Lan Lin

Background: White matter (WM) aging is commonly characterized as a largely unitary process of progressive microstructural decline, yet substantial interindividual variation suggests that this view may obscure distinct biological dimensions. Whether WM aging retains a multidimensional organization after minimizing variation attributable to overt clinical pathology remains unknown. Methods: We constructed a clinically defined aging reference cohort (CDAR; n = 17,310) from the UK Biobank (UKB) by excluding overt clinical pathology and analyzed 144 skeleton-based WM microstructural metrics. An additional 10,272 participants with overt clinical diagnoses were included for downstream association analyses, yielding a total analytic sample of 27,582. Surreal-GAN was applied to disentangle latent dimensions of WM aging, with reproducibility-based model selection and cross-sample and cross-sex validation. The biological relevance of the identified dimensions was evaluated through associations with cognitive, lifestyle, peripheral, and disease-related outcomes. Results: Three reproducible and coexisting dimensions of WM aging emerged, comprising distinct spatial configurations rather than a single diffuse aging axis. R₁ displayed an anterior-vulnerable/posterior-preserved gradient, R₂ reflected a widespread diffuse aging load, and R₃ showed a dissociation between the limbic–frontal system and sensorimotor pathways. The three patterns exhibited good cross-sample (r = 0.887–0.935) and cross-sex (r = 0.891–0.937) reproducibility, with strong bilateral hemispheric correspondence (r > 0.86). All three patterns were associated with age but exhibited partially shared and partially independent associations with cognition, lifestyle factors, peripheral phenotypes, and disease risk. Conclusions: WM aging is not adequately captured by a single neuroanatomical process. Even after minimizing overt clinical pathology, it retains a reproducible multidimensional organization, with partially shared and partially independent phenotypic profiles across aging dimensions. These findings suggest that individual differences in WM aging reflect partially separable biological processes rather than variation along a single aging axis.

Brain aging exhibits substantial interindividual heterogeneity, yet separating aging-related neuroanatomical variation from pathological influences remains methodologically challenging. To address this issue, we constructed a Clinically Defined Aging Reference (CDAR) cohort from the UK Biobank by excluding individuals with overt clinical pathology and applied the Surreal-GAN framework to characterize latent patterns of age-associated structural variations. A total of 26,251 participants were included. The model identified two co-occurring dimensions of brain aging, referred to as R₁ and R₂, that were stable across subsamples (R₁: r = 0.873, R₂: r = 0.953) and remained consistent when refitted separately in males and females (female: r = 0.792, male: r = 0.818). R₁ was characterized by widespread gray matter reduction involving cortical, subcortical, and cerebellar regions and was associated with broadly poorer cognitive performance, adverse lifestyle profiles, metabolic and inflammatory alterations, and age-related diseases. R₂ exhibited relative preservation of subcortical structures together with widespread preservation of cortical surface area and more selective differences in cortical thickness. Compared with R₁, R₂ showed weaker associations with cognition and peripheral physiological measures but retained associations with cardiovascular-related outcomes. These findings suggest that brain aging within a clinically defined aging reference cohort may involve multiple partially dissociable neuroanatomical dimensions rather than a single pattern, providing an operational reference for studying aging-related structural heterogeneity under reduced clinical confounding.

Internet of Things (IoT), sensor-network, and cyber-physical system (CPS) software increasingly relies on large language models (LLMs) and autonomous agents for code generation, maintenance, and vulnerability repair. However, LLM-generated edge services, telemetry APIs, configuration handlers, and data-aggregation routines can introduce SQL injection, path traversal, command injection, hard-coded credentials, and unsafe device-control logic, which may compromise sensing data integrity and system safety. Existing approaches largely rely on static post hoc analysis and lack a unified modeling of the generation process, making it difficult to achieve a principled trade-off between functionality and security. To address this challenge, we propose SafeCodeRL, a framework that integrates multi-agent collaboration with constrained reinforcement learning for trustworthy LLM-generated IoT/CPS software. SafeCodeRL models code generation as a security-aware sequential decision process, where Planner, Code, Security, Test, and Critic agents jointly optimize task decomposition, code synthesis, vulnerability auditing, and sandbox-based validation. We design a constraint-aware policy based on Proximal Policy Optimization, augmented with a Lagrangian mechanism and a shielding strategy to explicitly enforce security constraints. Experiments on real-world engineering and security benchmarks, including SWE-bench, SecurityEval, and CyberSecEval, show that SafeCodeRL reduces high-risk vulnerabilities by over 60% while maintaining high functional correctness. A scenario-level IoT/CPS case study further demonstrates that SafeCodeRL substantially improves secure pass rates for sensor telemetry, edge gateway, configuration-management, and data-aggregation tasks, providing a practical path toward trustworthy AI-assisted software development for sensor-driven systems.

Background/Objectives: This study introduces Lightweight ScaleDense–Transformer (LST-Net), a design-driven framework for whole-lifespan brain age prediction, aiming to improve prediction accuracy while maintaining structural and biological plausibility. Methods: LST-Net utilizes a ScaleDense–Transformer architecture specifically engineered to mirror the dual-nature of brain aging: ScaleDense modules capture fine-grained local anatomical changes, while Transformers model global, network-level reorganizations. The framework was developed and evaluated using a modeling cohort of 22,271 subjects from 17 heterogeneous data sources, covering a 96-year lifespan. Additionally, an auxiliary quantum module (VQC) was incorporated within the latent bottleneck to facilitate nonlinear latent representation transformation and improve representation expressiveness. Performance was primarily assessed using Mean Absolute Error (MAE). Results: LST-Net achieved a competitive MAE of 2.71 years in brain age prediction, demonstrating stability across a wide lifespan (0–96 years) and varying data sources. Longitudinal assessments and test–retest analysis further confirmed the model’s high reproducibility, yielding an Intraclass Correlation Coefficient (ICC) of 0.994. Conclusions: The current findings establish LST-Net as a scalable tool for brain health monitoring and large-scale screenings.

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.

This study evaluates the impact of capillary refill time (CRT) modulation on photoplethysmography (PPG) signals for improved non-invasive continuous blood pressure (CBP) estimation. Data from 21 healthy participants were collected, applying a standardized 9 N pressure for 15 s to induce CRT during 6-min sessions. PPG signals were segmented into 252 paired 30-s intervals (CRT-modulated and standard). Three machine learning models—ResNetCNN, LSTM, and Transformer—were validated using leave-one-subject-out (LOSO) and non-LOSO methods. CRT modulation significantly enhanced accuracy across all models. ResNetCNN showed substantial improvements, reducing mean absolute error (MAE) by up to 35.6% and mean absolute percentage error (MAPE) by up to 40.6%. LSTM and Transformer models also achieved notable accuracy gains. All models met the Association for the Advancement of Medical Instrumentation (AAMI) criteria (mean error < 5 mmHg; standard deviation < 8 mmHg). The findings suggest CRT modulation’s strong potential to improve wearable CBP monitoring, especially in resource-limited settings.

Background/Objectives: Alzheimer’s disease (AD) has a prolonged preclinical phase and marked heterogeneity. Brain age and the Brain Age Gap (BAG), derived from neuroimaging and machine learning (ML), offer a non-invasive, system-level indicator of brain integrity, with potential relevance for early detection, risk stratification, and intervention monitoring. This review summarizes the conceptual basis, imaging characteristics, biological relevance, and explores its potential clinical utility of BAG across the AD continuum. Methods: We conducted a narrative synthesis of evidence from morphometric structural magnetic resonance imaging (sMRI), connectivity-based functional magnetic resonance imaging (fMRI), positron emission tomography (PET), and diffusion tensor imaging (DTI), alongside recent advances in deep learning architectures and multimodal fusion techniques. We further examined associations between BAG and the Amyloid/Tau/Neurodegeneration (A/T/N) framework, neuroinflammation, cognitive reserve, and lifestyle interventions. Results: BAG may reflect neurodegeneration associated with AD, showing greater deviations in individuals with mild cognitive impairment (MCI) and early AD, and is correlated with tau pathology, neuroinflammation, and metabolic or functional network dysregulation. Multimodal and deep learning approaches enhance the sensitivity of BAG to disease-related deviations. Longitudinal BAG changes outperform static BAG in forecasting cognitive decline, and lifestyle or exercise interventions can attenuate BAG acceleration. Conclusions: BAG emerges as a promising, dynamic, integrative, and modifiable complementary biomarker with the potential for assessing neurobiological resilience, disease staging, and personalized intervention monitoring in AD. While further standardization and large-scale validation are essential to support clinical translation, BAG provides a novel systems-level perspective on brain health across the AD continuum.

With the accelerating global population aging, establishing effective brain health assessment systems has emerged as a critical challenge in public health. Neuroimaging-based brain age prediction, serving as a potential biomarker for evaluating individual brain aging, has achieved remarkable breakthroughs in recent years. However, the accuracy of current brain age prediction models remains substantially dependent on the quality and representativeness of their training datasets. Consequently, constructing larger-scale, population-representative, and high-quality datasets is essential for enhancing the reliability of brain age prediction. This systematic review synthesizes findings from 70 peer-reviewed studies (2014–2024) that utilized the UK Biobank (UKB) for brain age prediction, focusing on paradigm-shifting advancements in machine learning and deep learning algorithms. We comprehensively analyze influential factors associated with brain age and their clinical implications, while critically evaluating the unique advantages and inherent limitations of the UKB dataset in this research domain. Furthermore, this work proposes future research directions to address existing methodological gaps and enhance clinical applicability. This study systematically elucidates the advancements in brain age prediction research based on the UKB dataset, aiming to promote deeper exploration in this field and provide theoretical foundations and practical guidance for the precise diagnosis and treatment of neurodegenerative diseases, as well as the formulation of individualized intervention strategies.

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.