connectivity was quantified between seven DPMS regions: periaqueductal grey (PAG), rostral ventromedial medulla (RVM), hypothalamus, amygdala, rostral and subgenual anterior cingulate cortex (rACC, sgACC), and dorsolateral prefrontal cortex (dlPFC). Multi-group structural equation models tested associations between FMI scores and connectivity, stratified by chronic pain status. Mediation models evaluated which aspects of nociplastic pain accounted for the observed associations: widespread pain and SPACE symptoms (sleep disturbance, pain, affect, cognitive problems, and low energy). To assess specificity, we repeated analyses using the Douleur Neuropathique 4 (DN4), a measure of neuropathic pain, and average pain intensity as comparison outcomes. In 22 139 individuals with chronic pain (58% female; mean age 64.8, standard deviation 7.59), FMI scores were associated with altered structural connectivity between the PAG and amygdala [β = 0.023, 95% confidence interval (CI): 0.0087 to 0.039; Pcorr = 0.0125] and between the PAG and hypothalamus (β = -0.029, 95% CI: -0.043 to -0.015; Pcorr = 0.0013). Functional connectivity in the same circuits showed smaller effects. These associations were not observed in individuals without chronic pain. Mediation analyses revealed that PAG-amygdala and PAG-hypothalamus connectivity were partially explained by fatigue, sleep duration, and widespread pain. DPMS connectivity was not significantly associated with neuropathic pain or average pain intensity. These findings suggest that FMI scores reflect biologically meaningful changes in brain connectivity, particularly in subcortical DPMS circuits implicated in affective and homeostatic dimensions of pain. Structural connectivity was more strongly associated with FMI than functional measures, possibly reflecting cumulative effects of chronic pain on white matter architecture. The absence of similar associations for other pain outcomes supports the specificity of FMI as a marker of nociplastic pain severity. These results provide a neurobiological basis for the FMI and support its use in population research and biomarker development for nociplastic pain.

BACKGROUND:Cognitive impairments are common in depression and often persist beyond mood resolution. However, the relationship between cognitive performance, its neurological underpinnings, and future depression risk is unclear, limiting strategies for primary and secondary prevention. OBJECTIVE:Our objective was to determine whether cognition associates with subsequent depression, both relapse and first-episode occurrences. METHODS:1862 UK Biobank participants with a history of International Classification of Diseases (ICD)-10-defined depression in remission (RD) (mean (SD) age: 52.7 (7.13) years) were age-matched and sex-matched to 1862 participants without depression history or current antidepressant use. Cognitive scores were compared between groups at the composite (z-score), domain and task levels. MRI-derived phenotypes assessed brain network structure and functional connectivity. Longitudinal associations with future depression were assessed using logistic regression models and a Cox proportional hazards model controlling for key confounders. FINDINGS:Participants with RD had a higher risk of future depression (33%) than controls (13%), including when we accounted for temporal differences in longitudinal assessment (HR=3.16 (95% CI 2.71 to 3.67), global proportional hazard assumption p=0.07). Composite cognitive performance in controls was inversely associated with future depression risk (risk estimated marginal means: 0.25% at -1SD, 0.20% at mean, 0.15% at +1 SD). In RD, this relationship was reversed (0.74% at -1SD, 0.90% at mean, 1.10% at +1 SD). Executive functioning, processing speed and reasoning task scores all contributed. Higher grey matter in default mode network regions was associated with better concurrent cognitive performance across all participants, but not with future depression risk. Other MRI findings were limited. CONCLUSIONS:RD carried a threefold higher risk of future depression than controls. Cognitive performance was a risk marker for future depression in both groups but in opposing directions. Neuroimaging metrics provided little predictive value. CLINICAL IMPLICATIONS:Personalised risk factor assessment for depression is likely to be dependent on depression history. Those without previous history of diagnosed depression are at higher risk of future depression when cognitive performance is lower at baseline. RD is a high-risk group for future depression, and those with relatively higher cognitive performance may be more likely to report future depressive symptoms.

Neuroimaging has profoundly enhanced our understanding of the human brain by characterizing its structure, function, and connectivity through modalities like MRI, fMRI, EEG, and PET. These technologies have enabled major breakthroughs across the lifespan, from early brain development to neurodegenerative and neuropsychiatric disorders. Despite these advances, the brain is a complex, multiscale system, and neuroimaging measurements are correspondingly high-dimensional. This creates major statistical challenges, including measurement noise, motion-related artifacts, substantial inter-subject and site/scanner variability, and the sheer scale of modern studies. This paper explores statistical opportunities and challenges in neuroimaging across four key areas: (i) brain development from birth to age 20, (ii) the adult and aging brain, (iii) neurodegeneration and neuropsychiatric disorders, and (iv) brain encoding and decoding. After a quick tutorial on major imaging technologies, we review cutting-edge studies, underscore data and modeling challenges, and highlight research opportunities for statisticians. We conclude by emphasizing that close collaboration among statisticians, neuroscientists, and clinicians is essential for translating neuroimaging advances into improved diagnostics, deeper mechanistic insight, and more personalized treatments.

The underlying mechanisms for neurodegeneration in multiple sclerosis are complex and incompletely understood. Multivariate and multimodal investigations integrating demographic, clinical, multi-omics, and neuroimaging data provide opportunities for nuanced analyses, aimed to define disease progression markers. We used data from a 12-year longitudinal cohort of 88 people with multiple sclerosis, to test the predictive value of multi-omics, MRI, clinical examinations, self-reports on quality of life, demographics, and general health-related variables for future functional and cognitive disability. Progressive functional loss beyond an Expanded Disability Status Scale score≥4 was used to define a functional loss group. A cognitive decline group was defined by a ≥25% decrease from the maximum (cognitive) Paced Auditory Serial Addition Test score. We used a multiverse approach to identify which baseline variables were most predictive for functional and cognitive loss group memberships, independent of analysis bias. We identified several factors predicting an increased risk of future functional loss (FLG) and cognitive decline groups (CDG) within the next 12 years from baseline: functional score (0-10, median Odds Ratio per baseline unit increase [mORFLG=2.15± 0.51; mORCDG=2.46± 1.60]), cognitive scores (1-60 [mORFLG=0.98± 0.03; mORCDG=0.91± 0.06]), the number of previous relapses [mORFLG=1.56± 0.26; mORCDG=1.44± 0.60], serum vitamin A levels (umol/l [mORFLG=0.92± 0.06; mORCDG=0.33± 0.36]), self-reported mental health (1-100 [mORFLG=0.96± 0.02; mORCDG=0.91± 0.09]) and physical functioning (1-100 [mORFLG=0.99± 0.01; mORCDG=0.97± 0.03]). Our results suggest that clinical assessment of physical function and cognition, self-reported mental health, and potentially vitamin A levels are the best predictors for risk-group stratifications of people with MS at baseline. While these findings are promising, we also want to underscore the observed analysis-choice induced variability which necessitates both an increase in transparency when reporting study findings as well as strategies which are robust to the many researcher degrees of freedom. ### Competing Interest Statement OAA has received a speaker's honorarium from Lundbeck, Janssen, Otsuka and Lilly, and is a consultant to Coretechs.ai and Precision Health. LTW is a minor shareholder of baba.vision. KMM has served on scientific advisory board for Alexion, received speaker honoraria from Biogen, Novartis, Roche and Sanofi, and has participated in clinical trials organized by Biogen, Merck, Novartis, Otivio, Roche and Sanofi. EAH received honoraria for advisory board activity from Sanofi-Genzyme, and his department has received honoraria for lecturing from Biogen and Merck. OT received speaker honoraria from and served on scientific advisory boards of Biogen, Sanofi-Aventis, Merck, and Novartis, and has participated in clinical trials organized by Merck, Novartis, Roche and Sanofi. The remaining authors declare no other competing interests. ### Funding Statement This project was funded by the Norwegian MS-union (no reference). Model training was performed on the Service for Sensitive Data (TSD) platform, owned by the University of Oslo, operated and developed by the TSD service group at the University of Oslo IT-Department (USIT). Computations were performed using resources provided by UNINETT Sigma2 (#NS9666S) - the National Infrastructure for High Performance Computing and Data Storage in Norway, supported by the Norwegian Research Council (#223273). ### Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Yes The details of the IRB/oversight body that provided approval or exemption for the research described are given below: The study was approved by the Norwegian Regional Committees for Medical and Health Research Ethics (REK, 814351). The OFAMS-study and the 10-year follow-up were previously approved by REK (2016/1906) and registered as clinical trial (clinicaltrials.gov identifier: [NCT00360906][1]). Ethical approval for the different brain age training datasets was obtained (REK 567301, PVO 17/21624), as well as for the longitudinal validation set (Bergen Breakfast Scanning Club, REK 238310), and the cross-sectional MS data (REK 2011/1846, REK 2016/102). All participants gave their written informed consent according to the Declaration of Helsinki. I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals. Yes I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Yes I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable. Yes Brain age model training data are available from the respective websites of the databases either openly or after application (see supplemental material). The main study data (OFAMS data) can be shared after receiving a new ethics approval upon reasonable request to the authors. Brain age models and analysis code are freely available at https://github.com/MaxKorbmacher/OFAMS\_Brain\_Age. [https://github.com/MaxKorbmacher/OFAMS\_Brain\_Age][2] [1]: /lookup/external-ref?link_type=CLINTRIALGOV&access_num=NCT00360906&atom=%2Fmedrxiv%2Fearly%2F2025%2F02%2F12%2F2025.02.09.25321961.atom [2]: https://github.com/MaxKorbmacher/OFAMS_Brain_Age

Tensor-based morphometry (TBM) aims at showing local differences in brain volumes with respect to a common template. TBM images are smooth, but they exhibit (especially in diseased groups) higher values in some brain regions called lateral ventricles. More specifically, our voxelwise analysis shows both a mean-variance relationship in these areas and evidence of spatially dependent skewness. We propose a model for three-dimensional imaging data where mean, variance and skewness functions vary smoothly across brain locations. We model the voxelwise distributions as skew-normal. We illustrate an interpolation-based approach to obtain smooth parameter functions based on a subset of voxels. The effects of age and sex are estimated on a reference population of cognitively normal subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) data set and mapped across the whole brain. The three parameter functions allow transforming each TBM image (in the reference population as well as in a test set) into a normative map based on Gaussian distributions. These subject-specific normative maps are used to derive indices of deviation from a healthy condition to assess the individual risk of pathological degeneration.

How brain networks and cognition co-evolve during development remains poorly understood. Here, we use resting-state functional magnetic resonance imaging (rs-fMRI) and cognitive data at baseline and Year 2 of 2,949 individuals in the Adolescent Brain Cognitive Development (ABCD) Study to examine how stable and changing features of brain network organization predict cognitive development during early adolescence. We find that baseline resting-state functional connectivity (FC) more strongly predicts future cognitive ability than baseline cognitive ability. Models trained on baseline FC to predict baseline cognition generalize better to Year 2 FC and cognition, suggesting that brain-cognition relationships strengthen over time. Intriguingly, baseline FC outperforms longitudinal FC change in predicting future cognitive ability. One potential reason is the lower reliability of FC change compared to baseline FC: ICC = 0.24 vs. 0.56. However, reducing baseline FC's reliability by shortening scan duration only partially narrows the predictive gap, suggesting reliability alone cannot be the full explanation. Furthermore, neither baseline FC nor FC change meaningfully predicts longitudinal change in cognitive ability. We also identify converging and diverging predictive network features across cross-sectional and longitudinal models of brain-cognition relationships, revealing a multivariate twist on Simpson's paradox. Together, these findings suggest that during early adolescence, stable individual differences in brain functional network organization play a more critical role than dynamic changes in shaping future cognitive outcomes.