Brain Age Gap as a Predictive Biomarker: Linking Aging, Lifestyle, and Neuropsychiatric Health

Brain Age Gap as a Predictive Biomarker: Linking Aging, Lifestyle, and Neuropsychiatric Health | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Brain Age Gap as a Predictive Biomarker: Linking Aging, Lifestyle, and Neuropsychiatric Health Zhengxing Huang, Ruixia Zhang, Fan Yi, Junhang Zhang, Hongjing Mao, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6283338/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 24 Oct, 2025 Read the published version in Communications Medicine → Version 1 posted You are reading this latest preprint version Abstract Background The brain age gap (BAG), a neuroimaging-derived biomarker of accelerated brain aging, faces translational challenges due to model inaccuracies and unclear disease-mechanism linkages. We systematically evaluated BAG's clinical relevance across neuropsychiatric disorders, cognitive trajectories, mortality, and lifestyle interventions. Methods Using multi-cohort data (UK Biobank [n = 38,967], Alzheimer’s Disease Neuroimaging Initiative [ADNI; n = 1,402], Parkinson’s Progression Markers Initiative [PPMI; n = 1,182]), we developed a 3D Vision Transformer (3D-ViT) model for whole-brain age estimation. Survival analyses, restricted cubic splines, and stratified regressions assessed BAG’s associations with cognition, 16 neuropsychiatric disorders, and mortality. Lifestyle modulation effects were quantified through longitudinal BAG progression. Results The 3D Vision Transformer demonstrated robust predictive accuracy, achieving a mean absolute error (MAE) of 2.68 years in the UK Biobank cohort and 2.99–3.20 years in external validation cohorts (ADNI/PPMI). Per 1-year increment in BAG was linearly associated with elevated risks of Alzheimer's disease (HR = 1.165, 95% CI = 1.086–1.249; +16.5% risk/year), mild cognitive impairment (HR = 1.040, 95% CI = 1.030–1.050; +4.0%), and all-cause mortality (HR = 1.12, 1.09–1.15; +12%; all p < 0.001). Individuals in the highest BAG quartile (Q4) faced substantially amplified risks: 2.8-fold for Alzheimer's disease (HR = 2.801), 6.4-fold for multiple sclerosis (HR = 6.417), and 1.5-fold for major depressive disorder (HR = 1.466). Notably, prodromal Parkinson's disease exhibited paradoxical BAG rejuvenation (mean Δ=−1.441 years, p < 0.001), contrasting with nonsignificant associations in incident Parkinson's cases (HR = 1.830, p = 0.154). Cognitive decline followed nonlinear trajectories, with critical thresholds for domain-specific cognitive decline emerging at Q4 (BAG > 2.48 years). Lifestyle interventions synergistically attenuated BAG progression in advanced neurodegeneration (Q3–Q4; p < 0.05), particularly through smoking cessation, moderated alcohol consumption, and physical activity. Interpretation : BAG robustly predicts accelerated brain aging, neuropsychiatric multimorbidity, and mortality. Its nonlinear cognitive thresholds and stage-dependent lifestyle modifiability underscore clinical utility for risk stratification and personalized prevention strategies. Biological sciences/Neuroscience Biological sciences/Neuroscience/Cognitive neuroscience Brain Age Gap (BAG) Neuropsychiatric Disorders 3D Vision Transformer Cognitive Decline Lifestyle Interventions Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction The aging brain undergoes progressive structural and functional changes, which can be categorized as either normative aging or accelerated aging, influenced by genetic factors, nutritional status, and environmental exposures 1 – 3 . The Brain Age Gap (BAG), defined as the difference between neuroimaging-derived brain age and chronological age, has been established as a robust biomarker for assessing accelerated brain aging 4 – 6 . Advances in structural MRI (sMRI)-based deep learning frameworks have significantly improved the precision of brain age estimation, enabling the classification of individuals into accelerated or delayed brain aging trajectories 7 , 8 . BAG serves a dual purpose: it not only quantifies accelerated neuroaging but also predicts neuropsychiatric morbidity, offering a comprehensive metric for evaluating brain health 9 – 11 . The integration of BAG into clinical practice enhances the understanding of aging trajectories and supports the development of personalized intervention strategies. Despite the growing interest in brain age estimation, existing methods face significant limitations. Traditional approaches, such as convolutional neural networks (CNNs), often struggle to capture the complex spatial and temporal patterns of brain aging, particularly in the presence of disease-related structural changes 12 , 13 . Moreover, many models are constrained by their reliance on small, homogeneous datasets, which limits their generalizability to diverse populations 14 , 15 . To address these challenges, we developed a 3D Vision Transformer (3D-ViT) model 16 , 17 , which leverages the self-attention mechanism of transformers to better capture global and local neuroanatomical features associated with aging. This advanced architecture not only improves the accuracy of brain age estimation but also enhances the model’s ability to detect disease-specific patterns of brain aging, making it a powerful tool for both research and clinical applications. Brain age estimation has become a critical tool in the study of neuropsychiatric diseases 18 – 21 , as well as in the investigation of cognitive 22 , 23 , physiological 24 , genetic 25 , 26 , and environmental factors 27 that influence brain health. As the human brain ages, it undergoes characteristic structural and functional changes, which can be accelerated or decelerated by various conditions, including neurodegenerative diseases, psychiatric disorders, and lifestyle factors. Numerous studies 27 – 30 have demonstrated that brain age estimation is instrumental in identifying risk and protective factors across the lifespan, providing valuable insights into the mechanisms underlying healthy and pathological aging. By monitoring changes in individual brain age estimates over time, researchers can assess the effectiveness of interventions, such as lifestyle modifications, pharmacological treatments, or cognitive training, in promoting brain health and mitigating age-related decline. This approach has the potential to facilitate the development of personalized strategies for healthy aging and the prevention of neuropsychiatric disorders. While the relationship between brain age and the risk of neuropsychiatric disorders has garnered considerable attention, the specific role and predictive utility of the BAG in age-related diseases remain insufficiently elucidated. For instance, it remains uncertain whether BAG can serve as a reliable biomarker for predicting the risk of neuropsychiatric disorders or whether it can effectively monitor disease progression and treatment response. Additionally, the associations between BAG and diverse cognitive functions, as well as the potential of lifestyle interventions to mitigate increases in brain age, warrant further exploration. To address these gaps, we developed a 3D-ViT model for precise brain age estimation, leveraging MRI imaging to assess the brain’s aging status. By calculating the difference between estimated brain age and chronological age, our study aims to evaluate individuals’ aging rates relative to their peers and to investigate the applicability of BAG in assessing the risk of neuropsychiatric disorders and all-cause mortality. Furthermore, we examined the relationship between BAG and cognitive functions and analyzed the potential impact of healthy lifestyle interventions on decelerating brain aging. This research seeks to clarify the role of BAG in neuropsychiatric risk assessment and to inform strategies for promoting cognitive health through lifestyle modifications, ultimately contributing to the development of non-invasive, cost-effective tools for early detection and intervention in age-related diseases. Materials and methods 2.1 Study Design and Participants We developed and validated a brain age estimation model using multi-cohort neuroimaging data. The primary cohort included T1-weighted MRI scans from the UK Biobank 31 (UKB; application ID 89757). External validation was performed using the Alzheimer’s Disease Neuroimaging Initiative (ADNI) 32 and Parkinson’s Progression Markers Initiative (PPMI) 33 datasets. Ethical approval was obtained from the UK Biobank Access Committee and institutional review boards for ADNI/PPMI. All participants provided written informed consent. Cohort Descriptions 1. Primary Cohort (UK Biobank): A total of 42,904 participants (age range: 45–82 years; 52.54% female) with baseline T1-weighted MRI scans were initially included. To mitigate population stratification bias, analyses were restricted to 38,967 individuals of genetically confirmed White British ancestry. This cohort was stratified into three subgroups (Supplementary Table 1): • Healthy controls (CN): 6,036 individuals without ICD-10 neurological/psychiatric diagnoses (G00–G99, F00–F99, I60-I63). • Non-brain disease group: 25,416 individuals with non-neurological comorbidities. • Brain disorder group : 7,515 individuals with brain injury, neurodegeneration, or neuropsychiatric disorders (ICD-10 codes G00–G99, F00–F99, I60-I63, Supplementary Table 2). 2. Disease Enrichment Cohorts: • ADNI Cohort : A longitudinal multicenter study comprising 1,402 participants at baseline (CN: n = 332, 23.7%; MCI: n = 785, 56.0%; AD: n = 285, 20.3%, age range: 55–96 years) with serial T1-weighted MRI scans. Over the follow-up period, the cohort contributed 5,058 MRI scans stratified as follows (Supplementary Table 3): • Cognitively Normal (CN): 1,480 scans (29.3% of total scans) • Mild Cognitive Impairment (MCI): 2,836 scans (56.1%) • Alzheimer’s Disease (AD): 7 42 scans (14.7%) Diagnoses followed the National Institute on Aging–Alzheimer’s Association (NIA-AA) criteria. Longitudinal data were harmonized across ADNI phases (ADNI1/GO/2/3) to account for scanner variability. • PPMI Cohort : A longitudinal multicenter study comprising 1,182 participants at baseline (CN: n = 185, 15.7%; Prodromal-stage: n = 381, 32.2%; PD: n = 616, 52.1%, age range: 45–83 years) with serial T1-weighted MRI scans. Over the follow-up period, the cohort contributed 2,907 MRI scans stratified as follows (Supplementary Table 4): • Healthy Controls: (CN): 279 scans (9.6% of total scans) • Prodromal-stage (Prodromal: 1,326 scans (45.6%) • Parkinson’s Disease: (PD): 1,303 scans (44.8%) Diagnoses followed the Movement Disorder Society (MDS) clinical diagnostic criteria for PD and prodromal markers. Longitudinal MRI data were harmonized across 50 participating sites using standardized preprocessing pipelines. 3. External Validation Cohorts: To evaluate model generalizability in neurologically intact populations, two independent datasets were analyzed: • ADNI Validation Subset: 1,480 scans from cognitively normal individuals (age 55–96 years), excluding those with subsequent clinical progression. • PPMI Healthy Controls: 279 scans from baseline and longitudinal assessments of healthy participants (age 45–83 years). 2.2 MRI Acquisition and Preprocessing Scanners: • UKB: 3T Siemens Skyra scanner equipped with a 32-channel head coil (repetition time [TR]/echo time [TE] = 2,000/2.01 ms). • ADNI: Multisite 3T scanners with harmonized imaging protocols. • PPMI: 3T Siemens Prisma scanner with a 64-channel head coil, utilizing the magnetization-prepared rapid gradient-echo (MPRAGE) sequence. Processing Pipeline: 1. UK Biobank Data: Preprocessed T1-weighted (T1w) scans were utilized, which were registered to the Montreal Neurological Institute (MNI152) 34, 35 standard space with an isotropic voxel size of 1 mm³. 2. External Cohorts: Data were processed using FMRIB Software Library (FSL) version 6.0.5, with the following steps: • Reorientation: Raw MRI volumes were reoriented to the standard anatomical orientation. • Cropping: Images were cropped to remove the neck and subcranial structures. • Bias Field Correction: Bias field correction was applied without segmentation to maintain anatomical integrity. • Non-Brain Tissue Removal: Non-brain tissues were removed using FSL’s Brain Extraction Tool (BET). • Registration: Linear and non-linear registration techniques (6 degrees of freedom) were employed to align images with the MNI152 standard template. • Resampling: Preprocessed MRI volumes were resampled to a voxel size of 182 × 218 × 182 (1 mm³ isotropic resolution). Quality Control: Scans exhibiting motion artifacts (framewise displacement >0.5 mm) or registration errors (Dice coefficient <0.85 compared to the template) were excluded from further analysis. 2.3 Brain Age Estimation Data Preprocessing and Model Architecture In this study, we trained a 3D-ViT model on MRI samples from the UKB for brain age estimation (Fig. 1A, Supplementary Table 5). The ViT model has demonstrated its remarkable performance improvements compared to traditional machine learning models and CNN-based architectures in various medical image analysis tasks. Briefly, we first resized T1 weighted MR images to a resolution of 128 × 128 × 128 as input to the model. We utilized the multi-head attention mechanism of the 3D-ViT to divide the original MRI image into a set of 3D-patches (patch size: 16 × 16 × 16). These patches were processed through a transformer encoder, which measured attention scores between them. A multilayer perceptron header was then employed to summarize the outputs from the transformer encoder for brain age estimation. Training Strategy and Validation The model was optimized by minimizing the mean squared error (MSE) between predicted brain age and chronological age using the Adam optimizer (learning rate: 2 × 10⁻⁵, max epochs: 200). Training utilized 25,162 participants (80% of total N=31,452), comprising 6,036 healthy controls and 19,126 non-brain disorder subjects, with early stopping triggered if validation loss plateaued for 10 consecutive epochs. A nested 5-fold cross-validation protocol was implemented: at each fold, the training cohort was randomly divided into five subsets (4 for parameter optimization, 1 for internal validation). Following cross-validation, the finalized model was evaluated on an independent test set (n=6,290, 20% of total cohort) and externally validated using ADNI and PPMI datasets. Performance metrics included mean absolute error (MAE), MSE, R², and Pearson's correlation coefficient (r) between predicted and chronological ages (Supplementary Table 6). Computational implementation employed Python 3.9 with PyTorch 1.13.1 on CUDA 8.6-enabled hardware. 2.4 Quartile-Based Stratification of BAG The Brain Age Gap (BAG), calculated as the disparity between neuroimaging-derived brain age and chronological age, provides a biomarker of accelerated neurobiological aging. Elevated BAG values (positive deviations) reflect disproportionately rapid brain aging relative to biological age, whereas negative values indicate neuroprotective resilience. To refine its clinical utility in predicting age-related disease trajectories, we stratified participants into quartiles based on population-level BAG distributions: Q1 (slowest aging, 1st quartile), Q2 (2nd quartile), Q3 (3rd quartile), and Q4 (fastest aging, 4th quartile). This risk stratification framework enhances prognostic precision for neuropsychiatric disorders while enabling targeted prioritization of high-risk subgroups (Q4) in population health frameworks. 2.5 Association Between BAG and Cognitive Performance We analyzed multi-domain cognitive assessments from 25,617 UK Biobank participants (mean age 64.5±7.6 years; 52.3% female) across seven core domains 37,38 : short-term memory, visual reasoning, abstract reasoning, processing speed, associative learning, executive function, and visual memory. Cognitive measures included: • Numeric Memory Test (NMT): Digit span recall (0–12) • Matrix Pattern Completion (MPCT): Visual pattern accuracy (0–15) • Fluid Intelligence Test (FIT): Abstract problem-solving (13 items) • Reaction Time Test (RTT): Median response latency (ms) • Paired Associate Learning (PALT): Word-image pairs recalled (0–10) • Tower Rearrangement (TR): Optimal puzzle solutions (3–24 steps) • Symbol Digit Substitution (SDST): Correct matches/minute • Pairs Matching Test (PMT): Visual recognition accuracy (0–6) Sample sizes varied slightly across tests (n = 24,872–25,617) due to incomplete responses, with missing data addressed through listwise deletion to maintain analytical rigor (2.9–5.2% missingness per test). To account for potential bias from missing data, we performed multiple imputation via chained equations 39 (MICE; 20 iterations), incorporating MRI quality parameters, socioeconomic covariates, and cognitive test performance metrics. Imputation results were consistent with primary analyses (Supplementary Table 7). To characterize nonlinear associations between BAG and cognitive performance, we implemented restricted cubic splines (RCS) regression 40,41 . BAG was stratified into population quartiles (Q1: slowest aging to Q4: fastest aging), and knot positions were optimized using the Akaike Information Criterion (AIC) 42,43 to ensure appropriate smoothing. Models were adjusted for chronological age, sex, BMI, smoking status, and alcohol consumption to mitigate potential confounding. The inflection points of the dose-response curves were identified to determine critical thresholds of BAG associated with cognitive decline. Statistical significance was assessed using 95% confidence intervals, and effect sizes were standardized to facilitate cross-domain comparisons. Sensitivity analyses confirmed the robustness of findings across imputed and complete-case datasets. 2.6 BAG as a Risk Marker for Neurodegenerative, Psychiatric, and Neurological Disorders Case Definitions We categorized 16 neuropsychiatric disorders into three pathophysiological groups based on mechanistic profiles (Supplementary Table 2, 3 and 4): 1. Neurodegenerative disorders : Alzheimer's disease (AD; ICD-10: G30), vascular dementia (VaD; F01), and Prodromal Parkinson's and Parkinson's disease (PD; G20) 2. Psychiatric disorders : Major depressive disorder (MDD; F32), anxiety disorders (F41), bipolar disorder (BD; F31), anorexia nervosa (AN; F50.0), obsessive-compulsive disorder (OCD; F42), post-traumatic stress disorder (PTSD; F43.1), and schizophrenia (SCZ; F20) 3. Neurological disorders : Stroke (I63), epilepsy (G40), multiple sclerosis (MS; G35), and sleep disorders (SD; G47) Cases were identified through ICD-10 codes from hospital records. In the brain disorder cohort (N=7,515), prevalence ranged from 26 cases (schizophrenia) to 4,490 cases (major depressive disorder). Cox Regression and Follow-up Parameters Key analytical parameters (Supplementary Tables 11-13) included 46,47 : • Cohort characteristics : • Baseline distribution across BAG quartiles (Q1-Q4) • Incident vs censored case proportions • Median follow-up durations: • UK Biobank: 9.00-9.07 years (max 26.97) • ADNI: 1.91-3.04 years (max 10.50) • PPMI: 2.43-2.59 years (fixed 2-year intervals) • Exposure quantification : • Person-years accumulation patterns • Observation windows (min-max) • Disease-specific attrition : • Highest in UKB depression cohort (4,490 incident cases) • Longest tracking in UKB stroke cohort (median 9.90 person-years) Primary exposure was defined as 1-year increase in BAG. All models adjusted for chronological age, sex, BMI, smoking status, and alcohol consumption. 2.7 Survival Analysis of BAG and Mortality Risk We conducted survival analyses using Cox proportional hazards regression, with age as the underlying time scale. The study cohort comprised 38,967 UK Biobank participants with complete mortality linkage data obtained from national death registries. Detailed analytic parameters are provided in Supplementary Tables 16-18. The primary exposure variable was defined as a 1-year increase in the BAG. All models were adjusted for chronological age, sex, body mass index (BMI), smoking status, and alcohol consumption. Censoring and Validation Participants were right-censored under the following conditions: 1. Death verification : Confirmed through registry linkage. 2. Loss to follow-up : Occurred in less than 0.2% of participants. 3. Study termination : The end of the study period (October 23, 2021). To ensure the validity of the proportional hazards assumption, we conducted systematic validation using the following methods: 1. Schoenfeld residuals : Global and covariate-specific analyses were performed. 2. Stratified survival analysis: Kaplan-Meier curves stratified by BAG quartiles demonstrated proportional hazards across strata (log-rank test: p < 0.001; multivariable-adjusted for chronological age, sex, BMI, smoking status, and alcohol consumption). 3. Non-linear association testing : Restricted cubic splines (RCS) revealed significant non-linearity in the BAG-mortality relationship (likelihood ratio test for non-linearity: p < 0.001). Cox proportional hazards regression with RCS-derived hazard ratios (HR) and 95% confidence intervals (CI) confirmed this pattern. Stratified Analysis Censoring patterns across BAG quartiles revealed a dose-dependent relationship with mortality risk, with the lowest risk observed in the first quartile (Q1: 0.8%) and the highest in the fourth quartile (Q4: 1.8%) (Supplementary Tables 17). Follow-up duration was consistent across quartiles, with a median of 3.4 years (interquartile range [IQR]: 1.8–5.1 years). 2.8 Lifestyle-BAG Interaction Analysis Composite Lifestyle Score Development We constructed a 7-domain lifestyle score 48-51 for 38,967 participants using validated criteria: Figure 1. Seven modifiable factors were scored (0-1 per domain) Factor Healthy Criteria Assessment Tool Smoking Never smoker NHS/NICE 2015 guidelines Physical activity ≥150 min/week moderate or ≥75 min vigorous IPAQ-SF Alcohol ≤14 g/day (women), ≤28 g/day (men) U.S. Dietary Guidelines Diet ≥4/7 healthy food groups† 24-hr dietary recall Sleep 7-9 hours/night AASM/SRS consensus Sedentary time ≤4 hr/day non-work screen time GPAQ Social connection Social isolation index ≤1‡ Social network survey †Fruits, vegetables, fish, whole grains, unprocessed meats, Low-fat dairy products, Nuts ‡Composite of household size, social visit frequency, activity participation The composite lifestyle score demonstrated a continuous distribution from 0 (least healthy) to 7 (most healthy). Categorical stratification was performed as: • Unfavorable : 0-1 • Intermediate : 2-4 • Favorable : 5-7 BAG Interaction Analysis To examine the interaction between lifestyle and the BAG, the following analytical approach was implemented: 1. Stratification • Participants were grouped into BAG quartiles (Q1–Q4) based on their BAG values. 2. Comparative Testing • Between-group comparisons: Student’s t-test was used to assess differences between healthy (favorable lifestyle) and unhealthy (unfavorable lifestyle) groups. • Cross-quartile trends: The Kruskal-Wallis test 52 was applied to evaluate trends across BAG quartiles, with Dunn’s post-hoc adjustment for multiple comparisons to identify specific differences in lifestyle score distributions. Results Performance of 3D-ViT in Brain Age Estimation Overall Model Performance The 3D-ViT model was trained on structural MRI data from 25,162 healthy adults in the UKB, demonstrating robust performance in brain age estimation (Figs. 1; Supplementary Table 6). On the training set, the model achieved a mean absolute error (MAE) of 2.627 years (MSE = 10.988, R²=0.810, Pearson's r = 0.901), indicating strong concordance between predicted and chronological ages (Supplementary Fig. 1). Five-fold cross-validation confirmed stability (average MAE = 2.610 years) without evidence of overfitting. In an independent UKB test cohort ( N = 6,290), performance remained consistent (MAE = 2.684 years, R²=0.805, r = 0.898; Supplementary Fig. 2B), underscoring generalizability. Gender-stratified analyses revealed comparable accuracy for females (MAE = 2.567 years, r = 0.902) and males (MAE = 2.714 years, r = 0.897), with a minimal intergroup difference (< 0.2 years; Supplementary Figs. 2E-F). Age-stratified evaluations demonstrated MAE values of 2.609 years ( r = 0.760) for middle-aged participants (< 65 years) and 2.666 years ( r = 0.687) for elderly individuals (≥ 65 years; Supplementary Figs. 2). Overall, the model exhibited consistent accuracy and robustness across demographic subgroups. Cross-Dataset Validation External validation using ADNI ( N = 1,480 cognitively normal subjects) and PPMI ( N = 279 healthy controls) yielded MAE values of 2.998 years (MSE = 16.000, r = 0.758; Supplementary Fig. 3) and 3.205 years (MSE = 17.640, r = 0.936; Supplementary Fig. 4), respectively. The marginally higher error in ADNI (Δ ≈ 0.3 years vs. UKB) likely originated from protocol differences and an older population baseline, while PPMI's elevated correlation may reflect its narrower age distribution. Optimal performance in UKB was attributable to its expansive sample size and broad age representation. These findings collectively validate the model's cross-dataset adaptability, albeit with protocol- and cohort-dependent variance. Disease-Associated Brain Age Acceleration The model systematically overestimated brain age in neurodegenerative disorders, with error magnitudes correlating with disease severity. In ADNI cohorts, mild cognitive impairment (MCI) subjects ( N = 2,836) exhibited an MAE of 4.77 years (1.6× controls; r = 0.675; Supplementary Fig. 3), while AD dementia patients ( N = 742) showed pronounced acceleration (MAE = 6.08 years, 2× controls; r = 0.51; Supplementary Fig. 3), with predicted brain ages exceeding chronological ages by ~ 6 years. This pattern aligns with AD-specific atrophy trajectories diverging from healthy aging. PD patients ( N = 1,302) displayed modest error increases (MAE = 3.48 years, + 0.3 vs. controls; r = 0.960; Supplementary Fig. 4). Notably, UKB subgroups with mixed neurological conditions ( N = 7,515) maintained robust performance (MAE = 2.77 years vs. 2.68 in controls; r = 0.895), suggesting resilience to multifactorial pathologies. Association between BAG and cognitive decline Cognitive performance was stratified by BAG quartiles in the UK Biobank cohort (n = 25,617, Supplementary Table 8). Significant differences were observed across quartiles for several cognitive domains. Matrix pattern completion scores were higher in Q1 (8.10 ± 2.10) and Q2 (8.12 ± 2.07) compared to Q3 (7.95 ± 2.11) and Q4 (7.95 ± 2.12) (F = 12.40, p < 0.001). Reaction time was significantly slower in Q4 (604.88 ± 112.78 ms) compared to Q1 (591.74 ± 106.26 ms) (F = 19.27, p < 0.001). Paired associate learning scores were lower in Q4 (6.85 ± 2.52) compared to Q1 (7.14 ± 2.52) (F = 19.82, p < 0.001). Symbol digit substitution scores were also lower in Q4 (18.45 ± 5.21) compared to Q1 (19.29 ± 5.15) (F = 40.67, p < 0.001). No significant differences were observed for numeric memory (F = 0.52, p = 0.671) or fluid intelligence (F = 1.93, p = 0.122). Linear regression models revealed significant associations between biological aging quartiles and cognitive performance (Supplementary Table 9 and Supplementary Fig. 6). Compared to Q1, Q3 and Q4 showed significantly lower scores in matrix pattern completion (Q3: β = -0.15, p < 0.001; Q4: β = -0.15, p < 0.001), paired associate learning (Q3: β = -0.14, p = 0.001; Q4: β = -0.28, p < 0.001), and symbol digit substitution (Q3: β = -0.52, p < 0.001; Q4: β = -0.84, p < 0.001). Reaction time was significantly slower in Q3 (β = 8.06, p < 0.001) and Q4 (β = 13.15, p < 0.001) compared to Q1. No significant associations were found for numeric memory or fluid intelligence after adjusting for multiple comparisons using the Benjamini-Hochberg procedure. Non-linear modeling identified inflection points and maximum effects of BAG on cognitive domains (Fig. 2 and Supplementary Table 10). The maximum effects for all cognitive domains were observed in Q4, indicating that the most pronounced cognitive deficits occurred in individuals with the highest BAG. For example, reaction time exhibited an inflection point at BAG ≈ -1.75, with a maximum effect of 3.13 ms in Q4. Similarly, paired associate learning showed a maximum effect of 3.11 in Q4, with an inflection point at BAG ≈ 0.01. These findings suggest that cognitive performance remains relatively stable at lower BAG levels but deteriorates more rapidly beyond a critical threshold, particularly in Q4. BAG as a Risk Marker for Neuropsychiatric Disorders Cross-sectional analysis based on the BAG revealed significant heterogeneity in BAG levels between disease groups and the healthy control group (CN, BAG: 0.122 ± 3.039; Fig. 3B, Supplementary Table 14). Among neurodegenerative disorders, Alzheimer’s disease (AD: 3.242 ± 6.635, p < 0.001) and mild cognitive impairment (MCI: 2.063 ± 5.619, p < 0.001) exhibited significantly elevated BAG, whereas vascular dementia (VaD: 0.279 ± 3.463, p = 0.442) and Parkinson’s disease (PD: -0.003 ± 4.463, p = 0.321) showed no significant differences compared to CN. Notably, prodromal Parkinson’s disease (ProdPD: -1.441 ± 4.880, p < 0.001) demonstrated a unique negative BAG shift. In psychiatric disorders, bipolar disorder (BD: 1.913 ± 4.051, p < 0.001), major depressive disorder (MDD: 0.516 ± 3.409, p < 0.001), and anxiety disorder (ANX: 0.525 ± 3.425, p < 0.001) displayed significantly higher BAG than CN, whereas obsessive-compulsive disorder (OCD: 1.161 ± 3.231, p = 0.074), post-traumatic stress disorder (PTSD: 0.310 ± 3.539, p = 0.645), and anorexia nervosa (AN: 0.369 ± 3.547, p = 0.631) showed no significant differences. Schizophrenia (SCZ: 2.068 ± 3.341, p = 0.007) exhibited elevated BAG despite a limited sample size ( n = 26). Among neurological disorders, multiple sclerosis (MS: 4.069 ± 5.328, p < 0.001) showed the most pronounced BAG elevation, with stroke (Stroke: 0.631 ± 3.585, p < 0.001), epilepsy (Epilepsy: 1.059 ± 3.478, p < 0.001), and sleep disorders (SD: 0.641 ± 3.379, p < 0.001) also demonstrating significantly higher BAG compared to controls. Our study investigated the association between BAG and the risk of 16 neuropsychiatric disorders (Fig. 3C, Supplementary Fig. 7 and Supplementary Tables 15), categorized into neurodegenerative, psychiatric, and neurological groups, using Cox regression models adjusted for chronological age, sex, BMI, smoking status, and alcohol consumption. Elevated BAG was strongly associated with neurodegenerative diseases, particularly Alzheimer’s disease (AD) and mild cognitive impairment (MCI), with each 1-year increase in BAG conferring a 16.5% higher risk of AD (HR = 1.165, 95% CI = 1.086–1.249, p < 0.001) and a 4.0% higher risk of MCI (HR = 1.040, 1.030–1.050, p < 0.001). Participants in the highest BAG quartile (Q4) had 2.8-fold increased AD risk (HR = 2.801, p = 0.011) and 1.7-fold increased MCI risk (HR = 1.691, p < 0.001) compared to Q1, while vascular dementia (VaD) showed no significant BAG association (HR = 1.022, p = 0.323), aligning with its distinct cerebrovascular pathophysiology. Parkinson’s disease (PD) exhibited mixed results: prodromal PD had reduced BAG (mean = − 1.441 vs. controls, p < 0.001), while incident PD showed no significant association (HR = 1.038, p = 0.385), reflecting its focal nigrostriatal degeneration. In psychiatric disorders, BAG independently predicted major depressive disorder (MDD) (HR = 1.046 per year, p < 0.001), bipolar disorder (BD) (HR = 1.174, p < 0.001), and anxiety disorders (ANX) (Q4 HR = 1.442, p < 0.001), with Q4 participants facing 1.5-fold higher MDD risk (HR = 1.466, p < 0.001) and 2.4-fold higher BD risk (HR = 2.431, p = 0.038) versus Q1. Schizophrenia (SCZ) risk escalated sharply in Q4 (HR = 7.504, p = 0.022), albeit with limited case numbers (n = 26). Among neurological disorders, stroke risk increased by 5.5% per BAG year (HR = 1.055, p < 0.001), with Q4 participants facing 1.6-fold higher risk (HR = 1.612, p < 0.001), while multiple sclerosis (MS) demonstrated the most pronounced BAG association (Q4 HR = 6.417, p < 0.001), consistent with its inflammatory-driven accelerated aging. Epilepsy risk doubled in Q4 (HR = 2.189, p < 0.001), likely reflecting cumulative neurotoxic effects. Survival analysis of BAG and all-cause mortality We conducted a comprehensive survival analysis to evaluate the association between BAG and all-cause mortality using data from national death registries linked to participant records in the UKB. The cohort (Supplementary Tables 16–17) was followed for a median of 3.4 years (range: 0–7.5 years), with a total follow-up time of 148,774 person-years. During the follow-up period, 469 participants (1.2%) died from various causes, while 38,498 participants (98.8%) were right-censored. Censoring rates exhibited a graded pattern across BAG quartiles: Q1 (99.2%), Q2 (99.1%), Q3 (98.8%), and Q4 (98.2%). Mortality rates increased progressively across BAG quartiles: Q1 (0.8%), Q2 (0.9%), Q3 (1.2%), and Q4 (1.8%). Cox regression models adjusted for chronological age, sex, BMI, smoking, and alcohol consumption revealed significant mortality associations. Per 1-year increase in BAG conferred a 12% elevated mortality risk (adjusted HR 1.12, 95% CI 1.09–1.15; p < 0.001). Quartile-based analysis demonstrated a dose-response relationship: compared with Q1 (reference), Q2 showed non-significant risk elevation (HR 1.18, 0.87–1.59; p = 0.293), while Q3 (HR 2.11, 1.19–2.59; p = 0.002) and Q4 (HR 2.36, 1.81–3.08; p 0.68, Supplementary Tables 18). Kaplan-Meier survival curves confirmed significant inter-quartile divergence (log-rank p < 0.001). Q4 participants demonstrated substantially reduced survival probability versus Q1 (absolute difference 3.0% at 7 years), with intermediate quartiles showing graded risk progression (Fig. 4A). Nonlinear restricted cubic spline analysis demonstrated a robust and statistically significant association between BAG and mortality risk (Fig. 4B, likelihood ratio test for nonlinearity, p < 0.001). Mortality risk remained stable at BAG values below 2 years (HR ≈ 1.0) but increased sharply thereafter, with a HR of 1.2 at 3 years and exceeding 2.0 at BAG values of 6 to 8 years. These findings underscore the importance of monitoring and addressing elevated BAG levels to mitigate associated mortality risks. Lifestyle-BAG Interaction Stratified analyses demonstrated a significant dose-response relationship between composite lifestyle scores and BAG progression across quartiles ( p < 0.001 for trend; Fig. 5, Supplementary Tables 19–20). The strongest neuroprotective effects emerged in the highest brain-age correlation quartile (Q4): individuals with favorable lifestyles (scores 5–7) exhibited a 0.35-year BAG reduction compared to those with unfavorable profiles (scores 0–1) (4.15 ± 1.63 vs. 4.50 ± 1.81, p < 0.001), while intermediate scores (2–4) showed attenuated progression (annual reduction rate: 1.7%, p = 0.013). Domain-specific analyses identified three key protective factors in Q4: moderate alcohol consumption demonstrated the most substantial effect (-0.20-year BAG reduction, p < 0.001), followed by regular physical activity (-0.14-year, p = 0.03), and never smoking (-0.11-year reduction; 4.16 ± 1.65 vs. 4.27 ± 1.69 in smokers, p = 0.002). Stage-dependent patterns revealed transient sleep benefits restricted to Q1 (-0.09-year, p = 0.03) and U-shaped social connectivity effects peaking in Q3 (1.13 ± 0.61 vs. 1.27 ± 0.62, p = 0.012). Threshold phenomena were observed for sedentary behavior (2.3% BAG reduction in Q1 [ p = 0.046], nonsignificant in Q3–Q4), while dietary patterns showed no significant associations across strata ( p ≥ 0.27). These results delineate a dynamic neuroprotective hierarchy, where lifestyle efficacy shifts across brain aging stages, emphasizing smoking abstinence and alcohol moderation as critical intervention targets in advanced neurodegeneration. Discussion Our study demonstrates the effectiveness of the proposed 3D-ViT framework in estimating brain age, revealing significant advantages over traditional CNN methods. The analysis identifies robust associations between an elevated BAG and domain-specific cognitive decline, increased risks of neuropsychiatric disorders, and heightened all-cause mortality within a large, population-based cohort. Furthermore, our findings indicate that healthy lifestyle interventions can effectively mitigate the negative impacts of accelerated brain aging. These results collectively support the utility of BAG as a reliable biomarker for assessing biological aging, identifying high-risk individuals, and guiding early clinical interventions and public health strategies aimed at promoting brain health and longevity. The proposed 3D-ViT framework for brain age estimation demonstrates superior performance in capturing comprehensive neuroanatomical features compared with conventional CNN approaches. Traditional CNN methods, although capable of extracting global features from whole-brain MRI, typically emphasize prominent regions and overlook subtle anatomical details 53 , 54 . Conversely, patch-based CNN methods retain localized anatomical information but suffer from contextual fragmentation due to limited receptive fields 55 . Our 3D-ViT model addresses these limitations by integrating global and local neuroanatomical information through a self-attention mechanism, effectively capturing both broad structural patterns and fine-grained features within a unified architecture 56 . This multi-scale feature fusion aligns closely with emerging hybrid methods that combine CNN-derived local descriptors and Transformer-based contextual aggregation 57 but avoids the computational redundancy typically associated with multi-network ensembles. The proposed model achieves state-of-the-art performance in large-scale healthy populations and exhibits robust generalizability across external validation cohorts and disease-specific groups. Specifically, our 3D-ViT model demonstrates several key advantages: (1) it leverages Transformer-based architectures to analyze whole-brain 3D MRI scans, yielding a mean absolute error (MAE) of only 2.68 years in healthy adults, substantially outperforming conventional CNN methods (e.g., 2D ResNet 58 : MAE = 6.8 years; 3D VGGNet 59 , 60 : MAE = 4.45 years; 3D EfficientNet 61 : MAE = 3.31 years); (2) it maintains consistent and unbiased performance across diverse demographic groups (gender and age), highlighting the universality of learned neuroanatomical features; (3) it demonstrates robust cross-cohort generalization validated on external ADNI (MAE = 2.99 years) and PPMI (MAE = 3.20 years) datasets, despite variability in MRI acquisition protocols; and (4) it exhibits sensitivity to pathological brain aging, as evidenced by significantly elevated brain age estimates in patients with neuropsychiatric disorders, closely aligning with clinically observed cognitive decline and neurodegeneration trajectories. Collectively, these findings highlight the considerable potential of brain age estimation as an auxiliary clinical tool for disease screening and progression monitoring. Our study provides robust evidence of a significant association between the BAG and domain-specific cognitive decline in the UK Biobank cohort. These findings enhance our understanding of neurobiological aging by identifying critical thresholds and nonlinear relationships between structural brain changes and cognitive performance. Stratification by BAG quartiles revealed a clear gradient in cognitive outcomes, with the most pronounced deficits observed in the highest quartile (Q4; BAG > 2.48 years). Individuals in Q4 exhibited significantly slower reaction times, reduced executive function (symbol digit substitution: β = -0.84 vs. Q1), and impaired associative memory (paired associate learning: β = -0.28 vs. Q1). These results align with prior neuroimaging studies 62 – 64 implicating frontostriatal and hippocampal networks in age-related cognitive decline. Importantly, our analysis indicates distinct cognitive trajectories across BAG quartiles. Accelerated brain aging correlates with measurable cognitive impairment even among apparently healthy middle-aged and older adults. Previous studies 65 – 68 have reported similar associations, demonstrating that higher BAG is linked to poorer cognitive and motor functions in populations such as Parkinson’s disease patients and older adults with decreased processing speed and memory performance. Identifying domain-specific vulnerabilities through BAG highlights its potential as a biomarker for early detection of subclinical cognitive decline. Individuals surpassing a critical BAG threshold (> 2.48 years) appear especially vulnerable to substantial cognitive deterioration, suggesting a critical intervention window for targeted preventive strategies. Our large-scale study demonstrates that the BAG exhibits distinct pathophysiological associations across neurodegenerative, psychiatric, and neurological disorders, while also reflecting shared mechanisms underlying neuropsychiatric conditions. The robust association between elevated BAG and AD (HR = 1.165/year, p < 0.001) aligns with prior evidence of global brain atrophy and metabolic dysregulation in AD pathogenesis 69 . Intriguingly, the paradoxical reduction in BAG observed in prodromal Parkinson’s disease (Q4 HR = 0.925, p = 0.105) may reflect compensatory neuroplasticity 70 , 71 during early nigrostriatal degeneration, whereas the absence of BAG effects (Q4 HR = 1.830, p = 0.154) in incident PD suggests that focal dopaminergic loss minimally impacts global brain aging metrics 72 . The dose-dependent relationship between BAG and psychiatric disorders further highlights its transdiagnostic utility. For MDD, the quartile-dependent risk escalation (Q4 HR = 1.466, p < 0.001) supports the hypothesis of accelerated aging mediated by chronic stress-induced glucocorticoid toxicity and mitochondrial dysfunction 73 , 74 . Notably, SCZ exhibited an exponential risk surge in the highest BAG quartile (HR = 7.504, p = 0.022), suggesting that neurodevelopmental deficits in synaptic pruning may synergize with aging-related cortical thinning to amplify vulnerability 75 , 76 . In neurological disorders, the pronounced association between BAG and MS (Q4 HR = 6.417, p < 0.001) points to inflammaging—a convergence of chronic neuroinflammation and epigenetic aging—as a driver of disability progression 77 , 78 . Similarly, the 5.5% annual increase in stroke risk per BAG year ( p < 0.001) implicates endothelial glycocalyx degradation and blood-brain barrier disruption as mediators of neurovascular aging 79 , 80 . These findings support the concept of brain age as a comprehensive indicator of cerebral health: when the brain appears "older" than its chronological age, it often suggests the presence of underlying pathological processes or neurological damage. Although elevated brain age may not serve as a disease-specific marker but rather reflects a nonspecific indicator of cumulative pathological impacts on the brain, its significant association with disease risk provides critical insights for early identification of high-risk individuals. This research highlights the intricate interplay between systemic aging and neural circuitry, offering novel perspectives for the prevention and early intervention of neuropsychiatric disorders. Our study provides compelling evidence that the BAG independently predicts all-cause mortality within a large, population-based cohort. Three key findings highlight the clinical and prognostic relevance of BAG. First, each 1-year increase in BAG corresponded to a 12% elevated risk of mortality, even after adjusting for chronological age, sex, BMI, smoking status, and alcohol consumption. Second, a clear dose-response relationship was evident across BAG quartiles; participants in Q4 exhibited a 2.36-fold higher mortality risk compared to those in Q1. Third, nonlinear threshold effects were observed, indicating stable mortality risks when BAG was below 2 years, with significant increases beyond 3 years, reaching a HR greater than 2.0 between 6 to 8 years. Our findings position BAG as a valuable biomarker for biological aging with substantial prognostic implications. Our results align with emerging evidence linking accelerated biological aging to adverse health outcomes, including cardiovascular mortality and frailty progression. The 12% annual increase in mortality risk per year of elevated BAG mirrors previously reported associations between epigenetic age acceleration and cardiovascular mortality 81 (HR range: 1.58–1.59 per year). Additionally, the pronounced threshold effect at BAG values exceeding 6 years likely reflects cumulative neurological damage surpassing compensatory mechanisms. This interpretation is consistent with neuroimaging studies 82 , 83 reporting accelerated cortical thinning and increased white matter hyperintensities in individuals with higher BAG values. Overall, our findings underscore the importance of monitoring elevated BAG levels, offering potential targets for early interventions aimed at mitigating associated mortality risks. This study further elucidates the significant association between healthy lifestyle factors and brain age. The results demonstrate that individuals who adhere to regular physical activity, a balanced diet, non-smoking, and moderate alcohol consumption tend to exhibit a "younger" brain age compared to those with less favorable lifestyle habits, suggesting that healthy behaviors may effectively decelerate the brain aging process. Specifically, within the Q4 of brain age, non-smokers showed a reduction in BAG of 0.11 years compared to smokers (p = 0.002), while moderate alcohol consumption (BAG reduction of 0.20 years, p < 0.001) and regular exercise (BAG reduction of 0.14 years, p = 0.03) also demonstrated significant attenuation of neurodegenerative effects. These findings are highly consistent with recent research, such as the UK Biobank cohort analysis, which revealed that although individuals with diabetes or prediabetes exhibited accelerated brain aging, this trend was markedly mitigated in subgroups maintaining optimal lifestyle practices, including non-smoking, high physical activity, and moderate alcohol consumption 84 . Notably, while previous large-scale studies in the general population have shown relatively weak correlations between individual lifestyle factors and brain age, this study underscores the synergistic protective effects of comprehensive healthy behaviors on brain age. This discovery holds significant public health implications: interventions promoting exercise, smoking cessation, alcohol moderation, and balanced nutrition may effectively counteract brain aging at the population level, thereby mitigating the negative impacts of genetic or disease-related factors on brain age. The present study provides new empirical support for this hypothesis and lays a theoretical foundation for future research on lifestyle-based interventions targeting brain aging. These findings not only deepen our understanding of the relationship between lifestyle and brain health but also offer scientific evidence for the development of preventive public health strategies. Conclusion The 3D-ViT model represents a transformative advancement in brain age estimation, offering unprecedented accuracy and generalizability through its innovative integration of self-attention mechanisms and volumetric MRI analysis. By capturing disease-specific neuropathological signatures and hierarchical associations with cognitive decline, this framework provides critical insights into accelerated brain aging across neurodegenerative, psychiatric, and neurological disorders. Its independent association with all-cause mortality underscores its utility as a robust biomarker of biological aging, while the identification of stage-specific lifestyle interventions highlights the potential for tailored strategies to mitigate neurodegeneration. This study advances our understanding of brain aging and establishes a precision framework for risk stratification, intervention timing, and public health recommendations to optimize brain health across the lifespan. Limitations This study has several methodological constraints that warrant consideration. Firstly, the generalizability of findings is constrained by sampling bias in data sources. The training and validation datasets were predominantly derived from large-scale cohorts (UK Biobank, ADNI, PPMI) characterized by demographic homogeneity—particularly the overrepresentation of European ancestry individuals with above-average health literacy in UK Biobank, and disease-specific selection biases in ADNI (Alzheimer's-focused) and PPMI (Parkinson's-enriched). Secondly, technical heterogeneity in neuroimaging protocols introduces potential measurement confounding. Variations in MRI hardware specifications (e.g., 1.5T vs. 3T scanners) and acquisition parameters across datasets, despite rigorous intensity normalization and harmonization procedures, could systematically influence brain age estimations—a persistent challenge in multicenter neuroimaging research that necessitates advanced cross-scanner calibration techniques. Thirdly, the cross-sectional design precludes causal inference regarding the temporal dynamics between brain aging acceleration and clinical outcomes. While significant associations were observed between elevated brain age and cognitive decline/disease risk, the directionality of these relationships remains ambiguous, as residual confounding from unmeasured genetic, epigenetic, or environmental factors cannot be excluded. Declarations Data availability The data generated in this study and a data dictionary (Supplementary Data) are provided in the Supplementary Data. The data that support training and validating the proposed brain age estimation model were obtained from the UK Biobank (https://www.ukbiobank.ac.uk/enable-your-research/register), the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset (https://adni.loni.usc.edu/), and the Parkinson's Progression Markers Initiative (PPMI) dataset (https://www.ppmi-info.org/), upon registration and compliance with the data use agreement. Code availability The source codes pertaining to both the brain age estimation model and data analysis in this manuscript are provided at https://github.com/ZRX-MedAI/Brain_Age_Estimation. Funding This research was supported by Key R&D Program of Zhejiang No.2024C04024. Natural Science Foundation of Xinjiang Autonomous Region No.2022D01C434. Natural Science Foundation of Xinjiang Autonomous Region No.2022D01C434. Author contributions ZH and MH jointly supervised research. ZH and ZJ designed this study. FY developed a deep learning model. SZ and FY performed the model interpretation. SZ, FY and ZH performed data analysis. SZ, FY and ZH interpreted the results. SZ, FY, ZJ, MH and ZH prepared the first draft of the manuscript. All authors contributed and approved the final draft. Competing Interests The authors declare no competing interests. Acknowledgements The UK Biobank resource was used under application number 89757. The MEGASTROKE project received funding from sources specified at http://www.megastroke.org/acknowledgments.html. Supplementary material Supplementary material is available at Supplementary material . References Zhao B, Li T, Fan Z, et al. Heart-brain connections: Phenotypic and genetic insights from magnetic resonance images. Science 2023; 380 : abn6598. Tarkhov AE, Lindstrom-Vautrin T, Zhang S, et al. Nature of epigenetic aging from a single-cell perspective. Nat Aging 2024; 4 : 854–70. Seitz-Holland J, Haas SS, Penzel N, Reichenberg A, Pasternak O. BrainAGE, brain health, and mental disorders: A systematic review. Neurosci Biobehav Rev 2024; 159 : 105581. Seitz-Holland J, Haas SS, Penzel N, Reichenberg A, Pasternak O. BrainAGE, brain health, and mental disorders: A systematic review. Neurosci Biobehav Rev 2024; 159 : 105581. Moguilner S, Baez S, Hernandez H, et al. Brain clocks capture diversity and disparities in aging and dementia across geographically diverse populations. Nat Med 2024; published online Aug 26. DOI:10.1038/s41591-024-03209-x. Azzam M, Xu Z, Liu R, et al. A review of artificial intelligence-based brain age estimation and its applications for related diseases. Brief Funct Genomics 2024; published online Oct 22. DOI:10.1093/bfgp/elae042. Bellantuono L, Marzano L, La Rocca M, et al. Predicting brain age with complex networks: From adolescence to adulthood. Neuroimage 2021; 225 : 117458. Hong J, Feng Z, Wang S-H, et al. Brain age prediction of children using routine brain MR images via deep learning. Front Neurol 2020; 11 : 584682. Han LKM, Schnack HG, Brouwer RM, et al. Contributing factors to advanced brain aging in depression and anxiety disorders. Transl Psychiatry 2021; 11 : 402. Luo Y, Chen W, Qiu J, Jia T. Accelerated functional brain aging in major depressive disorder: evidence from a large scale fMRI analysis of Chinese participants. Transl Psychiatry 2022; 12 : 397. Chen Y-S, Kuo C-Y, Lu C-H, Wang Y-W, Chou K-H, Lin W-C. Multiscale brain age prediction reveals region-specific accelerated brain aging in Parkinson’s disease. Neurobiol Aging 2024; 140 : 122–9. Hwang I, Yeon EK, Lee JY, et al. Prediction of brain age from routine T2-weighted spin-echo brain magnetic resonance images with a deep convolutional neural network. Neurobiol Aging 2021; 105 : 78–85. Kolbeinsson A, Filippi S, Panagakis Y, et al. Accelerated MRI-predicted brain ageing and its associations with cardiometabolic and brain disorders. Sci Rep 2020; 10 : 19940. Varatharajah Y, Baradwaj S, Kiraly A, et al. Predicting brain age using structural neuroimaging and deep learning. bioRxiv. 2018; published online Dec 17. DOI:10.1101/497925. Wang J, Knol MJ, Tiulpin A, et al. Gray matter age prediction as a biomarker for risk of dementia. Proc Natl Acad Sci U S A 2019; 116 : 21213–8. Lahoud J, Cao J, Khan FS, et al. 3D vision with transformers: A survey. arXiv [cs.CV]. 2022; published online Aug 8. http://arxiv.org/abs/2208.04309. Scabini L, Sacilotti A, Zielinski KM, Ribas LC, De Baets B, Bruno OM. A comparative survey of Vision Transformers for feature extraction in texture analysis. arXiv [cs.CV]. 2024; published online June 10. http://arxiv.org/abs/2406.06136. Laguna A, Peñuelas N, Vila M. Modelling human neuronal catecholaminergic pigmentation in rodents recapitulates age-related multisystem neurodegenerative deficits v2. 2024; published online June 4. DOI:10.17504/protocols.io.4r3l2qq1ql1y/v2. Tian YE, Cropley V, Maier AB, Lautenschlager NT, Breakspear M, Zalesky A. Heterogeneous aging across multiple organ systems and prediction of chronic disease and mortality. Nat Med 2023; 29 : 1221–31. Ballester PL, Suh JS, Ho NCW, et al. Gray matter volume drives the brain age gap in schizophrenia: a SHAP study. Schizophrenia (Heidelb) 2023; 9 : 3. Siddiqi SH, Schaper FLWVJ, Horn A, et al. Brain stimulation and brain lesions converge on common causal circuits in neuropsychiatric disease. Nat Hum Behav 2021; 5 : 1707–16. Teselink J, Bawa KK, Koo GK, et al. Efficacy of non-invasive brain stimulation on global cognition and neuropsychiatric symptoms in Alzheimer’s disease and mild cognitive impairment: A meta-analysis and systematic review. Ageing Res Rev 2021; 72 : 101499. Operto FF, Child Neuropsychiatry Unit, Department of Medicine, Surgery and Dentistry, University of Salerno, Salerno, Italy, Pastorino GMG, et al. Social Cognition in neuropsychiatric disorders in pediatric age. Brain (Bacau) 2020; 11 : 81–8. Honea RA, Wilkins H, Hunt SL, et al. TOMM40 may mediate GFAP, neurofilament light Protein, pTau181, and brain morphometry in aging. Aging Brain 2025; 7 : 100134. Hernandez AR, Parker E, Babar M, et al. Microbiome-driven alterations in metabolic pathways and impaired cognition in aged female TgF344-AD rats. Aging Brain 2024; 5 : 100119. Luo S, Wang P-Y, Liu W-H, et al. Age at onset of genetic disease and genetic dependent stage: evidence from cases with SCN1A variants. Brain. 2025; published online Feb 19. DOI:10.1093/brain/awaf064. Smith S, Hopp SC. The 5XFAD mouse model of Alzheimer’s disease displays age-dependent deficits in habituation to a novel environment. Aging Brain 2023; 3 : 100078. Correction to: GDF5 as a rejuvenating treatment for age-related neuromuscular failure. Brain 2025; 148 : e21. de Fátima Dias M, Duarte JV, de Carvalho P, Castelo-Branco M. Unravelling pathological ageing with brain age gap estimation in Alzheimer’s disease, diabetes, and schizophrenia. Brain Commun 2025; published online March 11. DOI:10.1093/braincomms/fcaf109. Wang X, Zhu Z, Xu X, et al. Construction of brain age models based on structural and white matter information. Brain Res 2025; 1851 : 149458. Gibbon S, Breen DP, MacGillivray TJ, UK Biobank Eye & Vision Consortium. Optic disc pallor in Parkinson’s disease: A UK Biobank study. Mov Disord 2025; published online Jan 30. DOI:10.1002/mds.30127. Hanazawa R, Sato H, Hirakawa A, Alzheimer’s Disease Neuroimaging Initiative. Mixture disease progression model to predict and cluster the long-term trajectory of cognitive decline in Alzheimer’s disease. Ther Innov Regul Sci 2025; 59 : 264–77. Nudelman KNH, Brumm MC, Marek K, Foroud TM, for the Parkinson’s Progression Markers Initiative (PPMI) Study. TREM2 variants in Parkinson’s disease: Results from the Parkinson’s progression markers initiative (PPMI) study. Alzheimers Dement 2022; 18 . DOI:10.1002/alz.062316. Giff A, Noren G, Magnotti J, et al. Spatial normalization discrepancies between native and MNI152 brain template scans in gamma ventral capsulotomy patients. Psychiatry Res Neuroimaging 2023; 329 : 111595. Giff A, Noren G, Magnotti J, et al. 19. Spatial normalization discrepancies between native and MNI152 brain template scans in gamma ventral capsulotomy patients. Biol Psychiatry 2023; 93 : S101–2. Chung BS, Park JS. Automatic segmentation of true color sectioned images using FMRIB Software Library: First trial in brain, gray matter, and white matter. Clin Anat 2020; 33 : 1197–203. Zuo Q, Song L, Gao X, et al. Associations of metabolic syndrome with cognitive function and dementia risk: Evidence from the UK Biobank cohort. Diabetes Obes Metab 2024; 26 : 6023–33. Schulz C-A, Weinhold L, Schmid M, Nöthen MM, Nöthlings U. Analysis of associations between dietary patterns, genetic disposition, and cognitive function in data from UK Biobank. Eur J Nutr 2023; 62 : 511–21. Susetyo B, Fitrianto A. Estimating missing panel data with regression and multivariate imputation by chained equations (MICE). CAUCHY 2024; 9 : 94–105. Austin PC. Graphical methods to illustrate the nature of the relation between a continuous variable and the outcome when using restricted cubic splines with a Cox proportional hazards model. Stat Methods Med Res 2025; 34 : 277–85. Erratum to “A dose-effect network meta-analysis model with application in antidepressants using restricted cubic splines.” Stat Methods Med Res 2024; 33 : NP1. Fuh C-D, Kao C-LM, Pang T. Kullback-Leibler divergence and Akaike information criterion in general hidden Markov models. IEEE Trans Inf Theory 2024; 70 : 5888–909. Saumard A, Navarro F. Finite sample improvement of Akaike’s information criterion. IEEE Trans Inf Theory 2021; 67 : 6328–43. Xia L, Nan B, Li Y. Statistical inference for Cox proportional hazards models with a diverging number of covariates. Scand Stat Theory Appl 2023; 50 : 550–71. Hahn G, Prokopenko D, Hecker J, et al. Polygenic hazard score models for the prediction of Alzheimer’s free survival using the lasso for Cox’s proportional hazards model. Genet Epidemiol 2025; 49 : e22581. Hepsomali P, Groeger JA. Diet, sleep, and mental health: Insights from the UK Biobank study. Nutrients 2021; 13 : 2573. Huang S-Y, Li Y-Z, Zhang Y-R, et al. Sleep, physical activity, sedentary behavior, and risk of incident dementia: a prospective cohort study of 431,924 UK Biobank participants. Mol Psychiatry 2022; 27 : 4343–54. Chudasama YV, Khunti K, Gillies CL, et al. Healthy lifestyle and life expectancy in people with multimorbidity in the UK Biobank: A longitudinal cohort study. PLoS Med 2020; 17 : e1003332. Zhang Y-B, Chen C, Pan X-F, et al. Associations of healthy lifestyle and socioeconomic status with mortality and incident cardiovascular disease: two prospective cohort studies. BMJ 2021; 373 : n604. Rodriguez D, Sued M, Valdora M. A Kruskal-Wallis type test for functional data. Commun Stat Simul Comput 2025; : 1–15. He S, Grant PE, Ou Y. Global-local transformer for brain age estimation. IEEE Trans Med Imaging 2022; 41 : 213–24. Qiu S, Joshi PS, Miller MI, et al. Development and validation of an interpretable deep learning framework for Alzheimer’s disease classification. Brain 2020; 143 : 1920–33. Brendel W, Bethge M. Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNet. arXiv [cs.CV]. 2019; published online March 20. http://arxiv.org/abs/1904.00760. Zhang X, Yang H, Young EFY. Attentional transfer is all you need: Technology-aware layout pattern generation. In: 2021 58th ACM/IEEE Design Automation Conference (DAC). IEEE, 2021. DOI:10.1109/dac18074.2021.9586227. He S, Grant PE, Ou Y. Global-local transformer for brain age estimation. IEEE Trans Med Imaging 2022; 41 : 213–24. Borkar K, Chaturvedi A, Vinod PK, Bapi RS. Ayu-Characterization of healthy aging from neuroimaging data with deep learning and rsfMRI. Front Comput Neurosci 2022; 16 : 940922. Cheng Y, Zhang X-D, Chen C, et al. Dynamic evolution of brain structural patterns in liver transplantation recipients: a longitudinal study based on 3D convolutional neuronal network model. Eur Radiol 2023; 33 : 6134–44. Kuchcinski G, Rumetshofer T, Zervides KA, et al. MRI BrainAGE demonstrates increased brain aging in systemic lupus erythematosus patients. Front Aging Neurosci 2023; 15 : 1274061. Poloni KM, Ferrari RJ. A deep ensemble hippocampal CNN model for brain age estimation applied to Alzheimer’s diagnosis. Expert Syst Appl 2022; 195 : 116622. Cadena EJ, White DM, Kraguljac NV, Reid MA, Lahti AC. Evaluation of fronto-striatal networks during cognitive control in unmedicated patients with schizophrenia and the effect of antipsychotic medication. NPJ Schizophr 2018; 4 : 8. Averbeck B, O’Doherty JP. Reinforcement-learning in fronto-striatal circuits. Neuropsychopharmacology 2022; 47 : 147–62. Tang Y, Yan Y, Mao J, Ni J, Qing H. The hippocampus associated GABAergic neural network impairment in early-stage of Alzheimer’s disease. Ageing Res Rev 2023; 86 : 101865. Pal G, O’Keefe J, Robertson-Dick E, Bernard B, Anderson S, Hall D. Global cognitive function and processing speed are associated with gait and balance dysfunction in Parkinson’s disease. J Neuroeng Rehabil 2016; 13 : 94. Ebaid D, Crewther SG, MacCalman K, Brown A, Crewther DP. Cognitive processing speed across the lifespan: Beyond the influence of motor speed. Front Aging Neurosci 2017; 9 : 62. Albrecht F, Johansson H, Ekman U, et al. Investigating underlying brain structures and influence of mild and subjective cognitive impairment on dual-task performance in people with Parkinson’s disease. Sci Rep 2024; 14 : 9513. Abd-Alrazaq A, Ahmed A, Alali H, Aldardour AM, Househ M. The effectiveness of serious games on cognitive processing speed among older adults with cognitive impairment: Systematic review and meta-analysis. JMIR Serious Games 2022; 10 : e36754. Jung YH, Chae CW, Han HJ. The potential role of gut microbiota-derived metabolites as regulators of metabolic syndrome-associated mitochondrial and endolysosomal dysfunction in Alzheimer’s disease. Exp Mol Med 2024; 56 : 1691–702. Bogdanova I, Sokolik V, Voloshyna N, et al. The effectiveness of rehabilitation programs for the mobilization of compensatory-adaptive neuroplasticity processes in patients with Parkinson’s disease according to indicators of neurotrophic factors. Ukrains’kyi Visnyk Psykhonevrolohii 2022; : 18–23. Passaretti M, Cilia R, Rinaldo S, et al. Neurophysiological markers of motor compensatory mechanisms in early Parkinson’s disease. Brain 2024; 147 : 3714–26. Rademacher K, Nakamura K. Role of dopamine neuron activity in Parkinson’s disease pathophysiology. Exp Neurol 2024; 373 : 114645. Yegorov YE, Poznyak AV, Nikiforov NG, Sobenin IA, Orekhov AN. The link between chronic stress and accelerated aging. Biomedicines 2020; 8 : 198. Bobba-Alves N, Sturm G, Lin J, et al. Cellular allostatic load is linked to increased energy expenditure and accelerated biological aging. Psychoneuroendocrinology 2023; 155 : 106322. Jamea AA, Alblowi M, Alghamdi J, et al. Altered default mode network activity and cortical thickness as vulnerability indicators for SCZ: a preliminary resting state MRI study. Eur Rev Med Pharmacol Sci 2021; 25 : 669–77. Wang H, Zhao Q, Zhang Y, et al. Shared genetic architecture of cortical thickness alterations in major depressive disorder and schizophrenia. Prog Neuropsychopharmacol Biol Psychiatry 2024; 135 : 111121. Klotz L, Antel J, Kuhlmann T. Inflammation in multiple sclerosis: consequences for remyelination and disease progression. Nat Rev Neurol 2023; 19 : 305–20. Psenicka MW, Smith BC, Tinkey RA, Williams JL. Connecting neuroinflammation and neurodegeneration in multiple sclerosis: Are oligodendrocyte precursor cells a nexus of disease? Front Cell Neurosci 2021; 15 : 654284. Nian K, Harding IC, Herman IM, Ebong EE. Blood-brain barrier damage in ischemic stroke and its regulation by endothelial mechanotransduction. Front Physiol 2020; 11 : 605398. Li X, Simo L, Zhao Q, Kim EG, Ding Y, Geng X. Endothelial cells and the blood-brain barrier: Critical determinants of ineffective reperfusion in stroke. Eur J Neurosci 2025; 61 : e16663. Gong Y, Song Y, Xu J, et al. Progression of frailty and cardiovascular outcomes among Medicare beneficiaries. medRxiv 2024; published online Feb 13. DOI:10.1101/2024.02.09.24302612. Bernal J, Menze I, Yakupov R, et al. Longitudinal evidence for a mutually reinforcing relationship between white matter hyperintensities and cortical thickness in cognitively unimpaired older adults. medRxiv. 2024; published online July 10. DOI:10.1101/2024.07.08.24309994. Jiménez-Balado J, Habeck C, Stern Y, Eich T. The relationship between cortical thickness and white matter hyperintensities in mid to late life. Neurobiol Aging 2024; 141 : 129–39. Dove A, Wang J, Huang H, et al. Diabetes, prediabetes, and brain aging: The role of healthy lifestyle. Diabetes Care 2024; 47 : 1794–802. Additional Declarations There is NO Competing Interest. Supplementary Files Supplementary.docx Supplementary Tables and Figures for Cite Share Download PDF Status: Published Journal Publication published 24 Oct, 2025 Read the published version in Communications Medicine → Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6283338","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":434487416,"identity":"f32dcb66-c521-46c0-8d4f-b21a984d878f","order_by":0,"name":"Zhengxing Huang","email":"data:image/png;base64,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","orcid":"https://orcid.org/0000-0002-2644-8642","institution":"Zhejiang University","correspondingAuthor":true,"prefix":"","firstName":"Zhengxing","middleName":"","lastName":"Huang","suffix":""},{"id":434487417,"identity":"dc5962da-eff4-4cd2-ae64-0148ca3fa380","order_by":1,"name":"Ruixia Zhang","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Ruixia","middleName":"","lastName":"Zhang","suffix":""},{"id":434487418,"identity":"46b68c98-8091-437f-9144-1d30aa9bbd80","order_by":2,"name":"Fan Yi","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Fan","middleName":"","lastName":"Yi","suffix":""},{"id":434487419,"identity":"556080d6-f8f8-4b50-b043-91019902cf37","order_by":3,"name":"Junhang Zhang","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Junhang","middleName":"","lastName":"Zhang","suffix":""},{"id":434487420,"identity":"8ee90379-9f98-4772-812e-4b12733ef233","order_by":4,"name":"Hongjing Mao","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Hongjing","middleName":"","lastName":"Mao","suffix":""},{"id":434487421,"identity":"1ea8c42a-c96c-4d46-8a30-9106c1b76483","order_by":5,"name":"Kai Wang","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Kai","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2025-03-22 11:35:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6283338/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6283338/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s43856-025-01100-5","type":"published","date":"2025-10-24T04:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":80707436,"identity":"2b90bbf8-10e3-43e5-9ce8-973c048072b7","added_by":"auto","created_at":"2025-04-16 08:45:51","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":83098,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparative Performance Analysis of the 3D-ViT Neural Network in Brain Age Estimation. (A) \u003c/strong\u003eArchitectural overview of the 3D-ViT model. (B) Performance metrics across demographic and clinical subgroups.\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6283338/v1/321eb5c8d5906f83416317d0.jpg"},{"id":80707435,"identity":"28193c22-6091-4441-8a3f-6f86e931c98e","added_by":"auto","created_at":"2025-04-16 08:45:51","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":295977,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDose-Response Relationships Between BAG and Cognitive Performance Across Neuropsychological Domains. (A) \u003c/strong\u003eNumeric Memory; \u003cstrong\u003e(B) \u003c/strong\u003eMatrix Pattern Completion; \u003cstrong\u003e(C)\u003c/strong\u003e Fluid Intelligence; \u003cstrong\u003e(D)\u003c/strong\u003e Reaction Time;\u003cstrong\u003e (E)\u003c/strong\u003e Paired Associate Learning; \u003cstrong\u003e(F)\u003c/strong\u003e Tower Rearranging;\u003cstrong\u003e (G)\u003c/strong\u003e Symbol Digit Substitution;\u003cstrong\u003e (H)\u003c/strong\u003e Pairs Matching. Restricted cubic spline (RCS) curves illustrate the non-linear associations between BAG and cognitive performance, adjusted for chronological age, sex, BMI, smoking status, and alcohol consumption. Each RCS curve is accompanied by a 95% confidence interval (CI), demonstrating the statistical significance of the observed trends.\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6283338/v1/f0203721e8efe3d1e5a0161b.jpg"},{"id":80709116,"identity":"57098ece-9b2b-4e33-b3f1-8469a12a4b7c","added_by":"auto","created_at":"2025-04-16 08:53:52","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":435752,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEvaluating the Ability of BAG to assess neuropsychiatric disorders. (A)\u003c/strong\u003e Conceptual framework of BAG and its stratification: The BAG is defined as the difference between predicted brain age and chronological age, serving as a biomarker of accelerated brain aging. Individuals were stratified into quartiles (Q1–Q4) based on BAG distribution, with Q4 representing the highest deviation from expected aging patterns.\u003cstrong\u003e(B) \u003c/strong\u003eDistribution of BAG across neuropsychiatric disorders: Violin plots illustrate the distribution of BAG across various disorders, including mild cognitive impairment (MCI), multiple sclerosis (MS), Alzheimer's disease (AD), Parkinson’s disease (PD), stroke, anxiety disorders (ANX), bipolar disorder (BD), major depressive disorder (MDD), schizophrenia (SCZ), epilepsy, and post-traumatic stress disorder (PTSD). Healthy controls (CN) are included as a reference group. Wider sections of each plot indicate higher data density, with color coding corresponding to disease classification. \u003cstrong\u003e(C) \u003c/strong\u003eAssociation between BAG and disease risk across neuropsychiatric disorders: Hazard ratios (HR) with 95% confidence intervals (CIs) from Cox proportional hazard models estimating the risk of neuropsychiatric and neurological diseases per unit increase in BAG and across BAG quartiles (Q1–Q4). Asterisks indicate levels of statistical significance (*p \u0026lt; 0.05, **p \u0026lt; 0.01, ***p \u0026lt; 0.001).\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6283338/v1/bdd1c31885a3d9f4482e2676.jpg"},{"id":80707437,"identity":"00bff274-9c21-4b9e-a85d-b1c8d8d02878","added_by":"auto","created_at":"2025-04-16 08:45:52","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":155986,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBrain Age Gap Predicts All-Cause Mortality Through Non-Linear Associations\u003c/strong\u003e\u003cbr\u003e\n \u003cstrong\u003e(A)\u003c/strong\u003e Kaplan-Meier survival curves stratified by BAG quartiles (\u003cem\u003ep\u003c/em\u003e\u0026lt; 0.001, two-sided log-rank test with multivariable adjustment for chronological age, sex, BMI, smoking status, and alcohol consumption). \u003cstrong\u003e(B)\u003c/strong\u003e Restricted cubic splines (RCS) analysis showing non-linear association between BAG and mortality risk (likelihood ratio test \u003cem\u003ep\u003c/em\u003e\u0026lt;0.001 for non-linearity; knots=1). Hazard ratios (HR) and 95% confidence intervals (CI) were estimated using Cox proportional hazards regression with RCS.\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6283338/v1/3fcd2d41ee0fbf3c2d89c3a5.jpg"},{"id":80707440,"identity":"871f8386-db31-4028-9837-a016c563bb1d","added_by":"auto","created_at":"2025-04-16 08:45:52","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":313088,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAssociations between lifestyle factors and BAG across quartiles: (A) \u003c/strong\u003eModerate Alcohol Consumption. \u003cstrong\u003e(B) \u003c/strong\u003eNever Smoking. \u003cstrong\u003e(C) \u003c/strong\u003eLow-to-Moderate Sedentary Behavior. \u003cstrong\u003e(D) \u003c/strong\u003eRegular Physical Activity. \u003cstrong\u003e(E) \u003c/strong\u003eHealthy Sleep. \u003cstrong\u003e(F) \u003c/strong\u003eHealthy Diet. \u003cstrong\u003e(G) \u003c/strong\u003eFrequent Social Connection. \u003cstrong\u003e(H) \u003c/strong\u003eOverall Lifestyle Classification. Panels (A) through (G) illustrate the individual effects of specific lifestyle factors on BAG, while panel (H) demonstrates the combined influence of these factors, categorized into three lifestyle classes: favorable (5–7 points), intermediate (2–4 points), and unfavorable (0–1 points). Statistical significance levels are indicated as follows: *p \u0026lt; 0.05, **p \u0026lt; 0.01, ***p \u0026lt; 0.001, and ns (not significant).\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6283338/v1/09b4fda792b2225ffaa5a223.jpg"},{"id":94432792,"identity":"10ebd53f-1804-42fd-aea4-eb53f79a7974","added_by":"auto","created_at":"2025-10-27 14:18:24","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3012784,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6283338/v1/3e55a758-f537-4782-b573-99c227e15caa.pdf"},{"id":80709114,"identity":"22fbc2cf-c258-4491-a29e-4cae5b56b080","added_by":"auto","created_at":"2025-04-16 08:53:52","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":1823977,"visible":true,"origin":"","legend":"Supplementary Tables and Figures for","description":"","filename":"Supplementary.docx","url":"https://assets-eu.researchsquare.com/files/rs-6283338/v1/ac02ab3afca93cc0e358e83c.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Brain Age Gap as a Predictive Biomarker: Linking Aging, Lifestyle, and Neuropsychiatric Health","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe aging brain undergoes progressive structural and functional changes, which can be categorized as either normative aging or accelerated aging, influenced by genetic factors, nutritional status, and environmental exposures\u003csup\u003e\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. The Brain Age Gap (BAG), defined as the difference between neuroimaging-derived brain age and chronological age, has been established as a robust biomarker for assessing accelerated brain aging\u003csup\u003e\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. Advances in structural MRI (sMRI)-based deep learning frameworks have significantly improved the precision of brain age estimation, enabling the classification of individuals into accelerated or delayed brain aging trajectories\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. BAG serves a dual purpose: it not only quantifies accelerated neuroaging but also predicts neuropsychiatric morbidity, offering a comprehensive metric for evaluating brain health\u003csup\u003e\u003cspan additionalcitationids=\"CR10\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. The integration of BAG into clinical practice enhances the understanding of aging trajectories and supports the development of personalized intervention strategies.\u003c/p\u003e \u003cp\u003eDespite the growing interest in brain age estimation, existing methods face significant limitations. Traditional approaches, such as convolutional neural networks (CNNs), often struggle to capture the complex spatial and temporal patterns of brain aging, particularly in the presence of disease-related structural changes\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e,\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. Moreover, many models are constrained by their reliance on small, homogeneous datasets, which limits their generalizability to diverse populations\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. To address these challenges, we developed a 3D Vision Transformer (3D-ViT) model\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e,\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e, which leverages the self-attention mechanism of transformers to better capture global and local neuroanatomical features associated with aging. This advanced architecture not only improves the accuracy of brain age estimation but also enhances the model\u0026rsquo;s ability to detect disease-specific patterns of brain aging, making it a powerful tool for both research and clinical applications.\u003c/p\u003e \u003cp\u003eBrain age estimation has become a critical tool in the study of neuropsychiatric diseases\u003csup\u003e\u003cspan additionalcitationids=\"CR19 CR20\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e, as well as in the investigation of cognitive\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e,\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e, physiological\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e, genetic\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e, and environmental factors\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e that influence brain health. As the human brain ages, it undergoes characteristic structural and functional changes, which can be accelerated or decelerated by various conditions, including neurodegenerative diseases, psychiatric disorders, and lifestyle factors. Numerous studies\u003csup\u003e\u003cspan additionalcitationids=\"CR28 CR29\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e have demonstrated that brain age estimation is instrumental in identifying risk and protective factors across the lifespan, providing valuable insights into the mechanisms underlying healthy and pathological aging. By monitoring changes in individual brain age estimates over time, researchers can assess the effectiveness of interventions, such as lifestyle modifications, pharmacological treatments, or cognitive training, in promoting brain health and mitigating age-related decline. This approach has the potential to facilitate the development of personalized strategies for healthy aging and the prevention of neuropsychiatric disorders.\u003c/p\u003e \u003cp\u003eWhile the relationship between brain age and the risk of neuropsychiatric disorders has garnered considerable attention, the specific role and predictive utility of the BAG in age-related diseases remain insufficiently elucidated. For instance, it remains uncertain whether BAG can serve as a reliable biomarker for predicting the risk of neuropsychiatric disorders or whether it can effectively monitor disease progression and treatment response. Additionally, the associations between BAG and diverse cognitive functions, as well as the potential of lifestyle interventions to mitigate increases in brain age, warrant further exploration. To address these gaps, we developed a 3D-ViT model for precise brain age estimation, leveraging MRI imaging to assess the brain\u0026rsquo;s aging status. By calculating the difference between estimated brain age and chronological age, our study aims to evaluate individuals\u0026rsquo; aging rates relative to their peers and to investigate the applicability of BAG in assessing the risk of neuropsychiatric disorders and all-cause mortality. Furthermore, we examined the relationship between BAG and cognitive functions and analyzed the potential impact of healthy lifestyle interventions on decelerating brain aging. This research seeks to clarify the role of BAG in neuropsychiatric risk assessment and to inform strategies for promoting cognitive health through lifestyle modifications, ultimately contributing to the development of non-invasive, cost-effective tools for early detection and intervention in age-related diseases.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cp\u003e\u003cstrong\u003e2.1 Study Design and Participants\u003c/strong\u003e\u003cbr\u003eWe developed and validated a brain age estimation model using multi-cohort neuroimaging data. The primary cohort included T1-weighted MRI scans from the UK Biobank\u003csup\u003e31\u003c/sup\u003e (UKB; application ID 89757). External validation was performed using the Alzheimer\u0026rsquo;s Disease Neuroimaging Initiative (ADNI)\u003csup\u003e32\u003c/sup\u003e and Parkinson\u0026rsquo;s Progression Markers Initiative (PPMI)\u003csup\u003e33\u003c/sup\u003e datasets. Ethical approval was obtained from the UK Biobank Access Committee and institutional review boards for ADNI/PPMI. All participants provided written informed consent.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCohort Descriptions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e1. \u003cstrong\u003ePrimary Cohort (UK Biobank):\u003c/strong\u003e\u003cbr\u003e\u0026nbsp;A total of 42,904 participants (age range: 45\u0026ndash;82 years; 52.54% female) with baseline T1-weighted MRI scans were initially included. To mitigate population stratification bias, analyses were restricted to 38,967 individuals of genetically confirmed White British ancestry. This cohort was stratified into three subgroups (Supplementary Table 1):\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cem\u003eHealthy controls (CN):\u003c/em\u003e 6,036 individuals without ICD-10 neurological/psychiatric diagnoses (G00\u0026ndash;G99, F00\u0026ndash;F99, I60-I63).\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cem\u003eNon-brain disease group:\u003c/em\u003e 25,416 individuals with non-neurological comorbidities.\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cem\u003eBrain disorder group\u003c/em\u003e\u003cem\u003e:\u003c/em\u003e 7,515 individuals with brain injury, neurodegeneration, or neuropsychiatric disorders (ICD-10 codes G00\u0026ndash;G99, F00\u0026ndash;F99, I60-I63, Supplementary Table 2).\u003c/p\u003e\n\u003cp\u003e2. \u003cstrong\u003eDisease Enrichment Cohorts:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eADNI Cohort\u003c/strong\u003e: A longitudinal multicenter study comprising 1,402 participants at baseline (CN: \u003cem\u003en\u003c/em\u003e = 332, 23.7%; MCI: \u003cem\u003en\u003c/em\u003e = 785, 56.0%; AD: \u003cem\u003en\u003c/em\u003e = 285, 20.3%, age range: 55\u0026ndash;96 years) with serial T1-weighted MRI scans. Over the follow-up period, the cohort contributed 5,058 MRI scans stratified as follows (Supplementary Table 3):\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cem\u003eCognitively Normal (CN):\u0026nbsp;\u003c/em\u003e1,480 scans (29.3% of total scans)\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cem\u003eMild Cognitive Impairment (MCI):\u0026nbsp;\u003c/em\u003e2,836 scans (56.1%)\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cem\u003eAlzheimer\u0026rsquo;s Disease (AD):\u0026nbsp;7\u003c/em\u003e42 scans (14.7%)\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eDiagnoses followed the National Institute on Aging\u0026ndash;Alzheimer\u0026rsquo;s Association (NIA-AA) criteria. Longitudinal data were harmonized across ADNI phases (ADNI1/GO/2/3) to account for scanner variability.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003ePPMI Cohort\u003c/strong\u003e: A longitudinal multicenter study comprising 1,182 participants at baseline (CN: \u003cem\u003en\u003c/em\u003e = 185, 15.7%; Prodromal-stage: \u003cem\u003en\u003c/em\u003e = 381, 32.2%; PD: \u003cem\u003en\u003c/em\u003e = 616, 52.1%, age range: 45\u0026ndash;83 years) with serial T1-weighted MRI scans. Over the follow-up period, the cohort contributed 2,907 MRI scans stratified as follows (Supplementary Table 4):\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cem\u003eHealthy Controls: (CN):\u0026nbsp;\u003c/em\u003e279 scans (9.6% of total scans)\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cem\u003eProdromal-stage (Prodromal:\u0026nbsp;\u003c/em\u003e1,326 scans (45.6%)\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cem\u003eParkinson\u0026rsquo;s Disease: (PD):\u0026nbsp;\u003c/em\u003e1,303 scans (44.8%)\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eDiagnoses followed the Movement Disorder Society (MDS) clinical diagnostic criteria for PD and prodromal markers. Longitudinal MRI data were harmonized across 50 participating sites using standardized preprocessing pipelines.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e3. \u003cstrong\u003eExternal Validation Cohorts:\u003c/strong\u003e\u003cbr\u003e\u0026nbsp;To evaluate model generalizability in neurologically intact populations, two independent datasets were analyzed:\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eADNI Validation Subset:\u003c/strong\u003e 1,480 scans from cognitively normal individuals (age 55\u0026ndash;96 years), excluding those with subsequent clinical progression.\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003ePPMI Healthy Controls:\u003c/strong\u003e 279 scans from baseline and longitudinal assessments of healthy participants (age 45\u0026ndash;83 years).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.2 MRI Acquisition and Preprocessing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eScanners:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eUKB:\u003c/strong\u003e 3T Siemens Skyra scanner equipped with a 32-channel head coil (repetition time [TR]/echo time [TE] = 2,000/2.01 ms).\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eADNI:\u003c/strong\u003e Multisite 3T scanners with harmonized imaging protocols.\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003ePPMI:\u003c/strong\u003e 3T Siemens Prisma scanner with a 64-channel head coil, utilizing the magnetization-prepared rapid gradient-echo (MPRAGE) sequence.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eProcessing Pipeline:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e1. \u003cstrong\u003eUK Biobank Data:\u003c/strong\u003e Preprocessed T1-weighted (T1w) scans were utilized, which were registered to the Montreal Neurological Institute (MNI152)\u003csup\u003e34, 35\u003c/sup\u003e standard space with an isotropic voxel size of 1 mm\u0026sup3;.\u003c/p\u003e\n\u003cp\u003e2. \u003cstrong\u003eExternal Cohorts:\u003c/strong\u003e Data were processed using FMRIB Software Library (FSL) version 6.0.5, with the following steps:\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eReorientation:\u003c/strong\u003e Raw MRI volumes were reoriented to the standard anatomical orientation.\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eCropping:\u003c/strong\u003e Images were cropped to remove the neck and subcranial structures.\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eBias Field Correction:\u003c/strong\u003e Bias field correction was applied without segmentation to maintain anatomical integrity.\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eNon-Brain Tissue Removal:\u003c/strong\u003e Non-brain tissues were removed using FSL\u0026rsquo;s Brain Extraction Tool (BET).\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eRegistration:\u003c/strong\u003e Linear and non-linear registration techniques (6 degrees of freedom) were employed to align images with the MNI152 standard template.\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eResampling:\u003c/strong\u003e Preprocessed MRI volumes were resampled to a voxel size of 182 \u0026times; 218 \u0026times; 182 (1 mm\u0026sup3; isotropic resolution).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eQuality Control:\u003c/strong\u003e Scans exhibiting motion artifacts (framewise displacement \u0026gt;0.5 mm) or registration errors (Dice coefficient \u0026lt;0.85 compared to the template) were excluded from further analysis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.3 Brain Age Estimation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Preprocessing and Model Architecture\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this study, we trained a 3D-ViT model on MRI samples from the UKB for brain age estimation (Fig. 1A, Supplementary Table 5). The ViT model has demonstrated its remarkable performance improvements compared to traditional machine learning models and CNN-based architectures in various medical image analysis tasks. Briefly, we first resized T1 weighted MR images to a resolution of 128 \u0026times; 128 \u0026times; 128 as input to the model. We utilized the multi-head attention mechanism of the 3D-ViT to divide the original MRI image into a set of 3D-patches (patch size: 16 \u0026times; 16 \u0026times; 16). These patches were processed through a transformer encoder, which measured attention scores between them. A multilayer perceptron header was then employed to summarize the outputs from the transformer encoder for brain age estimation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTraining Strategy and Validation\u003c/strong\u003e\u003cbr\u003e\u0026nbsp;The model was optimized by minimizing the mean squared error (MSE) between predicted brain age and chronological age using the Adam optimizer (learning rate: 2 \u0026times; 10⁻⁵, max epochs: 200). Training utilized 25,162 participants (80% of total N=31,452), comprising 6,036 healthy controls and 19,126 non-brain disorder subjects, with early stopping triggered if validation loss plateaued for 10 consecutive epochs. A nested 5-fold cross-validation protocol was implemented: at each fold, the training cohort was randomly divided into five subsets (4 for parameter optimization, 1 for internal validation). Following cross-validation, the finalized model was evaluated on an independent test set (n=6,290, 20% of total cohort) and externally validated using ADNI and PPMI datasets. Performance metrics included mean absolute error (MAE), MSE, R\u0026sup2;, and Pearson\u0026apos;s correlation coefficient (r) between predicted and chronological ages (Supplementary Table 6). Computational implementation employed Python 3.9 with PyTorch 1.13.1 on CUDA 8.6-enabled hardware.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.4 Quartile-Based Stratification of BAG\u003c/strong\u003e\u003cstrong\u003e\u003cbr\u003e\u003c/strong\u003eThe Brain Age Gap (BAG), calculated as the disparity between neuroimaging-derived brain age and chronological age, provides a biomarker of accelerated neurobiological aging. Elevated BAG values (positive deviations) reflect disproportionately rapid brain aging relative to biological age, whereas negative values indicate neuroprotective resilience. To refine its clinical utility in predicting age-related disease trajectories, we stratified participants into quartiles based on population-level BAG distributions: Q1 (slowest aging, 1st quartile), Q2 (2nd quartile), Q3 (3rd quartile), and Q4 (fastest aging, 4th quartile). This risk stratification framework enhances prognostic precision for neuropsychiatric disorders while enabling targeted prioritization of high-risk subgroups (Q4) in population health frameworks.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.5 Association Between BAG and Cognitive Performance\u003c/strong\u003e\u003cstrong\u003e\u003cbr\u003e\u003c/strong\u003eWe analyzed multi-domain cognitive assessments from 25,617 UK Biobank participants (mean age 64.5\u0026plusmn;7.6 years; 52.3% female) across seven core domains\u003csup\u003e37,38\u003c/sup\u003e: short-term memory, visual reasoning, abstract reasoning, processing speed, associative learning, executive function, and visual memory. \u003cstrong\u003eCognitive measures included:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u0026bull; Numeric Memory Test\u0026nbsp;(NMT): Digit span recall (0\u0026ndash;12)\u003c/p\u003e\n\u003cp\u003e\u0026bull; Matrix Pattern Completion\u0026nbsp;(MPCT): Visual pattern accuracy (0\u0026ndash;15)\u003c/p\u003e\n\u003cp\u003e\u0026bull; Fluid Intelligence Test\u0026nbsp;(FIT): Abstract problem-solving (13 items)\u003c/p\u003e\n\u003cp\u003e\u0026bull; Reaction Time Test\u0026nbsp;(RTT): Median response latency (ms)\u003c/p\u003e\n\u003cp\u003e\u0026bull; Paired Associate Learning\u0026nbsp;(PALT): Word-image pairs recalled (0\u0026ndash;10)\u003c/p\u003e\n\u003cp\u003e\u0026bull; Tower Rearrangement\u0026nbsp;(TR): Optimal puzzle solutions (3\u0026ndash;24 steps)\u003c/p\u003e\n\u003cp\u003e\u0026bull; Symbol Digit Substitution\u0026nbsp;(SDST): Correct matches/minute\u003c/p\u003e\n\u003cp\u003e\u0026bull; Pairs Matching Test\u0026nbsp;(PMT): Visual recognition accuracy (0\u0026ndash;6)\u003c/p\u003e\n\u003cp\u003eSample sizes varied slightly across tests (n = 24,872\u0026ndash;25,617) due to incomplete responses, with missing data addressed through\u0026nbsp;listwise deletion\u0026nbsp;to maintain analytical rigor (2.9\u0026ndash;5.2% missingness per test). To account for potential bias from missing data, we performed\u0026nbsp;multiple imputation via chained equations\u003csup\u003e39\u003c/sup\u003e (MICE; 20 iterations), incorporating MRI quality parameters, socioeconomic covariates, and cognitive test performance metrics. Imputation results were consistent with primary analyses (Supplementary Table 7). To characterize nonlinear associations between BAG and cognitive performance, we implemented restricted cubic splines (RCS) regression\u003csup\u003e40,41\u003c/sup\u003e. BAG was stratified into\u0026nbsp;population quartiles (Q1: slowest aging to Q4: fastest aging), and knot positions were optimized using the\u0026nbsp;Akaike Information Criterion (AIC)\u003csup\u003e42,43\u003c/sup\u003e to ensure appropriate smoothing. Models were adjusted for chronological age, sex, BMI, smoking status, and alcohol consumption to mitigate potential confounding. The inflection points of the dose-response curves were identified to determine critical thresholds of BAG associated with cognitive decline. Statistical significance was assessed using 95% confidence intervals, and effect sizes were standardized to facilitate cross-domain comparisons. Sensitivity analyses confirmed the robustness of findings across imputed and complete-case datasets.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.6 BAG as a Risk Marker for Neurodegenerative, Psychiatric, and Neurological Disorders\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCase Definitions\u003c/strong\u003e\u003cbr\u003e\u0026nbsp;We categorized 16 neuropsychiatric disorders into three pathophysiological groups based on mechanistic profiles (Supplementary Table 2, 3 and 4):\u003c/p\u003e\n\u003cp\u003e1. \u003cstrong\u003eNeurodegenerative disorders\u003c/strong\u003e: Alzheimer\u0026apos;s disease (AD; ICD-10: G30), vascular dementia (VaD; F01), and Prodromal Parkinson\u0026apos;s and Parkinson\u0026apos;s disease (PD; G20)\u003c/p\u003e\n\u003cp\u003e2. \u003cstrong\u003ePsychiatric disorders\u003c/strong\u003e: Major depressive disorder (MDD; F32), anxiety disorders (F41), bipolar disorder (BD; F31), anorexia nervosa (AN; F50.0), obsessive-compulsive disorder (OCD; F42), post-traumatic stress disorder (PTSD; F43.1), and schizophrenia (SCZ; F20)\u003c/p\u003e\n\u003cp\u003e3. \u003cstrong\u003eNeurological disorders\u003c/strong\u003e: Stroke (I63), epilepsy (G40), multiple sclerosis (MS; G35), and sleep disorders (SD; G47)\u003c/p\u003e\n\u003cp\u003eCases were identified through ICD-10 codes from hospital records. In the brain disorder cohort (N=7,515), prevalence ranged from 26 cases (schizophrenia) to 4,490 cases (major depressive disorder).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCox Regression and Follow-up Parameters\u003c/strong\u003e\u003cbr\u003eKey analytical parameters (Supplementary Tables 11-13) included\u003csup\u003e46,47\u003c/sup\u003e:\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eCohort characteristics\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003e\u0026bull; Baseline distribution across BAG quartiles (Q1-Q4)\u003c/p\u003e\n\u003cp\u003e\u0026bull; Incident vs censored case proportions\u003c/p\u003e\n\u003cp\u003e\u0026bull; Median follow-up durations:\u003cbr\u003e\u0026nbsp;\u0026bull; UK Biobank: 9.00-9.07 years (max 26.97)\u003cbr\u003e\u0026nbsp;\u0026bull; ADNI: 1.91-3.04 years (max 10.50)\u003cbr\u003e\u0026nbsp;\u0026bull; PPMI: 2.43-2.59 years (fixed 2-year intervals)\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eExposure quantification\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003e\u0026bull; Person-years accumulation patterns\u003c/p\u003e\n\u003cp\u003e\u0026bull; Observation windows (min-max)\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eDisease-specific attrition\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003e\u0026bull; Highest in UKB depression cohort (4,490 incident cases)\u003c/p\u003e\n\u003cp\u003e\u0026bull; Longest tracking in UKB stroke cohort (median 9.90 person-years)\u003c/p\u003e\n\u003cp\u003ePrimary exposure was defined as 1-year increase in BAG. All models adjusted for chronological age, sex, BMI, smoking status, and alcohol consumption.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.7 Survival Analysis of BAG and Mortality Risk\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe conducted survival analyses using Cox proportional hazards regression, with age as the underlying time scale. The study cohort comprised 38,967 UK Biobank participants with complete mortality linkage data obtained from national death registries. Detailed analytic parameters are provided in\u0026nbsp;Supplementary Tables 16-18. The primary exposure variable was defined as a 1-year increase in the BAG. All models were adjusted for chronological age, sex, body mass index (BMI), smoking status, and alcohol consumption.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCensoring and Validation\u003c/strong\u003e\u003cbr\u003e\u0026nbsp;Participants were right-censored under the following conditions:\u003c/p\u003e\n\u003cp\u003e1. \u003cstrong\u003eDeath verification\u003c/strong\u003e: Confirmed through registry linkage.\u003c/p\u003e\n\u003cp\u003e2. \u003cstrong\u003eLoss to follow-up\u003c/strong\u003e: Occurred in less than 0.2% of participants.\u003c/p\u003e\n\u003cp\u003e3. \u003cstrong\u003eStudy termination\u003c/strong\u003e: The end of the study period (October 23, 2021).\u003c/p\u003e\n\u003cp\u003eTo ensure the validity of the proportional hazards assumption, we conducted systematic validation using the following methods:\u003c/p\u003e\n\u003cp\u003e1. \u003cstrong\u003eSchoenfeld residuals\u003c/strong\u003e: Global and covariate-specific analyses were performed.\u003c/p\u003e\n\u003cp\u003e2. \u003cstrong\u003eStratified survival analysis:\u0026nbsp;\u003c/strong\u003eKaplan-Meier curves stratified by BAG quartiles demonstrated proportional hazards across strata (log-rank test: p \u0026lt; 0.001; multivariable-adjusted for chronological age, sex, BMI, smoking status, and alcohol consumption).\u003c/p\u003e\n\u003cp\u003e3. \u003cstrong\u003eNon-linear association testing\u003c/strong\u003e: Restricted cubic splines (RCS) revealed significant non-linearity in the BAG-mortality relationship (likelihood ratio test for non-linearity: p \u0026lt; 0.001). Cox proportional hazards regression with RCS-derived hazard ratios (HR) and 95% confidence intervals (CI) confirmed this pattern.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStratified Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCensoring patterns across BAG quartiles revealed a dose-dependent relationship with mortality risk, with the lowest risk observed in the first quartile (Q1: 0.8%) and the highest in the fourth quartile (Q4: 1.8%) (Supplementary Tables 17). Follow-up duration was consistent across quartiles, with a median of 3.4 years (interquartile range [IQR]: 1.8\u0026ndash;5.1 years).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.8 Lifestyle-BAG Interaction Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eComposite Lifestyle Score Development\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe constructed a 7-domain lifestyle score\u003csup\u003e48-51\u003c/sup\u003e for 38,967 participants using validated criteria:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 1. Seven modifiable factors were scored (0-1 per domain)\u003c/strong\u003e\u003c/p\u003e\n\u003cdiv align=\"center\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"106%\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eFactor\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eHealthy Criteria\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eAssessment Tool\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSmoking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eNever smoker\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eNHS/NICE 2015 guidelines\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePhysical activity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ge;150 min/week moderate or \u0026ge;75 min vigorous\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eIPAQ-SF\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eAlcohol\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026le;14 g/day (women), \u0026le;28 g/day (men)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eU.S. Dietary Guidelines\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eDiet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ge;4/7 healthy food groups\u0026dagger;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e24-hr dietary recall\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSleep\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e7-9 hours/night\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAASM/SRS consensus\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSedentary time\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026le;4 hr/day non-work screen time\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eGPAQ\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSocial connection\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSocial isolation index \u0026le;1\u0026Dagger;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSocial network survey\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cem\u003e\u0026dagger;Fruits, vegetables, fish, whole grains, unprocessed meats,\u003c/em\u003e \u003cem\u003eLow-fat dairy products,\u003c/em\u003e \u003cem\u003eNuts\u003cbr\u003e\u0026nbsp;\u0026Dagger;Composite of household size, social visit frequency, activity participation\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe composite lifestyle score demonstrated a continuous distribution from 0 (least healthy) to 7 (most healthy). Categorical stratification was performed as:\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eUnfavorable\u003c/strong\u003e: 0-1\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eIntermediate\u003c/strong\u003e: 2-4\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eFavorable\u003c/strong\u003e: 5-7\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBAG Interaction Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo examine the interaction between lifestyle and the BAG, the following analytical approach was implemented:\u003c/p\u003e\n\u003cp\u003e1. \u003cstrong\u003eStratification\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u0026bull; Participants were grouped into BAG quartiles (Q1\u0026ndash;Q4) based on their BAG values.\u003c/p\u003e\n\u003cp\u003e2. \u003cstrong\u003eComparative Testing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u0026bull; Between-group comparisons: Student\u0026rsquo;s t-test was used to assess differences between healthy (favorable lifestyle) and unhealthy (unfavorable lifestyle) groups.\u003c/p\u003e\n\u003cp\u003e\u0026bull; Cross-quartile trends: The Kruskal-Wallis test\u003csup\u003e52\u003c/sup\u003e was applied to evaluate trends across BAG quartiles, with Dunn\u0026rsquo;s post-hoc adjustment for multiple comparisons to identify specific differences in lifestyle score distributions.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec31\" class=\"Section2\"\u003e \u003ch2\u003ePerformance of 3D-ViT in Brain Age Estimation\u003c/h2\u003e \u003cdiv id=\"Sec32\" class=\"Section3\"\u003e \u003ch2\u003eOverall Model Performance\u003c/h2\u003e \u003cp\u003eThe 3D-ViT model was trained on structural MRI data from 25,162 healthy adults in the UKB, demonstrating robust performance in brain age estimation (Figs.\u0026nbsp;1; Supplementary Table\u0026nbsp;6). On the training set, the model achieved a mean absolute error (MAE) of 2.627 years (MSE\u0026thinsp;=\u0026thinsp;10.988, R\u0026sup2;=0.810, Pearson's \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.901), indicating strong concordance between predicted and chronological ages (Supplementary Fig.\u0026nbsp;1). Five-fold cross-validation confirmed stability (average MAE\u0026thinsp;=\u0026thinsp;2.610 years) without evidence of overfitting. In an independent UKB test cohort (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;6,290), performance remained consistent (MAE\u0026thinsp;=\u0026thinsp;2.684 years, R\u0026sup2;=0.805, \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.898; Supplementary Fig.\u0026nbsp;2B), underscoring generalizability. Gender-stratified analyses revealed comparable accuracy for females (MAE\u0026thinsp;=\u0026thinsp;2.567 years, \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.902) and males (MAE\u0026thinsp;=\u0026thinsp;2.714 years, \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.897), with a minimal intergroup difference (\u0026lt;\u0026thinsp;0.2 years; Supplementary Figs.\u0026nbsp;2E-F). Age-stratified evaluations demonstrated MAE values of 2.609 years (\u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.760) for middle-aged participants (\u0026lt;\u0026thinsp;65 years) and 2.666 years (\u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.687) for elderly individuals (\u0026ge;\u0026thinsp;65 years; Supplementary Figs.\u0026nbsp;2). Overall, the model exhibited consistent accuracy and robustness across demographic subgroups.\u003c/p\u003e \u003cdiv id=\"Sec33\" class=\"Section4\"\u003e \u003ch2\u003eCross-Dataset Validation\u003c/h2\u003e \u003cp\u003eExternal validation using ADNI (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1,480 cognitively normal subjects) and PPMI (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;279 healthy controls) yielded MAE values of 2.998 years (MSE\u0026thinsp;=\u0026thinsp;16.000, \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.758; Supplementary Fig.\u0026nbsp;3) and 3.205 years (MSE\u0026thinsp;=\u0026thinsp;17.640, \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.936; Supplementary Fig.\u0026nbsp;4), respectively. The marginally higher error in ADNI (Δ\u0026thinsp;\u0026asymp;\u0026thinsp;0.3 years vs. UKB) likely originated from protocol differences and an older population baseline, while PPMI's elevated correlation may reflect its narrower age distribution. Optimal performance in UKB was attributable to its expansive sample size and broad age representation. These findings collectively validate the model's cross-dataset adaptability, albeit with protocol- and cohort-dependent variance.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec34\" class=\"Section3\"\u003e \u003ch2\u003eDisease-Associated Brain Age Acceleration\u003c/h2\u003e \u003cp\u003eThe model systematically overestimated brain age in neurodegenerative disorders, with error magnitudes correlating with disease severity. In ADNI cohorts, mild cognitive impairment (MCI) subjects (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2,836) exhibited an MAE of 4.77 years (1.6\u0026times; controls; \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.675; Supplementary Fig.\u0026nbsp;3), while AD dementia patients (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;742) showed pronounced acceleration (MAE\u0026thinsp;=\u0026thinsp;6.08 years, 2\u0026times; controls; \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.51; Supplementary Fig.\u0026nbsp;3), with predicted brain ages exceeding chronological ages by ~\u0026thinsp;6 years. This pattern aligns with AD-specific atrophy trajectories diverging from healthy aging. PD patients (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1,302) displayed modest error increases (MAE\u0026thinsp;=\u0026thinsp;3.48 years, +\u0026thinsp;0.3 vs. controls; \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.960; Supplementary Fig.\u0026nbsp;4). Notably, UKB subgroups with mixed neurological conditions (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;7,515) maintained robust performance (MAE\u0026thinsp;=\u0026thinsp;2.77 years vs. 2.68 in controls; \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.895), suggesting resilience to multifactorial pathologies.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e\n\u003ch3\u003eAssociation between BAG and cognitive decline\u003c/h3\u003e\n\u003cp\u003eCognitive performance was stratified by BAG quartiles in the UK Biobank cohort (n\u0026thinsp;=\u0026thinsp;25,617, Supplementary Table\u0026nbsp;8). Significant differences were observed across quartiles for several cognitive domains. Matrix pattern completion scores were higher in Q1 (8.10\u0026thinsp;\u0026plusmn;\u0026thinsp;2.10) and Q2 (8.12\u0026thinsp;\u0026plusmn;\u0026thinsp;2.07) compared to Q3 (7.95\u0026thinsp;\u0026plusmn;\u0026thinsp;2.11) and Q4 (7.95\u0026thinsp;\u0026plusmn;\u0026thinsp;2.12) (F\u0026thinsp;=\u0026thinsp;12.40, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Reaction time was significantly slower in Q4 (604.88\u0026thinsp;\u0026plusmn;\u0026thinsp;112.78 ms) compared to Q1 (591.74\u0026thinsp;\u0026plusmn;\u0026thinsp;106.26 ms) (F\u0026thinsp;=\u0026thinsp;19.27, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Paired associate learning scores were lower in Q4 (6.85\u0026thinsp;\u0026plusmn;\u0026thinsp;2.52) compared to Q1 (7.14\u0026thinsp;\u0026plusmn;\u0026thinsp;2.52) (F\u0026thinsp;=\u0026thinsp;19.82, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Symbol digit substitution scores were also lower in Q4 (18.45\u0026thinsp;\u0026plusmn;\u0026thinsp;5.21) compared to Q1 (19.29\u0026thinsp;\u0026plusmn;\u0026thinsp;5.15) (F\u0026thinsp;=\u0026thinsp;40.67, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). No significant differences were observed for numeric memory (F\u0026thinsp;=\u0026thinsp;0.52, p\u0026thinsp;=\u0026thinsp;0.671) or fluid intelligence (F\u0026thinsp;=\u0026thinsp;1.93, p\u0026thinsp;=\u0026thinsp;0.122).\u003c/p\u003e \u003cp\u003eLinear regression models revealed significant associations between biological aging quartiles and cognitive performance (Supplementary Table\u0026nbsp;9 and Supplementary Fig.\u0026nbsp;6). Compared to Q1, Q3 and Q4 showed significantly lower scores in matrix pattern completion (Q3: β = -0.15, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001; Q4: β = -0.15, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), paired associate learning (Q3: β = -0.14, p\u0026thinsp;=\u0026thinsp;0.001; Q4: β = -0.28, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and symbol digit substitution (Q3: β = -0.52, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001; Q4: β = -0.84, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Reaction time was significantly slower in Q3 (β\u0026thinsp;=\u0026thinsp;8.06, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and Q4 (β\u0026thinsp;=\u0026thinsp;13.15, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) compared to Q1. No significant associations were found for numeric memory or fluid intelligence after adjusting for multiple comparisons using the Benjamini-Hochberg procedure.\u003c/p\u003e \u003cp\u003eNon-linear modeling identified inflection points and maximum effects of BAG on cognitive domains (Fig.\u0026nbsp;2 and Supplementary Table\u0026nbsp;10). The maximum effects for all cognitive domains were observed in Q4, indicating that the most pronounced cognitive deficits occurred in individuals with the highest BAG. For example, reaction time exhibited an inflection point at BAG \u0026asymp; -1.75, with a maximum effect of 3.13 ms in Q4. Similarly, paired associate learning showed a maximum effect of 3.11 in Q4, with an inflection point at BAG\u0026thinsp;\u0026asymp;\u0026thinsp;0.01. These findings suggest that cognitive performance remains relatively stable at lower BAG levels but deteriorates more rapidly beyond a critical threshold, particularly in Q4.\u003c/p\u003e\n\u003ch3\u003eBAG as a Risk Marker for Neuropsychiatric Disorders\u003c/h3\u003e\n\u003cp\u003eCross-sectional analysis based on the BAG revealed significant heterogeneity in BAG levels between disease groups and the healthy control group (CN, BAG: 0.122\u0026thinsp;\u0026plusmn;\u0026thinsp;3.039; Fig.\u0026nbsp;3B, Supplementary Table\u0026nbsp;14). Among neurodegenerative disorders, Alzheimer\u0026rsquo;s disease (AD: 3.242\u0026thinsp;\u0026plusmn;\u0026thinsp;6.635, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and mild cognitive impairment (MCI: 2.063\u0026thinsp;\u0026plusmn;\u0026thinsp;5.619, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) exhibited significantly elevated BAG, whereas vascular dementia (VaD: 0.279\u0026thinsp;\u0026plusmn;\u0026thinsp;3.463, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.442) and Parkinson\u0026rsquo;s disease (PD: -0.003\u0026thinsp;\u0026plusmn;\u0026thinsp;4.463, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.321) showed no significant differences compared to CN. Notably, prodromal Parkinson\u0026rsquo;s disease (ProdPD: -1.441\u0026thinsp;\u0026plusmn;\u0026thinsp;4.880, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) demonstrated a unique negative BAG shift. In psychiatric disorders, bipolar disorder (BD: 1.913\u0026thinsp;\u0026plusmn;\u0026thinsp;4.051, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), major depressive disorder (MDD: 0.516\u0026thinsp;\u0026plusmn;\u0026thinsp;3.409, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and anxiety disorder (ANX: 0.525\u0026thinsp;\u0026plusmn;\u0026thinsp;3.425, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) displayed significantly higher BAG than CN, whereas obsessive-compulsive disorder (OCD: 1.161\u0026thinsp;\u0026plusmn;\u0026thinsp;3.231, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.074), post-traumatic stress disorder (PTSD: 0.310\u0026thinsp;\u0026plusmn;\u0026thinsp;3.539, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.645), and anorexia nervosa (AN: 0.369\u0026thinsp;\u0026plusmn;\u0026thinsp;3.547, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.631) showed no significant differences. Schizophrenia (SCZ: 2.068\u0026thinsp;\u0026plusmn;\u0026thinsp;3.341, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.007) exhibited elevated BAG despite a limited sample size (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;26). Among neurological disorders, multiple sclerosis (MS: 4.069\u0026thinsp;\u0026plusmn;\u0026thinsp;5.328, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) showed the most pronounced BAG elevation, with stroke (Stroke: 0.631\u0026thinsp;\u0026plusmn;\u0026thinsp;3.585, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), epilepsy (Epilepsy: 1.059\u0026thinsp;\u0026plusmn;\u0026thinsp;3.478, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and sleep disorders (SD: 0.641\u0026thinsp;\u0026plusmn;\u0026thinsp;3.379, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) also demonstrating significantly higher BAG compared to controls.\u003c/p\u003e \u003cp\u003eOur study investigated the association between BAG and the risk of 16 neuropsychiatric disorders (Fig.\u0026nbsp;3C, Supplementary Fig.\u0026nbsp;7 and Supplementary Tables\u0026nbsp;15), categorized into neurodegenerative, psychiatric, and neurological groups, using Cox regression models adjusted for chronological age, sex, BMI, smoking status, and alcohol consumption. Elevated BAG was strongly associated with neurodegenerative diseases, particularly Alzheimer\u0026rsquo;s disease (AD) and mild cognitive impairment (MCI), with each 1-year increase in BAG conferring a 16.5% higher risk of AD (HR\u0026thinsp;=\u0026thinsp;1.165, 95% CI\u0026thinsp;=\u0026thinsp;1.086\u0026ndash;1.249, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and a 4.0% higher risk of MCI (HR\u0026thinsp;=\u0026thinsp;1.040, 1.030\u0026ndash;1.050, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Participants in the highest BAG quartile (Q4) had 2.8-fold increased AD risk (HR\u0026thinsp;=\u0026thinsp;2.801, p\u0026thinsp;=\u0026thinsp;0.011) and 1.7-fold increased MCI risk (HR\u0026thinsp;=\u0026thinsp;1.691, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) compared to Q1, while vascular dementia (VaD) showed no significant BAG association (HR\u0026thinsp;=\u0026thinsp;1.022, p\u0026thinsp;=\u0026thinsp;0.323), aligning with its distinct cerebrovascular pathophysiology. Parkinson\u0026rsquo;s disease (PD) exhibited mixed results: prodromal PD had reduced BAG (mean\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;1.441 vs. controls, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), while incident PD showed no significant association (HR\u0026thinsp;=\u0026thinsp;1.038, p\u0026thinsp;=\u0026thinsp;0.385), reflecting its focal nigrostriatal degeneration. In psychiatric disorders, BAG independently predicted major depressive disorder (MDD) (HR\u0026thinsp;=\u0026thinsp;1.046 per year, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), bipolar disorder (BD) (HR\u0026thinsp;=\u0026thinsp;1.174, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and anxiety disorders (ANX) (Q4 HR\u0026thinsp;=\u0026thinsp;1.442, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), with Q4 participants facing 1.5-fold higher MDD risk (HR\u0026thinsp;=\u0026thinsp;1.466, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and 2.4-fold higher BD risk (HR\u0026thinsp;=\u0026thinsp;2.431, p\u0026thinsp;=\u0026thinsp;0.038) versus Q1. Schizophrenia (SCZ) risk escalated sharply in Q4 (HR\u0026thinsp;=\u0026thinsp;7.504, p\u0026thinsp;=\u0026thinsp;0.022), albeit with limited case numbers (n\u0026thinsp;=\u0026thinsp;26). Among neurological disorders, stroke risk increased by 5.5% per BAG year (HR\u0026thinsp;=\u0026thinsp;1.055, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), with Q4 participants facing 1.6-fold higher risk (HR\u0026thinsp;=\u0026thinsp;1.612, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), while multiple sclerosis (MS) demonstrated the most pronounced BAG association (Q4 HR\u0026thinsp;=\u0026thinsp;6.417, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), consistent with its inflammatory-driven accelerated aging. Epilepsy risk doubled in Q4 (HR\u0026thinsp;=\u0026thinsp;2.189, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), likely reflecting cumulative neurotoxic effects.\u003c/p\u003e \u003cdiv id=\"Sec37\" class=\"Section2\"\u003e \u003ch2\u003eSurvival analysis of BAG and all-cause mortality\u003c/h2\u003e \u003cp\u003eWe conducted a comprehensive survival analysis to evaluate the association between BAG and all-cause mortality using data from national death registries linked to participant records in the UKB. The cohort (Supplementary Tables\u0026nbsp;16\u0026ndash;17) was followed for a median of 3.4 years (range: 0\u0026ndash;7.5 years), with a total follow-up time of 148,774 person-years. During the follow-up period, 469 participants (1.2%) died from various causes, while 38,498 participants (98.8%) were right-censored. Censoring rates exhibited a graded pattern across BAG quartiles: Q1 (99.2%), Q2 (99.1%), Q3 (98.8%), and Q4 (98.2%). Mortality rates increased progressively across BAG quartiles: Q1 (0.8%), Q2 (0.9%), Q3 (1.2%), and Q4 (1.8%). Cox regression models adjusted for chronological age, sex, BMI, smoking, and alcohol consumption revealed significant mortality associations. Per 1-year increase in BAG conferred a 12% elevated mortality risk (adjusted HR 1.12, 95% CI 1.09\u0026ndash;1.15; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Quartile-based analysis demonstrated a dose-response relationship: compared with Q1 (reference), Q2 showed non-significant risk elevation (HR 1.18, 0.87\u0026ndash;1.59; p\u0026thinsp;=\u0026thinsp;0.293), while Q3 (HR 2.11, 1.19\u0026ndash;2.59; p\u0026thinsp;=\u0026thinsp;0.002) and Q4 (HR 2.36, 1.81\u0026ndash;3.08; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) exhibited progressively stronger associations. The proportional hazards assumption remained valid across all covariates (global Schoenfeld test p\u0026thinsp;=\u0026thinsp;0.88; covariate-specific p\u0026thinsp;\u0026gt;\u0026thinsp;0.68, Supplementary Tables\u0026nbsp;18). Kaplan-Meier survival curves confirmed significant inter-quartile divergence (log-rank p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Q4 participants demonstrated substantially reduced survival probability versus Q1 (absolute difference 3.0% at 7 years), with intermediate quartiles showing graded risk progression (Fig.\u0026nbsp;4A). Nonlinear restricted cubic spline analysis demonstrated a robust and statistically significant association between BAG and mortality risk (Fig.\u0026nbsp;4B, likelihood ratio test for nonlinearity, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Mortality risk remained stable at BAG values below 2 years (HR\u0026thinsp;\u0026asymp;\u0026thinsp;1.0) but increased sharply thereafter, with a HR of 1.2 at 3 years and exceeding 2.0 at BAG values of 6 to 8 years. These findings underscore the importance of monitoring and addressing elevated BAG levels to mitigate associated mortality risks.\u003c/p\u003e \u003cdiv id=\"Sec38\" class=\"Section3\"\u003e \u003ch2\u003eLifestyle-BAG Interaction\u003c/h2\u003e \u003cp\u003eStratified analyses demonstrated a significant dose-response relationship between composite lifestyle scores and BAG progression across quartiles (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001 for trend; Fig.\u0026nbsp;5, Supplementary Tables\u0026nbsp;19\u0026ndash;20). The strongest neuroprotective effects emerged in the highest brain-age correlation quartile (Q4): individuals with favorable lifestyles (scores 5\u0026ndash;7) exhibited a 0.35-year BAG reduction compared to those with unfavorable profiles (scores 0\u0026ndash;1) (4.15\u0026thinsp;\u0026plusmn;\u0026thinsp;1.63 vs. 4.50\u0026thinsp;\u0026plusmn;\u0026thinsp;1.81, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), while intermediate scores (2\u0026ndash;4) showed attenuated progression (annual reduction rate: 1.7%, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.013). Domain-specific analyses identified three key protective factors in Q4: moderate alcohol consumption demonstrated the most substantial effect (-0.20-year BAG reduction, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), followed by regular physical activity (-0.14-year, p\u0026thinsp;=\u0026thinsp;0.03), and never smoking (-0.11-year reduction; 4.16\u0026thinsp;\u0026plusmn;\u0026thinsp;1.65 vs. 4.27\u0026thinsp;\u0026plusmn;\u0026thinsp;1.69 in smokers, p\u0026thinsp;=\u0026thinsp;0.002). Stage-dependent patterns revealed transient sleep benefits restricted to Q1 (-0.09-year, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.03) and U-shaped social connectivity effects peaking in Q3 (1.13\u0026thinsp;\u0026plusmn;\u0026thinsp;0.61 vs. 1.27\u0026thinsp;\u0026plusmn;\u0026thinsp;0.62, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.012). Threshold phenomena were observed for sedentary behavior (2.3% BAG reduction in Q1 [\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.046], nonsignificant in Q3\u0026ndash;Q4), while dietary patterns showed no significant associations across strata (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026ge;\u0026thinsp;0.27). These results delineate a dynamic neuroprotective hierarchy, where lifestyle efficacy shifts across brain aging stages, emphasizing smoking abstinence and alcohol moderation as critical intervention targets in advanced neurodegeneration.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eOur study demonstrates the effectiveness of the proposed 3D-ViT framework in estimating brain age, revealing significant advantages over traditional CNN methods. The analysis identifies robust associations between an elevated BAG and domain-specific cognitive decline, increased risks of neuropsychiatric disorders, and heightened all-cause mortality within a large, population-based cohort. Furthermore, our findings indicate that healthy lifestyle interventions can effectively mitigate the negative impacts of accelerated brain aging. These results collectively support the utility of BAG as a reliable biomarker for assessing biological aging, identifying high-risk individuals, and guiding early clinical interventions and public health strategies aimed at promoting brain health and longevity.\u003c/p\u003e \u003cp\u003eThe proposed 3D-ViT framework for brain age estimation demonstrates superior performance in capturing comprehensive neuroanatomical features compared with conventional CNN approaches. Traditional CNN methods, although capable of extracting global features from whole-brain MRI, typically emphasize prominent regions and overlook subtle anatomical details\u003csup\u003e\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e53\u003c/span\u003e,\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e54\u003c/span\u003e\u003c/sup\u003e. Conversely, patch-based CNN methods retain localized anatomical information but suffer from contextual fragmentation due to limited receptive fields\u003csup\u003e\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e55\u003c/span\u003e\u003c/sup\u003e. Our 3D-ViT model addresses these limitations by integrating global and local neuroanatomical information through a self-attention mechanism, effectively capturing both broad structural patterns and fine-grained features within a unified architecture\u003csup\u003e\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e56\u003c/span\u003e\u003c/sup\u003e. This multi-scale feature fusion aligns closely with emerging hybrid methods that combine CNN-derived local descriptors and Transformer-based contextual aggregation\u003csup\u003e\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e57\u003c/span\u003e\u003c/sup\u003e but avoids the computational redundancy typically associated with multi-network ensembles. The proposed model achieves state-of-the-art performance in large-scale healthy populations and exhibits robust generalizability across external validation cohorts and disease-specific groups. Specifically, our 3D-ViT model demonstrates several key advantages: (1) it leverages Transformer-based architectures to analyze whole-brain 3D MRI scans, yielding a mean absolute error (MAE) of only 2.68 years in healthy adults, substantially outperforming conventional CNN methods (e.g., 2D ResNet\u003csup\u003e\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e58\u003c/span\u003e\u003c/sup\u003e: MAE\u0026thinsp;=\u0026thinsp;6.8 years; 3D VGGNet\u003csup\u003e\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e59\u003c/span\u003e,\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e: MAE\u0026thinsp;=\u0026thinsp;4.45 years; 3D EfficientNet\u003csup\u003e\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e61\u003c/span\u003e\u003c/sup\u003e: MAE\u0026thinsp;=\u0026thinsp;3.31 years); (2) it maintains consistent and unbiased performance across diverse demographic groups (gender and age), highlighting the universality of learned neuroanatomical features; (3) it demonstrates robust cross-cohort generalization validated on external ADNI (MAE\u0026thinsp;=\u0026thinsp;2.99 years) and PPMI (MAE\u0026thinsp;=\u0026thinsp;3.20 years) datasets, despite variability in MRI acquisition protocols; and (4) it exhibits sensitivity to pathological brain aging, as evidenced by significantly elevated brain age estimates in patients with neuropsychiatric disorders, closely aligning with clinically observed cognitive decline and neurodegeneration trajectories. Collectively, these findings highlight the considerable potential of brain age estimation as an auxiliary clinical tool for disease screening and progression monitoring.\u003c/p\u003e \u003cp\u003eOur study provides robust evidence of a significant association between the BAG and domain-specific cognitive decline in the UK Biobank cohort. These findings enhance our understanding of neurobiological aging by identifying critical thresholds and nonlinear relationships between structural brain changes and cognitive performance. Stratification by BAG quartiles revealed a clear gradient in cognitive outcomes, with the most pronounced deficits observed in the highest quartile (Q4; BAG\u0026thinsp;\u0026gt;\u0026thinsp;2.48 years). Individuals in Q4 exhibited significantly slower reaction times, reduced executive function (symbol digit substitution: β = -0.84 vs. Q1), and impaired associative memory (paired associate learning: β = -0.28 vs. Q1). These results align with prior neuroimaging studies\u003csup\u003e\u003cspan additionalcitationids=\"CR63\" citationid=\"CR60\" class=\"CitationRef\"\u003e62\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e64\u003c/span\u003e\u003c/sup\u003e implicating frontostriatal and hippocampal networks in age-related cognitive decline. Importantly, our analysis indicates distinct cognitive trajectories across BAG quartiles. Accelerated brain aging correlates with measurable cognitive impairment even among apparently healthy middle-aged and older adults. Previous studies\u003csup\u003e\u003cspan additionalcitationids=\"CR66 CR67\" citationid=\"CR63\" class=\"CitationRef\"\u003e65\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e68\u003c/span\u003e\u003c/sup\u003e have reported similar associations, demonstrating that higher BAG is linked to poorer cognitive and motor functions in populations such as Parkinson\u0026rsquo;s disease patients and older adults with decreased processing speed and memory performance. Identifying domain-specific vulnerabilities through BAG highlights its potential as a biomarker for early detection of subclinical cognitive decline. Individuals surpassing a critical BAG threshold (\u0026gt;\u0026thinsp;2.48 years) appear especially vulnerable to substantial cognitive deterioration, suggesting a critical intervention window for targeted preventive strategies.\u003c/p\u003e \u003cp\u003eOur large-scale study demonstrates that the BAG exhibits distinct pathophysiological associations across neurodegenerative, psychiatric, and neurological disorders, while also reflecting shared mechanisms underlying neuropsychiatric conditions. The robust association between elevated BAG and AD (HR\u0026thinsp;=\u0026thinsp;1.165/year, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) aligns with prior evidence of global brain atrophy and metabolic dysregulation in AD pathogenesis\u003csup\u003e\u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e69\u003c/span\u003e\u003c/sup\u003e. Intriguingly, the paradoxical reduction in BAG observed in prodromal Parkinson\u0026rsquo;s disease (Q4 HR\u0026thinsp;=\u0026thinsp;0.925, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.105) may reflect compensatory neuroplasticity\u003csup\u003e\u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e70\u003c/span\u003e,\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e71\u003c/span\u003e\u003c/sup\u003e during early nigrostriatal degeneration, whereas the absence of BAG effects (Q4 HR\u0026thinsp;=\u0026thinsp;1.830, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.154) in incident PD suggests that focal dopaminergic loss minimally impacts global brain aging metrics\u003csup\u003e\u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e72\u003c/span\u003e\u003c/sup\u003e. The dose-dependent relationship between BAG and psychiatric disorders further highlights its transdiagnostic utility. For MDD, the quartile-dependent risk escalation (Q4 HR\u0026thinsp;=\u0026thinsp;1.466, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) supports the hypothesis of accelerated aging mediated by chronic stress-induced glucocorticoid toxicity and mitochondrial dysfunction\u003csup\u003e\u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e73\u003c/span\u003e,\u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e74\u003c/span\u003e\u003c/sup\u003e. Notably, SCZ exhibited an exponential risk surge in the highest BAG quartile (HR\u0026thinsp;=\u0026thinsp;7.504, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.022), suggesting that neurodevelopmental deficits in synaptic pruning may synergize with aging-related cortical thinning to amplify vulnerability\u003csup\u003e\u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e75\u003c/span\u003e,\u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e76\u003c/span\u003e\u003c/sup\u003e. In neurological disorders, the pronounced association between BAG and MS (Q4 HR\u0026thinsp;=\u0026thinsp;6.417, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) points to inflammaging\u0026mdash;a convergence of chronic neuroinflammation and epigenetic aging\u0026mdash;as a driver of disability progression\u003csup\u003e\u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e77\u003c/span\u003e,\u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e78\u003c/span\u003e\u003c/sup\u003e. Similarly, the 5.5% annual increase in stroke risk per BAG year (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) implicates endothelial glycocalyx degradation and blood-brain barrier disruption as mediators of neurovascular aging\u003csup\u003e\u003cspan citationid=\"CR77\" class=\"CitationRef\"\u003e79\u003c/span\u003e,\u003cspan citationid=\"CR78\" class=\"CitationRef\"\u003e80\u003c/span\u003e\u003c/sup\u003e. These findings support the concept of brain age as a comprehensive indicator of cerebral health: when the brain appears \"older\" than its chronological age, it often suggests the presence of underlying pathological processes or neurological damage. Although elevated brain age may not serve as a disease-specific marker but rather reflects a nonspecific indicator of cumulative pathological impacts on the brain, its significant association with disease risk provides critical insights for early identification of high-risk individuals. This research highlights the intricate interplay between systemic aging and neural circuitry, offering novel perspectives for the prevention and early intervention of neuropsychiatric disorders.\u003c/p\u003e \u003cp\u003eOur study provides compelling evidence that the BAG independently predicts all-cause mortality within a large, population-based cohort. Three key findings highlight the clinical and prognostic relevance of BAG. First, each 1-year increase in BAG corresponded to a 12% elevated risk of mortality, even after adjusting for chronological age, sex, BMI, smoking status, and alcohol consumption. Second, a clear dose-response relationship was evident across BAG quartiles; participants in Q4 exhibited a 2.36-fold higher mortality risk compared to those in Q1. Third, nonlinear threshold effects were observed, indicating stable mortality risks when BAG was below 2 years, with significant increases beyond 3 years, reaching a HR greater than 2.0 between 6 to 8 years. Our findings position BAG as a valuable biomarker for biological aging with substantial prognostic implications. Our results align with emerging evidence linking accelerated biological aging to adverse health outcomes, including cardiovascular mortality and frailty progression. The 12% annual increase in mortality risk per year of elevated BAG mirrors previously reported associations between epigenetic age acceleration and cardiovascular mortality\u003csup\u003e\u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e81\u003c/span\u003e\u003c/sup\u003e (HR range: 1.58\u0026ndash;1.59 per year). Additionally, the pronounced threshold effect at BAG values exceeding 6 years likely reflects cumulative neurological damage surpassing compensatory mechanisms. This interpretation is consistent with neuroimaging studies\u003csup\u003e\u003cspan citationid=\"CR80\" class=\"CitationRef\"\u003e82\u003c/span\u003e,\u003cspan citationid=\"CR81\" class=\"CitationRef\"\u003e83\u003c/span\u003e\u003c/sup\u003e reporting accelerated cortical thinning and increased white matter hyperintensities in individuals with higher BAG values. Overall, our findings underscore the importance of monitoring elevated BAG levels, offering potential targets for early interventions aimed at mitigating associated mortality risks.\u003c/p\u003e \u003cp\u003eThis study further elucidates the significant association between healthy lifestyle factors and brain age. The results demonstrate that individuals who adhere to regular physical activity, a balanced diet, non-smoking, and moderate alcohol consumption tend to exhibit a \"younger\" brain age compared to those with less favorable lifestyle habits, suggesting that healthy behaviors may effectively decelerate the brain aging process. Specifically, within the Q4 of brain age, non-smokers showed a reduction in BAG of 0.11 years compared to smokers (p\u0026thinsp;=\u0026thinsp;0.002), while moderate alcohol consumption (BAG reduction of 0.20 years, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and regular exercise (BAG reduction of 0.14 years, p\u0026thinsp;=\u0026thinsp;0.03) also demonstrated significant attenuation of neurodegenerative effects. These findings are highly consistent with recent research, such as the UK Biobank cohort analysis, which revealed that although individuals with diabetes or prediabetes exhibited accelerated brain aging, this trend was markedly mitigated in subgroups maintaining optimal lifestyle practices, including non-smoking, high physical activity, and moderate alcohol consumption\u003csup\u003e\u003cspan citationid=\"CR82\" class=\"CitationRef\"\u003e84\u003c/span\u003e\u003c/sup\u003e. Notably, while previous large-scale studies in the general population have shown relatively weak correlations between individual lifestyle factors and brain age, this study underscores the synergistic protective effects of comprehensive healthy behaviors on brain age. This discovery holds significant public health implications: interventions promoting exercise, smoking cessation, alcohol moderation, and balanced nutrition may effectively counteract brain aging at the population level, thereby mitigating the negative impacts of genetic or disease-related factors on brain age. The present study provides new empirical support for this hypothesis and lays a theoretical foundation for future research on lifestyle-based interventions targeting brain aging. These findings not only deepen our understanding of the relationship between lifestyle and brain health but also offer scientific evidence for the development of preventive public health strategies.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThe 3D-ViT model represents a transformative advancement in brain age estimation, offering unprecedented accuracy and generalizability through its innovative integration of self-attention mechanisms and volumetric MRI analysis. By capturing disease-specific neuropathological signatures and hierarchical associations with cognitive decline, this framework provides critical insights into accelerated brain aging across neurodegenerative, psychiatric, and neurological disorders. Its independent association with all-cause mortality underscores its utility as a robust biomarker of biological aging, while the identification of stage-specific lifestyle interventions highlights the potential for tailored strategies to mitigate neurodegeneration. This study advances our understanding of brain aging and establishes a precision framework for risk stratification, intervention timing, and public health recommendations to optimize brain health across the lifespan.\u003c/p\u003e"},{"header":"Limitations","content":"\u003cp\u003eThis study has several methodological constraints that warrant consideration. Firstly, the generalizability of findings is constrained by sampling bias in data sources. The training and validation datasets were predominantly derived from large-scale cohorts (UK Biobank, ADNI, PPMI) characterized by demographic homogeneity\u0026mdash;particularly the overrepresentation of European ancestry individuals with above-average health literacy in UK Biobank, and disease-specific selection biases in ADNI (Alzheimer's-focused) and PPMI (Parkinson's-enriched). Secondly, technical heterogeneity in neuroimaging protocols introduces potential measurement confounding. Variations in MRI hardware specifications (e.g., 1.5T vs. 3T scanners) and acquisition parameters across datasets, despite rigorous intensity normalization and harmonization procedures, could systematically influence brain age estimations\u0026mdash;a persistent challenge in multicenter neuroimaging research that necessitates advanced cross-scanner calibration techniques. Thirdly, the cross-sectional design precludes causal inference regarding the temporal dynamics between brain aging acceleration and clinical outcomes. While significant associations were observed between elevated brain age and cognitive decline/disease risk, the directionality of these relationships remains ambiguous, as residual confounding from unmeasured genetic, epigenetic, or environmental factors cannot be excluded.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data generated in this study and a data dictionary (Supplementary Data) are provided in the Supplementary Data. The data that support training and validating the proposed brain age estimation model were obtained from the UK Biobank (https://www.ukbiobank.ac.uk/enable-your-research/register), the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset (https://adni.loni.usc.edu/), and the Parkinson's Progression Markers Initiative (PPMI) dataset (https://www.ppmi-info.org/), upon registration and compliance with the data use agreement. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe source codes pertaining to both the brain age estimation model and data analysis in this manuscript are provided at https://github.com/ZRX-MedAI/Brain_Age_Estimation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was supported by Key R\u0026amp;D Program of Zhejiang No.2024C04024. Natural Science Foundation of Xinjiang Autonomous Region No.2022D01C434. Natural Science Foundation of Xinjiang Autonomous Region No.2022D01C434.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eZH and MH jointly supervised research. ZH and ZJ designed this study. FY developed a deep learning model. SZ and FY performed the model interpretation. SZ, FY and ZH performed data analysis. SZ, FY and ZH interpreted the results. SZ, FY, ZJ, MH and ZH prepared the first draft of the manuscript. All authors contributed and approved the final draft. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe UK Biobank resource was used under application number 89757. The MEGASTROKE project received funding from sources specified at http://www.megastroke.org/acknowledgments.html. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary material \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSupplementary material is available at \u003cem\u003eSupplementary material\u003c/em\u003e.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eZhao B, Li T, Fan Z, \u003cem\u003eet al.\u003c/em\u003e Heart-brain connections: Phenotypic and genetic insights from magnetic resonance images. \u003cem\u003eScience\u003c/em\u003e 2023; \u003cstrong\u003e380\u003c/strong\u003e: abn6598.\u003c/li\u003e\n\u003cli\u003eTarkhov AE, Lindstrom-Vautrin T, Zhang S, \u003cem\u003eet al.\u003c/em\u003e Nature of epigenetic aging from a single-cell perspective. \u003cem\u003eNat Aging\u003c/em\u003e 2024; \u003cstrong\u003e4\u003c/strong\u003e: 854\u0026ndash;70.\u003c/li\u003e\n\u003cli\u003eSeitz-Holland J, Haas SS, Penzel N, Reichenberg A, Pasternak O. BrainAGE, brain health, and mental disorders: A systematic review. \u003cem\u003eNeurosci Biobehav Rev\u003c/em\u003e 2024; \u003cstrong\u003e159\u003c/strong\u003e: 105581.\u003c/li\u003e\n\u003cli\u003eSeitz-Holland J, Haas SS, Penzel N, Reichenberg A, Pasternak O. BrainAGE, brain health, and mental disorders: A systematic review. \u003cem\u003eNeurosci Biobehav Rev\u003c/em\u003e 2024; \u003cstrong\u003e159\u003c/strong\u003e: 105581.\u003c/li\u003e\n\u003cli\u003eMoguilner S, Baez S, Hernandez H, \u003cem\u003eet al.\u003c/em\u003e Brain clocks capture diversity and disparities in aging and dementia across geographically diverse populations. \u003cem\u003eNat Med\u003c/em\u003e 2024; published online Aug 26. DOI:10.1038/s41591-024-03209-x.\u003c/li\u003e\n\u003cli\u003eAzzam M, Xu Z, Liu R, \u003cem\u003eet al.\u003c/em\u003e A review of artificial intelligence-based brain age estimation and its applications for related diseases. \u003cem\u003eBrief Funct Genomics\u003c/em\u003e 2024; published online Oct 22. DOI:10.1093/bfgp/elae042.\u003c/li\u003e\n\u003cli\u003eBellantuono L, Marzano L, La Rocca M, \u003cem\u003eet al.\u003c/em\u003e Predicting brain age with complex networks: From adolescence to adulthood. \u003cem\u003eNeuroimage\u003c/em\u003e 2021; \u003cstrong\u003e225\u003c/strong\u003e: 117458.\u003c/li\u003e\n\u003cli\u003eHong J, Feng Z, Wang S-H, \u003cem\u003eet al.\u003c/em\u003e Brain age prediction of children using routine brain MR images via deep learning. \u003cem\u003eFront Neurol\u003c/em\u003e 2020; \u003cstrong\u003e11\u003c/strong\u003e: 584682.\u003c/li\u003e\n\u003cli\u003eHan LKM, Schnack HG, Brouwer RM, \u003cem\u003eet al.\u003c/em\u003e Contributing factors to advanced brain aging in depression and anxiety disorders. \u003cem\u003eTransl Psychiatry\u003c/em\u003e 2021; \u003cstrong\u003e11\u003c/strong\u003e: 402.\u003c/li\u003e\n\u003cli\u003eLuo Y, Chen W, Qiu J, Jia T. Accelerated functional brain aging in major depressive disorder: evidence from a large scale fMRI analysis of Chinese participants. \u003cem\u003eTransl Psychiatry\u003c/em\u003e 2022; \u003cstrong\u003e12\u003c/strong\u003e: 397.\u003c/li\u003e\n\u003cli\u003eChen Y-S, Kuo C-Y, Lu C-H, Wang Y-W, Chou K-H, Lin W-C. Multiscale brain age prediction reveals region-specific accelerated brain aging in Parkinson\u0026rsquo;s disease. \u003cem\u003eNeurobiol Aging\u003c/em\u003e 2024; \u003cstrong\u003e140\u003c/strong\u003e: 122\u0026ndash;9.\u003c/li\u003e\n\u003cli\u003eHwang I, Yeon EK, Lee JY, \u003cem\u003eet al.\u003c/em\u003e Prediction of brain age from routine T2-weighted spin-echo brain magnetic resonance images with a deep convolutional neural network. \u003cem\u003eNeurobiol Aging\u003c/em\u003e 2021; \u003cstrong\u003e105\u003c/strong\u003e: 78\u0026ndash;85.\u003c/li\u003e\n\u003cli\u003eKolbeinsson A, Filippi S, Panagakis Y, \u003cem\u003eet al.\u003c/em\u003e Accelerated MRI-predicted brain ageing and its associations with cardiometabolic and brain disorders. \u003cem\u003eSci Rep\u003c/em\u003e 2020; \u003cstrong\u003e10\u003c/strong\u003e: 19940.\u003c/li\u003e\n\u003cli\u003eVaratharajah Y, Baradwaj S, Kiraly A, \u003cem\u003eet al.\u003c/em\u003e Predicting brain age using structural neuroimaging and deep learning. bioRxiv. 2018; published online Dec 17. DOI:10.1101/497925.\u003c/li\u003e\n\u003cli\u003eWang J, Knol MJ, Tiulpin A, \u003cem\u003eet al.\u003c/em\u003e Gray matter age prediction as a biomarker for risk of dementia. \u003cem\u003eProc Natl Acad Sci U S A\u003c/em\u003e 2019; \u003cstrong\u003e116\u003c/strong\u003e: 21213\u0026ndash;8.\u003c/li\u003e\n\u003cli\u003eLahoud J, Cao J, Khan FS, \u003cem\u003eet al.\u003c/em\u003e 3D vision with transformers: A survey. arXiv [cs.CV]. 2022; published online Aug 8. http://arxiv.org/abs/2208.04309.\u003c/li\u003e\n\u003cli\u003eScabini L, Sacilotti A, Zielinski KM, Ribas LC, De Baets B, Bruno OM. A comparative survey of Vision Transformers for feature extraction in texture analysis. arXiv [cs.CV]. 2024; published online June 10. http://arxiv.org/abs/2406.06136.\u003c/li\u003e\n\u003cli\u003eLaguna A, Pe\u0026ntilde;uelas N, Vila M. Modelling human neuronal catecholaminergic pigmentation in rodents recapitulates age-related multisystem neurodegenerative deficits v2. 2024; published online June 4. DOI:10.17504/protocols.io.4r3l2qq1ql1y/v2.\u003c/li\u003e\n\u003cli\u003eTian YE, Cropley V, Maier AB, Lautenschlager NT, Breakspear M, Zalesky A. Heterogeneous aging across multiple organ systems and prediction of chronic disease and mortality. \u003cem\u003eNat Med\u003c/em\u003e 2023; \u003cstrong\u003e29\u003c/strong\u003e: 1221\u0026ndash;31.\u003c/li\u003e\n\u003cli\u003eBallester PL, Suh JS, Ho NCW, \u003cem\u003eet al.\u003c/em\u003e Gray matter volume drives the brain age gap in schizophrenia: a SHAP study. \u003cem\u003eSchizophrenia (Heidelb)\u003c/em\u003e 2023; \u003cstrong\u003e9\u003c/strong\u003e: 3.\u003c/li\u003e\n\u003cli\u003eSiddiqi SH, Schaper FLWVJ, Horn A, \u003cem\u003eet al.\u003c/em\u003e Brain stimulation and brain lesions converge on common causal circuits in neuropsychiatric disease. \u003cem\u003eNat Hum Behav\u003c/em\u003e 2021; \u003cstrong\u003e5\u003c/strong\u003e: 1707\u0026ndash;16.\u003c/li\u003e\n\u003cli\u003eTeselink J, Bawa KK, Koo GK, \u003cem\u003eet al.\u003c/em\u003e Efficacy of non-invasive brain stimulation on global cognition and neuropsychiatric symptoms in Alzheimer\u0026rsquo;s disease and mild cognitive impairment: A meta-analysis and systematic review. \u003cem\u003eAgeing Res Rev\u003c/em\u003e 2021; \u003cstrong\u003e72\u003c/strong\u003e: 101499.\u003c/li\u003e\n\u003cli\u003eOperto FF, Child Neuropsychiatry Unit, Department of Medicine, Surgery and Dentistry, University of Salerno, Salerno, Italy, Pastorino GMG, \u003cem\u003eet al.\u003c/em\u003e Social Cognition in neuropsychiatric disorders in pediatric age. \u003cem\u003eBrain (Bacau)\u003c/em\u003e 2020; \u003cstrong\u003e11\u003c/strong\u003e: 81\u0026ndash;8.\u003c/li\u003e\n\u003cli\u003eHonea RA, Wilkins H, Hunt SL, \u003cem\u003eet al.\u003c/em\u003e TOMM40 may mediate GFAP, neurofilament light Protein, pTau181, and brain morphometry in aging. \u003cem\u003eAging Brain\u003c/em\u003e 2025; \u003cstrong\u003e7\u003c/strong\u003e: 100134.\u003c/li\u003e\n\u003cli\u003eHernandez AR, Parker E, Babar M, \u003cem\u003eet al.\u003c/em\u003e Microbiome-driven alterations in metabolic pathways and impaired cognition in aged female TgF344-AD rats. \u003cem\u003eAging Brain\u003c/em\u003e 2024; \u003cstrong\u003e5\u003c/strong\u003e: 100119.\u003c/li\u003e\n\u003cli\u003eLuo S, Wang P-Y, Liu W-H, \u003cem\u003eet al.\u003c/em\u003e Age at onset of genetic disease and genetic dependent stage: evidence from cases with SCN1A variants. Brain. 2025; published online Feb 19. DOI:10.1093/brain/awaf064.\u003c/li\u003e\n\u003cli\u003eSmith S, Hopp SC. The 5XFAD mouse model of Alzheimer\u0026rsquo;s disease displays age-dependent deficits in habituation to a novel environment. \u003cem\u003eAging Brain\u003c/em\u003e 2023; \u003cstrong\u003e3\u003c/strong\u003e: 100078.\u003c/li\u003e\n\u003cli\u003eCorrection to: GDF5 as a rejuvenating treatment for age-related neuromuscular failure. \u003cem\u003eBrain\u003c/em\u003e 2025; \u003cstrong\u003e148\u003c/strong\u003e: e21.\u003c/li\u003e\n\u003cli\u003ede F\u0026aacute;tima Dias M, Duarte JV, de Carvalho P, Castelo-Branco M. Unravelling pathological ageing with brain age gap estimation in Alzheimer\u0026rsquo;s disease, diabetes, and schizophrenia. \u003cem\u003eBrain Commun\u003c/em\u003e 2025; published online March 11. DOI:10.1093/braincomms/fcaf109.\u003c/li\u003e\n\u003cli\u003eWang X, Zhu Z, Xu X, \u003cem\u003eet al.\u003c/em\u003e Construction of brain age models based on structural and white matter information. \u003cem\u003eBrain Res\u003c/em\u003e 2025; \u003cstrong\u003e1851\u003c/strong\u003e: 149458.\u003c/li\u003e\n\u003cli\u003eGibbon S, Breen DP, MacGillivray TJ, UK Biobank Eye \u0026amp; Vision Consortium. Optic disc pallor in Parkinson\u0026rsquo;s disease: A UK Biobank study. \u003cem\u003eMov Disord\u003c/em\u003e 2025; published online Jan 30. DOI:10.1002/mds.30127.\u003c/li\u003e\n\u003cli\u003eHanazawa R, Sato H, Hirakawa A, Alzheimer\u0026rsquo;s Disease Neuroimaging Initiative. Mixture disease progression model to predict and cluster the long-term trajectory of cognitive decline in Alzheimer\u0026rsquo;s disease. \u003cem\u003eTher Innov Regul Sci\u003c/em\u003e 2025; \u003cstrong\u003e59\u003c/strong\u003e: 264\u0026ndash;77.\u003c/li\u003e\n\u003cli\u003eNudelman KNH, Brumm MC, Marek K, Foroud TM, for the Parkinson\u0026rsquo;s Progression Markers Initiative (PPMI) Study. \u003cem\u003eTREM2\u003c/em\u003e variants in Parkinson\u0026rsquo;s disease: Results from the Parkinson\u0026rsquo;s progression markers initiative (PPMI) study. \u003cem\u003eAlzheimers Dement\u003c/em\u003e 2022; \u003cstrong\u003e18\u003c/strong\u003e. DOI:10.1002/alz.062316.\u003c/li\u003e\n\u003cli\u003eGiff A, Noren G, Magnotti J, \u003cem\u003eet al.\u003c/em\u003e Spatial normalization discrepancies between native and MNI152 brain template scans in gamma ventral capsulotomy patients. \u003cem\u003ePsychiatry Res Neuroimaging\u003c/em\u003e 2023; \u003cstrong\u003e329\u003c/strong\u003e: 111595.\u003c/li\u003e\n\u003cli\u003eGiff A, Noren G, Magnotti J, \u003cem\u003eet al.\u003c/em\u003e 19. Spatial normalization discrepancies between native and MNI152 brain template scans in gamma ventral capsulotomy patients. \u003cem\u003eBiol Psychiatry\u003c/em\u003e 2023; \u003cstrong\u003e93\u003c/strong\u003e: S101\u0026ndash;2.\u003c/li\u003e\n\u003cli\u003eChung BS, Park JS. Automatic segmentation of true color sectioned images using FMRIB Software Library: First trial in brain, gray matter, and white matter. \u003cem\u003eClin Anat\u003c/em\u003e 2020; \u003cstrong\u003e33\u003c/strong\u003e: 1197\u0026ndash;203.\u003c/li\u003e\n\u003cli\u003eZuo Q, Song L, Gao X, \u003cem\u003eet al.\u003c/em\u003e Associations of metabolic syndrome with cognitive function and dementia risk: Evidence from the UK Biobank cohort. \u003cem\u003eDiabetes Obes Metab\u003c/em\u003e 2024; \u003cstrong\u003e26\u003c/strong\u003e: 6023\u0026ndash;33.\u003c/li\u003e\n\u003cli\u003eSchulz C-A, Weinhold L, Schmid M, N\u0026ouml;then MM, N\u0026ouml;thlings U. Analysis of associations between dietary patterns, genetic disposition, and cognitive function in data from UK Biobank. \u003cem\u003eEur J Nutr\u003c/em\u003e 2023; \u003cstrong\u003e62\u003c/strong\u003e: 511\u0026ndash;21.\u003c/li\u003e\n\u003cli\u003eSusetyo B, Fitrianto A. Estimating missing panel data with regression and multivariate imputation by chained equations (MICE). \u003cem\u003eCAUCHY\u003c/em\u003e 2024; \u003cstrong\u003e9\u003c/strong\u003e: 94\u0026ndash;105.\u003c/li\u003e\n\u003cli\u003eAustin PC. Graphical methods to illustrate the nature of the relation between a continuous variable and the outcome when using restricted cubic splines with a Cox proportional hazards model. \u003cem\u003eStat Methods Med Res\u003c/em\u003e 2025; \u003cstrong\u003e34\u003c/strong\u003e: 277\u0026ndash;85.\u003c/li\u003e\n\u003cli\u003eErratum to \u0026ldquo;A dose-effect network meta-analysis model with application in antidepressants using restricted cubic splines.\u0026rdquo; \u003cem\u003eStat Methods Med Res\u003c/em\u003e 2024; \u003cstrong\u003e33\u003c/strong\u003e: NP1.\u003c/li\u003e\n\u003cli\u003eFuh C-D, Kao C-LM, Pang T. Kullback-Leibler divergence and Akaike information criterion in general hidden Markov models. \u003cem\u003eIEEE Trans Inf Theory\u003c/em\u003e 2024; \u003cstrong\u003e70\u003c/strong\u003e: 5888\u0026ndash;909.\u003c/li\u003e\n\u003cli\u003eSaumard A, Navarro F. Finite sample improvement of Akaike\u0026rsquo;s information criterion. \u003cem\u003eIEEE Trans Inf Theory\u003c/em\u003e 2021; \u003cstrong\u003e67\u003c/strong\u003e: 6328\u0026ndash;43.\u003c/li\u003e\n\u003cli\u003eXia L, Nan B, Li Y. Statistical inference for Cox proportional hazards models with a diverging number of covariates. \u003cem\u003eScand Stat Theory Appl\u003c/em\u003e 2023; \u003cstrong\u003e50\u003c/strong\u003e: 550\u0026ndash;71.\u003c/li\u003e\n\u003cli\u003eHahn G, Prokopenko D, Hecker J, \u003cem\u003eet al.\u003c/em\u003e Polygenic hazard score models for the prediction of Alzheimer\u0026rsquo;s free survival using the lasso for Cox\u0026rsquo;s proportional hazards model. \u003cem\u003eGenet Epidemiol\u003c/em\u003e 2025; \u003cstrong\u003e49\u003c/strong\u003e: e22581.\u003c/li\u003e\n\u003cli\u003eHepsomali P, Groeger JA. Diet, sleep, and mental health: Insights from the UK Biobank study. \u003cem\u003eNutrients\u003c/em\u003e 2021; \u003cstrong\u003e13\u003c/strong\u003e: 2573.\u003c/li\u003e\n\u003cli\u003eHuang S-Y, Li Y-Z, Zhang Y-R, \u003cem\u003eet al.\u003c/em\u003e Sleep, physical activity, sedentary behavior, and risk of incident dementia: a prospective cohort study of 431,924 UK Biobank participants. \u003cem\u003eMol Psychiatry\u003c/em\u003e 2022; \u003cstrong\u003e27\u003c/strong\u003e: 4343\u0026ndash;54.\u003c/li\u003e\n\u003cli\u003eChudasama YV, Khunti K, Gillies CL, \u003cem\u003eet al.\u003c/em\u003e Healthy lifestyle and life expectancy in people with multimorbidity in the UK Biobank: A longitudinal cohort study. \u003cem\u003ePLoS Med\u003c/em\u003e 2020; \u003cstrong\u003e17\u003c/strong\u003e: e1003332.\u003c/li\u003e\n\u003cli\u003eZhang Y-B, Chen C, Pan X-F, \u003cem\u003eet al.\u003c/em\u003e Associations of healthy lifestyle and socioeconomic status with mortality and incident cardiovascular disease: two prospective cohort studies. \u003cem\u003eBMJ\u003c/em\u003e 2021; \u003cstrong\u003e373\u003c/strong\u003e: n604.\u003c/li\u003e\n\u003cli\u003eRodriguez D, Sued M, Valdora M. A Kruskal-Wallis type test for functional data. \u003cem\u003eCommun Stat Simul Comput\u003c/em\u003e 2025; : 1\u0026ndash;15.\u003c/li\u003e\n\u003cli\u003eHe S, Grant PE, Ou Y. Global-local transformer for brain age estimation. \u003cem\u003eIEEE Trans Med Imaging\u003c/em\u003e 2022; \u003cstrong\u003e41\u003c/strong\u003e: 213\u0026ndash;24.\u003c/li\u003e\n\u003cli\u003eQiu S, Joshi PS, Miller MI, \u003cem\u003eet al.\u003c/em\u003e Development and validation of an interpretable deep learning framework for Alzheimer\u0026rsquo;s disease classification. \u003cem\u003eBrain\u003c/em\u003e 2020; \u003cstrong\u003e143\u003c/strong\u003e: 1920\u0026ndash;33.\u003c/li\u003e\n\u003cli\u003eBrendel W, Bethge M. Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNet. arXiv [cs.CV]. 2019; published online March 20. http://arxiv.org/abs/1904.00760.\u003c/li\u003e\n\u003cli\u003eZhang X, Yang H, Young EFY. Attentional transfer is all you need: Technology-aware layout pattern generation. In: 2021 58th ACM/IEEE Design Automation Conference (DAC). IEEE, 2021. DOI:10.1109/dac18074.2021.9586227.\u003c/li\u003e\n\u003cli\u003eHe S, Grant PE, Ou Y. Global-local transformer for brain age estimation. \u003cem\u003eIEEE Trans Med Imaging\u003c/em\u003e 2022; \u003cstrong\u003e41\u003c/strong\u003e: 213\u0026ndash;24.\u003c/li\u003e\n\u003cli\u003eBorkar K, Chaturvedi A, Vinod PK, Bapi RS. Ayu-Characterization of healthy aging from neuroimaging data with deep learning and rsfMRI. \u003cem\u003eFront Comput Neurosci\u003c/em\u003e 2022; \u003cstrong\u003e16\u003c/strong\u003e: 940922.\u003c/li\u003e\n\u003cli\u003eCheng Y, Zhang X-D, Chen C, \u003cem\u003eet al.\u003c/em\u003e Dynamic evolution of brain structural patterns in liver transplantation recipients: a longitudinal study based on 3D convolutional neuronal network model. \u003cem\u003eEur Radiol\u003c/em\u003e 2023; \u003cstrong\u003e33\u003c/strong\u003e: 6134\u0026ndash;44.\u003c/li\u003e\n\u003cli\u003eKuchcinski G, Rumetshofer T, Zervides KA, \u003cem\u003eet al.\u003c/em\u003e MRI BrainAGE demonstrates increased brain aging in systemic lupus erythematosus patients. \u003cem\u003eFront Aging Neurosci\u003c/em\u003e 2023; \u003cstrong\u003e15\u003c/strong\u003e: 1274061.\u003c/li\u003e\n\u003cli\u003ePoloni KM, Ferrari RJ. A deep ensemble hippocampal CNN model for brain age estimation applied to Alzheimer\u0026rsquo;s diagnosis. \u003cem\u003eExpert Syst Appl\u003c/em\u003e 2022; \u003cstrong\u003e195\u003c/strong\u003e: 116622.\u003c/li\u003e\n\u003cli\u003eCadena EJ, White DM, Kraguljac NV, Reid MA, Lahti AC. Evaluation of fronto-striatal networks during cognitive control in unmedicated patients with schizophrenia and the effect of antipsychotic medication. \u003cem\u003eNPJ Schizophr\u003c/em\u003e 2018; \u003cstrong\u003e4\u003c/strong\u003e: 8.\u003c/li\u003e\n\u003cli\u003eAverbeck B, O\u0026rsquo;Doherty JP. Reinforcement-learning in fronto-striatal circuits. \u003cem\u003eNeuropsychopharmacology\u003c/em\u003e 2022; \u003cstrong\u003e47\u003c/strong\u003e: 147\u0026ndash;62.\u003c/li\u003e\n\u003cli\u003eTang Y, Yan Y, Mao J, Ni J, Qing H. The hippocampus associated GABAergic neural network impairment in early-stage of Alzheimer\u0026rsquo;s disease. \u003cem\u003eAgeing Res Rev\u003c/em\u003e 2023; \u003cstrong\u003e86\u003c/strong\u003e: 101865.\u003c/li\u003e\n\u003cli\u003ePal G, O\u0026rsquo;Keefe J, Robertson-Dick E, Bernard B, Anderson S, Hall D. Global cognitive function and processing speed are associated with gait and balance dysfunction in Parkinson\u0026rsquo;s disease. \u003cem\u003eJ Neuroeng Rehabil\u003c/em\u003e 2016; \u003cstrong\u003e13\u003c/strong\u003e: 94.\u003c/li\u003e\n\u003cli\u003eEbaid D, Crewther SG, MacCalman K, Brown A, Crewther DP. Cognitive processing speed across the lifespan: Beyond the influence of motor speed. \u003cem\u003eFront Aging Neurosci\u003c/em\u003e 2017; \u003cstrong\u003e9\u003c/strong\u003e: 62.\u003c/li\u003e\n\u003cli\u003eAlbrecht F, Johansson H, Ekman U, \u003cem\u003eet al.\u003c/em\u003e Investigating underlying brain structures and influence of mild and subjective cognitive impairment on dual-task performance in people with Parkinson\u0026rsquo;s disease. \u003cem\u003eSci Rep\u003c/em\u003e 2024; \u003cstrong\u003e14\u003c/strong\u003e: 9513.\u003c/li\u003e\n\u003cli\u003eAbd-Alrazaq A, Ahmed A, Alali H, Aldardour AM, Househ M. The effectiveness of serious games on cognitive processing speed among older adults with cognitive impairment: Systematic review and meta-analysis. \u003cem\u003eJMIR Serious Games\u003c/em\u003e 2022; \u003cstrong\u003e10\u003c/strong\u003e: e36754.\u003c/li\u003e\n\u003cli\u003eJung YH, Chae CW, Han HJ. The potential role of gut microbiota-derived metabolites as regulators of metabolic syndrome-associated mitochondrial and endolysosomal dysfunction in Alzheimer\u0026rsquo;s disease. \u003cem\u003eExp Mol Med\u003c/em\u003e 2024; \u003cstrong\u003e56\u003c/strong\u003e: 1691\u0026ndash;702.\u003c/li\u003e\n\u003cli\u003eBogdanova I, Sokolik V, Voloshyna N, \u003cem\u003eet al.\u003c/em\u003e The effectiveness of rehabilitation programs for the mobilization of compensatory-adaptive neuroplasticity processes in patients with Parkinson\u0026rsquo;s disease according to indicators of neurotrophic factors. \u003cem\u003eUkrains\u0026rsquo;kyi Visnyk Psykhonevrolohii\u003c/em\u003e 2022; : 18\u0026ndash;23.\u003c/li\u003e\n\u003cli\u003ePassaretti M, Cilia R, Rinaldo S, \u003cem\u003eet al.\u003c/em\u003e Neurophysiological markers of motor compensatory mechanisms in early Parkinson\u0026rsquo;s disease. \u003cem\u003eBrain\u003c/em\u003e 2024; \u003cstrong\u003e147\u003c/strong\u003e: 3714\u0026ndash;26.\u003c/li\u003e\n\u003cli\u003eRademacher K, Nakamura K. Role of dopamine neuron activity in Parkinson\u0026rsquo;s disease pathophysiology. \u003cem\u003eExp Neurol\u003c/em\u003e 2024; \u003cstrong\u003e373\u003c/strong\u003e: 114645.\u003c/li\u003e\n\u003cli\u003eYegorov YE, Poznyak AV, Nikiforov NG, Sobenin IA, Orekhov AN. The link between chronic stress and accelerated aging. \u003cem\u003eBiomedicines\u003c/em\u003e 2020; \u003cstrong\u003e8\u003c/strong\u003e: 198.\u003c/li\u003e\n\u003cli\u003eBobba-Alves N, Sturm G, Lin J, \u003cem\u003eet al.\u003c/em\u003e Cellular allostatic load is linked to increased energy expenditure and accelerated biological aging. \u003cem\u003ePsychoneuroendocrinology\u003c/em\u003e 2023; \u003cstrong\u003e155\u003c/strong\u003e: 106322.\u003c/li\u003e\n\u003cli\u003eJamea AA, Alblowi M, Alghamdi J, \u003cem\u003eet al.\u003c/em\u003e Altered default mode network activity and cortical thickness as vulnerability indicators for SCZ: a preliminary resting state MRI study. \u003cem\u003eEur Rev Med Pharmacol Sci\u003c/em\u003e 2021; \u003cstrong\u003e25\u003c/strong\u003e: 669\u0026ndash;77.\u003c/li\u003e\n\u003cli\u003eWang H, Zhao Q, Zhang Y, \u003cem\u003eet al.\u003c/em\u003e Shared genetic architecture of cortical thickness alterations in major depressive disorder and schizophrenia. \u003cem\u003eProg Neuropsychopharmacol Biol Psychiatry\u003c/em\u003e 2024; \u003cstrong\u003e135\u003c/strong\u003e: 111121.\u003c/li\u003e\n\u003cli\u003eKlotz L, Antel J, Kuhlmann T. Inflammation in multiple sclerosis: consequences for remyelination and disease progression. \u003cem\u003eNat Rev Neurol\u003c/em\u003e 2023; \u003cstrong\u003e19\u003c/strong\u003e: 305\u0026ndash;20.\u003c/li\u003e\n\u003cli\u003ePsenicka MW, Smith BC, Tinkey RA, Williams JL. Connecting neuroinflammation and neurodegeneration in multiple sclerosis: Are oligodendrocyte precursor cells a nexus of disease? \u003cem\u003eFront Cell Neurosci\u003c/em\u003e 2021; \u003cstrong\u003e15\u003c/strong\u003e: 654284.\u003c/li\u003e\n\u003cli\u003eNian K, Harding IC, Herman IM, Ebong EE. Blood-brain barrier damage in ischemic stroke and its regulation by endothelial mechanotransduction. \u003cem\u003eFront Physiol\u003c/em\u003e 2020; \u003cstrong\u003e11\u003c/strong\u003e: 605398.\u003c/li\u003e\n\u003cli\u003eLi X, Simo L, Zhao Q, Kim EG, Ding Y, Geng X. Endothelial cells and the blood-brain barrier: Critical determinants of ineffective reperfusion in stroke. \u003cem\u003eEur J Neurosci\u003c/em\u003e 2025; \u003cstrong\u003e61\u003c/strong\u003e: e16663.\u003c/li\u003e\n\u003cli\u003eGong Y, Song Y, Xu J, \u003cem\u003eet al.\u003c/em\u003e Progression of frailty and cardiovascular outcomes among Medicare beneficiaries. \u003cem\u003emedRxiv\u003c/em\u003e 2024; published online Feb 13. DOI:10.1101/2024.02.09.24302612.\u003c/li\u003e\n\u003cli\u003eBernal J, Menze I, Yakupov R, \u003cem\u003eet al.\u003c/em\u003e Longitudinal evidence for a mutually reinforcing relationship between white matter hyperintensities and cortical thickness in cognitively unimpaired older adults. medRxiv. 2024; published online July 10. DOI:10.1101/2024.07.08.24309994.\u003c/li\u003e\n\u003cli\u003eJim\u0026eacute;nez-Balado J, Habeck C, Stern Y, Eich T. The relationship between cortical thickness and white matter hyperintensities in mid to late life. \u003cem\u003eNeurobiol Aging\u003c/em\u003e 2024; \u003cstrong\u003e141\u003c/strong\u003e: 129\u0026ndash;39.\u003c/li\u003e\n\u003cli\u003eDove A, Wang J, Huang H, \u003cem\u003eet al.\u003c/em\u003e Diabetes, prediabetes, and brain aging: The role of healthy lifestyle. \u003cem\u003eDiabetes Care\u003c/em\u003e 2024; \u003cstrong\u003e47\u003c/strong\u003e: 1794\u0026ndash;802.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Brain Age Gap (BAG), Neuropsychiatric Disorders, 3D Vision Transformer, Cognitive Decline, Lifestyle Interventions","lastPublishedDoi":"10.21203/rs.3.rs-6283338/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6283338/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe brain age gap (BAG), a neuroimaging-derived biomarker of accelerated brain aging, faces translational challenges due to model inaccuracies and unclear disease-mechanism linkages. We systematically evaluated BAG's clinical relevance across neuropsychiatric disorders, cognitive trajectories, mortality, and lifestyle interventions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eUsing multi-cohort data (UK Biobank [n = 38,967], Alzheimer’s Disease Neuroimaging Initiative [ADNI; n = 1,402], Parkinson’s Progression Markers Initiative [PPMI; n = 1,182]), we developed a 3D Vision Transformer (3D-ViT) model for whole-brain age estimation. Survival analyses, restricted cubic splines, and stratified regressions assessed BAG’s associations with cognition, 16 neuropsychiatric disorders, and mortality. Lifestyle modulation effects were quantified through longitudinal BAG progression.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe 3D Vision Transformer demonstrated robust predictive accuracy, achieving a mean absolute error (MAE) of 2.68 years in the UK Biobank cohort and 2.99–3.20 years in external validation cohorts (ADNI/PPMI). Per 1-year increment in BAG was linearly associated with elevated risks of Alzheimer's disease (HR = 1.165, 95% CI = 1.086–1.249; +16.5% risk/year), mild cognitive impairment (HR = 1.040, 95% CI = 1.030–1.050; +4.0%), and all-cause mortality (HR = 1.12, 1.09–1.15; +12%; all \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001). Individuals in the highest BAG quartile (Q4) faced substantially amplified risks: 2.8-fold for Alzheimer's disease (HR = 2.801), 6.4-fold for multiple sclerosis (HR = 6.417), and 1.5-fold for major depressive disorder (HR = 1.466). Notably, prodromal Parkinson's disease exhibited paradoxical BAG rejuvenation (mean Δ=−1.441 years, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001), contrasting with nonsignificant associations in incident Parkinson's cases (HR = 1.830, \u003cem\u003ep\u003c/em\u003e = 0.154). Cognitive decline followed nonlinear trajectories, with critical thresholds for domain-specific cognitive decline emerging at Q4 (BAG \u0026gt; 2.48 years). Lifestyle interventions synergistically attenuated BAG progression in advanced neurodegeneration (Q3–Q4; \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05), particularly through smoking cessation, moderated alcohol consumption, and physical activity.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInterpretation\u003c/strong\u003e: BAG robustly predicts accelerated brain aging, neuropsychiatric multimorbidity, and mortality. Its nonlinear cognitive thresholds and stage-dependent lifestyle modifiability underscore clinical utility for risk stratification and personalized prevention strategies.\u003c/p\u003e","manuscriptTitle":"Brain Age Gap as a Predictive Biomarker: Linking Aging, Lifestyle, and Neuropsychiatric Health","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-16 08:45:47","doi":"10.21203/rs.3.rs-6283338/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"communications-medicine","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"commsmed","sideBox":"Learn more about [Communications Medicine](http://www.nature.com/commsmed)","snPcode":"43856","submissionUrl":"https://mts-commsmed.nature.com/cgi-bin/main.plex","title":"Communications Medicine","twitterHandle":"@commsmedicine","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Communications Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"db0fb14e-c6cf-4691-b2c1-8922095d1131","owner":[],"postedDate":"April 16th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":46266591,"name":"Biological sciences/Neuroscience"},{"id":46266592,"name":"Biological sciences/Neuroscience/Cognitive neuroscience"}],"tags":[],"updatedAt":"2025-10-27T12:44:10+00:00","versionOfRecord":{"articleIdentity":"rs-6283338","link":"https://doi.org/10.1038/s43856-025-01100-5","journal":{"identity":"communications-medicine","isVorOnly":false,"title":"Communications Medicine"},"publishedOn":"2025-10-24 04:00:00","publishedOnDateReadable":"October 24th, 2025"},"versionCreatedAt":"2025-04-16 08:45:47","video":"","vorDoi":"10.1038/s43856-025-01100-5","vorDoiUrl":"https://doi.org/10.1038/s43856-025-01100-5","workflowStages":[]},"version":"v1","identity":"rs-6283338","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6283338","identity":"rs-6283338","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}