The concept of ‘brain age’, derived from neuroimaging data, serves as a crucial biomarker reflecting cognitive vitality and neurodegenerative trajectories. In the past decade, machine learning (ML) and deep learning (DL) integration has transformed the field, providing advanced models for brain age estimation. However, achieving precise brain age prediction across all ages remains a significant analytical challenge. This comprehensive review scrutinizes advancements in ML- and DL-based brain age prediction, analyzing 52 peer-reviewed studies from 2020 to 2024. It assesses various model architectures, highlighting their effectiveness and nuances in lifespan brain age studies. By comparing ML and DL, strengths in forecasting and methodological limitations are revealed. Finally, key findings from the reviewed articles are summarized and a number of major issues related to ML/DL-based lifespan brain age prediction are discussed. Through this study, we aim at the synthesis of the current state of brain age prediction, emphasizing both advancements and persistent challenges, guiding future research, technological advancements, and improving early intervention strategies for neurodegenerative diseases.

This study proposes a novel brain-region-level aging assessment paradigm based on Shapley value interpretation, aiming to overcome the interpretability limitations of traditional brain age prediction models. Although deep-learning-based brain age prediction models using neuroimaging data have become crucial tools for evaluating abnormal brain aging, their unidimensional brain age–chronological age discrepancy metric fails to characterize the regional heterogeneity of brain aging. Meanwhile, despite Shapley additive explanations having demonstrated potential for revealing regional heterogeneity, their application in complex deep learning algorithms has been hindered by prohibitive computational complexity. To address this, we innovatively developed a computational framework featuring efficient Shapley value approximation through a novel multi-stage computational strategy that significantly reduces complexity, thereby enabling an interpretable analysis of deep learning models. By establishing a reference system based on standard Shapley values from healthy populations, we constructed an anatomically specific Regional Brain Aging Deviation Index (RBADI) that maintains age-related validity. Experimental validation using UK Biobank data demonstrated that our framework successfully identified the thalamus (THA) and hippocampus (HIP) as core contributors to brain age prediction model decisions, highlighting their close associations with physiological aging. Notably, it revealed significant correlations between the insula (INS) and alcohol consumption, as well as between the inferior frontal gyrus opercular part (IFGoperc) and smoking history. Crucially, the RBADI exhibited superior performance in the tri-class classification of prodromal neurodegenerative diseases (HCs vs. MCI vs. AD: AUC = 0.92; HCs vs. pPD vs. PD: AUC = 0.86). This framework not only enables the practical implementation of Shapley additive explanations in brain age prediction deep learning models but also establishes anatomically interpretable biomarkers. These advancements provide a novel spatial analytical dimension for investigating brain aging mechanisms and demonstrate significant clinical translational value for early neurodegenerative disease screening, ultimately offering a new methodological tool for deciphering the neural mechanisms of aging.

Objectives: Given the escalating global prevalence of age-related cognitive impairments, identifying modifiable factors is crucial for developing targeted interventions. Methods: After excluding participants with dementia and substantial missing data, 453,950 individuals from UK Biobank (UKB) were included. Cognitive decline was assessed across four cognitive domains. The top 10% exhibiting the greatest decline were categorized as the “Cognitively At-Risk Population”. Eighty-three potential factors from three categories were analyzed. Univariate and multivariate Cox proportional hazards models were employed to assess the independent and joint effects of these factors on cognitive decline. Population Attributable Fractions (PAFs) were calculated to estimate the potential impact of eliminating each risk category. Results: Our findings revealed a significant impact of unfavorable medical and psychiatric histories on processing speed and visual episodic memory decline (Hazard Ratio (HR) = 1.34, 95% CI: 1.20–1.51, p = 6.06 × 10^⁻7; HR = 1.50, 95% CI: 1.22–1.86, p = 1.62 × 10^⁻4, respectively). Furthermore, PAF analysis indicated that physiological and biochemical markers were the most critical risk category for preventing processing speed decline (PAF = 7.03%), while social and behavioral factors exerted the greatest influence on preventing visual episodic memory decline (PAF = 9.68%). Higher education, socioeconomic status, and handgrip strength emerged as protective factors, whereas high body mass index (BMI), hypertension, and depression were detrimental. Conclusions: By identifying this high-risk group and quantifying the impact of modifiable factors, this study provides valuable insights for developing targeted interventions to delay cognitive decline and improve public health outcomes in middle-aged and older adults.

This study investigates the relationship between modifiable risk factors and dementia subtypes using data from 460,799 participants in the UK Biobank. Utilizing univariate Cox proportional hazards regression models, we examined the associations between 83 modifiable risk factors and the risks of all-cause dementia (ACD), Alzheimer’s disease (AD), and vascular dementia (VD). Composite scores for different domains were generated by aggregating risk factors associated with ACD, AD, and VD, respectively, and their joint associations were assessed in multivariable Cox models. Additionally, population attributable fractions (PAF) were utilized to estimate the potential impact of eliminating adverse characteristics of the risk domains. Our findings revealed that an unfavorable medical history significantly increased the risk of ACD, AD, and VD (hazard ratios (HR) = 1.88, 95% confidence interval (95% CI): 1.74–2.03, p < 0.001; HR = 1.80, 95% CI: 1.54–2.10, p < 0.001; HR = 2.39, 95% CI: 2.10–2.71, p < 0.001, respectively). Blood markers (PAF = 12.1%; 17.4%) emerged as the most important risk domain for preventing ACD and VD, while psychiatric factors (PAF = 18.3%) were the most important for preventing AD. This study underscores the potential for preventing dementia and its subtypes through targeted interventions for modifiable risk factors. The distinct insights provided by HR and PAF emphasize the importance of considering both the strength of the associations and the population-level impact of dementia prevention strategies. Our research provides valuable guidance for developing effective public health interventions aimed at reducing the burden of dementia, representing a significant advancement in the field.

Accurate estimation of brain age from structural magnetic resonance imaging (MRI) serves as a vital biomarker for quantifying individual neurobiological aging and identifying risks for neurological disorders. However, developing robust models that generalize across the entire lifespan (from infancy to aging) remains challenging due to heterogeneous maturation/degeneration patterns, limited cross-center generalizability, and insufficient temporal reliability evaluation. To address these limitations, we curated a large-scale, multi-center T1-weighted MRI dataset across 27 public cohorts. Of these, 22,271 scans from 17 cohorts (aged 0–96 years) formed the primary foundation for model development, complemented by 10 additional cohorts utilized for independent multi-center evaluation and robustness testing. We propose ResNet-CrossViT, a novel hybrid architecture that synergistically combines a 3D Contextual Transformer-ResNet (CoTResNet3D) backbone for enriched local feature extraction and a CrossVision Transformer (CrossViT) module for cross-scale global dependency modeling. The model was rigorously evaluated on an internal test set, an unseen external dataset for cross-center validation, a longitudinal dataset for assessing temporal consistency, and a test–retest dataset for measuring reproducibility. On the internal test set, ResNet-CrossViT achieved a mean absolute error (MAE) of 2.72 years and a maximal MAE (mMAE) of 5.10 years, demonstrating marked performance improvements, particularly within the challenging adolescent cohort. The model maintained strong generalization on the unseen dataset (MAE = 4.19 years) and exhibited superior longitudinal consistency (Mean Absolute Difference Error, MAdE = 3.68) and excellent test–retest reliability (Intraclass Correlation Coefficient, ICC = 0.994). By integrating a large-scale, heterogeneous lifespan dataset with a hybrid architecture that effectively captures both local structural details and global long-range interactions, our study provides a precise, generalizable, and reliable framework for brain age estimation.