Feature representation learning in graph neural networks (GNNs) is a dynamic process driven by progressive information exchange throughout the graph. Current GNNs typically apply pre-defined message-passing heuristics uniformly across all graph data, even when the assumed relational inductive bias conflicts with the intrinsic graph properties (e.g., heterophily). This limitation hinders a principled understanding of graph learning and motivates a shift from application-specific architectures toward a generalizable paradigm. In this work, we conceptualize GNN learning through the lens of mean-field control, where the desired learning outcome corresponds to a variational critical point, each node reaches a representation that is adaptively calibrated to both itself and the entire graph. Building on this formulation, we propose a mathematical framework based on mean-field control (MFC) to adaptively design GNNs for unseen graph data. Specifically, we jointly optimize two control patterns: diffusive mobility, governing information propagation across the graph, and reactive mobility, regulating feature transformation at individual nodes. Both controls are learned from each input instance by solving an MFC variational problem through Hamiltonian flows characterized by partial differential equations (PDE). Our variational framework unifies existing PDE-based GNNs as special mean-field games with fixed control patterns, and yields an end-to-end deep model, termed Nash-GNN. Extensive experiments show that Nash-GNN achieves state-of-the-art performance across diverse benchmarks, including heterophilic graphs and human connectomes. Beyond empirical gains, the MFC framework offers a principled lens to examine the mechanisms of graph representation learning through collective dynamics, opening promising directions for next-generation GNN design.

Modern neuroimaging enables the study of structural connectivity (SC) and functional connectivity (FC) in vivo, yet how alterations in SC-FC coupling relate to neurodegenerative progression remains elusive. Here, we introduce KM-Net, a brain-inspired deep network that models grounded in Kuramoto oscillatory dynamics, which simulates how oscillatory synchrony emerges from the anatomical connectome. The model learns the dynamic coupling between SC and FC, providing an interpretable, biophysically grounded representation of whole-brain dynamics. Through large-scale evaluation across three independent neuroimaging cohorts (ADNI, PPMI, NIFD; participants >1300), KM-Net achieves state-of-the-art diagnostic performance, outperforming leading graph neural networks. Beyond performance, KM-Net uncovers disease-specific subnetwork alterations. Together, this physics-informed framework bridges structural and functional domains to expose the altered coupling mechanisms underlying neurodegeneration, providing a mechanistic route toward explainable brain biomarkers.

Understanding complex interactions between brain regions is critical for early neurodegenerative disease classification such as Alzheimer’s Disease (AD) and Parkinson’s Disease (PD). While graph-based models are widely used to analyze brain networks, most existing approaches primarily focus on pairwise interactions between directly connected nodes, limiting their ability to capture higher-order dependencies across multiple regions. Although hypergraph-based methods have been proposed to model higher-order relations, many rely on predefined hyperedges or restrict learning to hyperedge weights, reducing flexibility and limiting their capacity to capture multi-resolution structural patterns. In this regard, we introduce an adaptive multi-scale hyperedge learning framework, i.e., MuHL, which constructs hierarchical node features and dynamically learns high-order interaction through continuous hyper-edge construction over multi-resolution graph signals. Extensive experiments on multiple brain network benchmarks demonstrate that MuHL consistently improves disease classification performance across different stages, and further identifies key regions of interest (ROIs) and their group-wise interactions from the learned hyperedges that are associated with disease progression, highlighting its potential as a powerful tool for brain network analysis with neurodegenerative disorders.

Human cognition emerges from coordinated spiking dynamics in distributed neural circuits, where information is encoded via both firing rates and precise spike timing determined by brain rhythms. Inspired by this notion, we propose a brain-inspired learning primitive in which cognition-level neural synchrony emerges through iterative bottom-up and top-down interactions between micro-scale dynamics of spiking neurons and a macro-scale mechanism of oscillatory synchronization. Specifically, we model each parcel (e.g., a cortical region or an image pixel) in the target system as a spiking neuron embedded in a predefined connectivity scaffold. Low-level information is encoded in a spatiotemporal domain, where neurons are selectively grouped and fire spontaneously over time through self-organized dynamics. In the bottom-up route, oscillatory synchronization is formed from past spiking activity accumulated over a finite memory window. Since brain dynamics operate in a regime of partial and transient synchronization rather than global phase locking, we model oscillatory coordination using a time-delayed synchronization formulation, which enables a top-down modulation of heterogeneous neural spiking for a large-scale distributed system. Together, we devise a spiking-by-synchronization neural network (S2-Net) that uses rhythmic timing as a control mechanism for efficient information processing. Promising results have been achieved across a broad range of tasks, including neural activity decoding, energy-efficient signal processing, temporal binding and semantic reasoning.

A reliable foundation model of functional neuroimages is critical to promote clinical applications where the performance of current AI models is significantly impeded by a limited sample size. To that end, tremendous efforts have been made to pretraining large models on extensive unlabeled fMRI data using scalable self-supervised learning. Since self-supervision is not necessarily aligned with the brain-to-outcome relationship, most foundation models are suboptimal to the downstream task, such as predicting disease outcomes. By capitalizing on rich environmental variables and demographic data along with an unprecedented amount of functional neuroimages, we form the brain modeling as a multitask learning and present a scalable model architecture for (i) multitask pretraining by tokenizing multiple brain-environment interactions (BEI) and (ii) semi-supervised finetuning by assigning pseudo-labels of pretrained BEI. We have evaluated our foundation model on a variety of applications, including sex prediction, human behavior recognition, and disease early diagnosis of Autism, Parkinson's disease, Alzheimer's disease, and {Schizophrenia}, where promising results indicate the great potential to facilitate current neuroimaging applications in clinical routines.

An unprecedented amount of existing functional Magnetic Resonance Imaging (fMRI) data provides a new opportunity to understand how functional fluctuations relate to human cognition/behavior using data-driven approaches. To this end, tremendous efforts have been made in machine learning to decode cognitive states from evolving volumetric images of blood-oxygen-level-dependent (BOLD) signals. However, due to the complex nature of brain function, the performance and findings of current deep learning models remain inconsistent across tasks, datasets, and evaluation settings. In this work, by capitalizing on large-scale existing neuroimaging data (39,784 fMRI samples from seven databases), we seek to establish a well-founded empirical guideline for designing deep models in functional neuroimaging by linking the methodology underpinning with neuroscientific understanding. Specifically, we put the spotlight on (1) What is the performance landscape of various models in cognitive task recognition and disease diagnosis? (2) What are the key limitations and trade-offs of current deep models? and (3) What is the general guideline for selecting the suitable machine learning backbone for specific neuroimaging applications? We have conducted comprehensive evaluations and statistical analyses across cognitive and clinical scenarios, to answer the above outstanding questions. Our findings demonstrate that no universal model dominates all scenarios; instead, model effectiveness depends on factors such as demographics, task type, and disease stage. Furthermore, we introduce an attention-based interpretability method to reveal spatial patterns of brain activation associated with tasks and disorders.

We present a dynamical-systems based model for resting-state functional magnetic resonance imaging (rs-fMRI), trained on a dataset of roughly $40$K rs-fMRI sequences covering a wide variety of public and available-by-permission datasets. While most existing proposals use transformer backbones, we utilize multi-resolution temporal modeling of the dynamics across parcellated brain regions. We show that MnemoDyn is compute efficient and generalizes very well across diverse populations and scanning protocols. When benchmarked against current state-of-the-art transformer-based approaches, MnemoDyn consistently delivers superior reconstruction quality. Overall, we find that with such large-scale pre-training on (non-proprietary) rs-fMRI datasets, we get a highly performant model for various downstream tasks. Our results also provide evidence of the efficacy of the model on small sample size studies which has implications for neuroimaging studies at large where resting state fMRI is a commonly acquired imaging modality.

Cognition emerges from the adaptive coordination of neural systems responding to internal demands and external stimuli. Capturing evolving cognitive states requires modeling large-scale neural dynamics in a latent low-dimensional space. We propose a Deep Riemannian State-Space Model (DRSSM) that characterizes the temporal evolution of functional connectivity (FC) on the manifold of symmetric positive definite (SPD) matrices. DRSSM integrates static and dynamic FC via an attention-based feature fusion module and employs a feature distillation network for compact yet informative representations. A Riemannian state-evolved network then models FC trajectories within geometric constraints. Evaluated on five fMRI datasets of Alzheimer's and Parkinson's disease, DRSSM outperforms baseline methods in identifying dynamic FC patterns and provides improved interpretability, offering a principled framework for deep learning on non-Euclidean manifolds in neuroimaging.

Transforming raw neuroimage archives into analysis-ready derivatives relies on three brittle stages: data standardization, modality-specific preprocessing, and quality control (QC). While individual neuroimaging tools are well developed, their orchestration requires project-specific scripts, environment-adaptive tuning, and labor-intensive manual QC. To address this, we introduce NeuroPilot, a multi-agent system that digitalizes the expertise of neuroimage processing, QC, and data management into three LLM-invocable skills: dcm2bids-skill, neuroimage-pre-skill, and qc-agent-skill. The LLM-driven agent autonomously orchestrates workflows, generalizing various infrastructure settings into a single configuration to achieve the highest scalability. Demonstrating the system's generalizability, we deployed NeuroPilot across 17 cohorts (>123,000 subjects) spanning infant to aging populations and multiple MRI modalities (structural, diffusion, functional). In practice, after standardizing data via the dcm2bids-skill, the agent dynamically routes datasets to the optimal neuroimage-pre-skill based on available modalities and cohort traits (e.g., dispatching T1w and fMRI data to fMRIPrep, or selecting specialized pipelines for infant cohorts). The qc-agent-skill then drives an evidence-based, semi-automated QC via a 3-D browser dashboard, utilizing a multi-tiered verification system to optimize failed cases and escalate complex issues for supervisor inspection. Quantitatively, our QC agent screened 558 production subjects, validating its automated flags against FreeSurfer's topology-defect metrics. The infant processing pipeline achieved a 100

Despite tremendous success in neuroimaging methodology, making large-scale, high-dimensional datasets ready for AI/ML applications remains a critical operational bottleneck. Conventional workflows require extensive manual effort across metadata curation, pipeline execution, post-processing quality control, and data management, a burden that disproportionately excludes laboratories with limited manpower and computational infrastructure. To address this real-world barrier, there is an urgent need for scalable, cost-effective computational platforms that democratize advanced neuroimaging analytics and accelerate discoveries in mental health and clinical translation. Capitalizing on multi-agent LLM breakthroughs, we introduce CyberNeuro, an agentic workbench with a tailored local LLM-model ('WandaMind') for automated neuroimaging and health-data analysis. Driven by four dedicated agents (Planner, Validator, Dispatcher, and Reporter) communicating via a secure MCP bridge and a pinned execution layer, CyberNeuro enables researchers to execute complex workflows using natural language while maintaining clinical-grade data privacy. On the public NeuroBench suite, CyberNeuro increases held-out domain accuracy from 40

Light-sheet fluorescence microscopy (LSFM) has emerged as a revolutionary imaging modality for investigating intact three-dimensional brain structures at the teravoxel scale. In parallel, high-throughput computational methods, especially deep learning approaches, have opened new avenues for uncovering the pathophysiological mechanisms of neurological diseases through LSFM technology. Recent advances in optics and tissue clearing methods have allowed whole-brain imaging at cellular resolution in three dimensions, and the integration of artificial intelligence has facilitated the identification of disease-related cellular profiles and morphological markers. Machine learning techniques for stitching, segmentation, classification, super-resolution, and registration, therefore, are promoted to uncover biological patterns that are not visible to human eyes yet are related to neuroinflammatory and neurodegenerative diseases. However, analytic pipelines have been designed differently for various animal models and brain structures, leading to challenges in feasibility and compatibility within this emerging field of data-driven LSFM image analysis. Here, we present an overview of current pipelines, examine existing and forthcoming challenges as the LSFM community advances, demonstrate their implications for neurological disease applications, and propose potential solutions.

Functional magnetic resonance imaging (fMRI) derived functional connectivity (FC) is represented as graphs and as correlation or covariance matrices that live on non-Euclidean spaces, cortical graphs and the Riemannian manifold of symmetric positive-definite (SPD) matrices, thus conventional Euclidean sequence models are misspecified. To this end, we introduce Geo-Mamba, a geometric variant of Mamba formulated on Riemannian manifolds. Geo-Mamba employs a dual-path selective state-space design, (1) a stacked path performs hierarchical modeling by aggregating pyramid multi-granular features to capture short- and long-range dependencies; and (2) a distillation path combats redundancy in high-dimensional SPD inputs via progressive, geometry-aware dimensionality reduction (operating in the manifold spaces) to produce compact states without violating Riemannian constraints. Their complementary outputs are fused through the tailored GeoMix operator to yield a compact, discriminative SPD representation. Geo-Mamba is evaluated on seven public fMRI datasets, including two Alzheimer’s disease cohorts, three Parkinson’s disease cohorts, one Autism dataset, as well as a longitudinal single-site, single-scanner study designed for detecting subtle changes in the brain due to a season of playing contact sports. To further evaluate the cross-modal applicability and scalability of the model, we apply Geo-Mamba to three electroencephalography (EEG) datasets. Across these benchmarks, it delivers consistently competitive accuracy and robustness, supporting the value of dual-path manifold modeling for neuroimaging and its potential for clinical translation. The code is released at https://github.com/acmlab/Geo-Mamba.

Neural coupling is a fundamental mechanism in neuroscience that facilitates the emergence of cognitive functions through dynamic interactions and synchronization among distributed brain regions. Inspired by this principle, we pose the question: Might the biological mechanism of neural oscillatory synchronization inspire the feature representation learning for neuroscience? By addressing this question through the Kuramoto model, renowned for simulating oscillatory dynamics, we present a novel physics-informed deep model, `SyncBrain`, it models brain regions as interacting oscillatory units and simulates their temporal dynamics and synchronization patterns to distinguish cognitive states. Furthermore, inspired by the brain's inherent ability to dynamically attend to critical temporal information, we incorporate an adaptive control module that introduces an attention-like mechanism to guide information flow. We evaluate our model on multiple functional neuroimaging datasets, it demonstrates promising performance and enhanced interpretability in both cognitive state decoding and early disease diagnosis, outperforming existing computational methods. These results demonstrate the effectiveness of neural oscillatory mechanisms in shaping robust and interpretable machine learning models for neuroscience applications.

Alzheimer's disease (AD) is a progressive neurodegenerative disorder that severely impairs survival and quality of life. While anti-amyloid beta (Aβ) therapies can slow disease progression, their efficacy depends on personalized dosing that maximizes benefits and minimizes risks, such as amyloid-related imaging abnormalities (ARIA). Mathematical modelling offers a powerful tool for understanding AD dynamics and optimizing treatment, yet most models focus solely on temporal behaviour, overlooking spatial heterogeneity within the brain. In this study, we propose a spatially explicit reaction-diffusion model to describe Aβ plaque dynamics. We formulate an optimal control problem to minimize plaque concentration while balancing therapeutic efficacy and treatment risk. Under reasonable assumptions, we establish well-posedness and uniqueness of the optimal solution. A finite element method (FEM)-based numerical framework is developed to compute personalized treatment strategies. Our model is calibrated using longitudinal Aβ positron emission tomography (PET) data from the Alzheimer's Disease Neuroimaging Initiative (ADNI), enabling estimation of patient-specific parameters, such as growth rate and effective diffusivity. Results show that optimized treatment strategies consistently outperform constant dosing regimens across patient groups, achieving substantial reductions in cumulative amyloid burden while minimizing side effects. This integrated, data-driven framework advances personalized, spatially informed therapeutic optimization for AD.

Traffic congestion remains a persistent barrier to mobility and efficiency, especially in developing regions with limited infrastructure. Addressing this challenge requires robust traffic flow modeling, which remains challenging due to nonlinear spatiotemporal dependencies and external perturbations. Current modeling approaches primarily characterize temporal fluctuations at the regional scale, without a comprehensive formulation of the governing dynamics. Here, we present TrafficFlowNet, a physics-informed deep learning model that formulates traffic flow as a transport process across interconnected road nodes. By embedding the transport dynamics within a graph neural network (GNN) backbone, TrafficFlowNet models directional flows driven by traffic-volume gradients. We further incorporate total variation (TV) regularization into a min-max optimization framework to address the oversmoothing issue common in GNNs and adopt a generative adversarial structure to capture realistic dynamics. In experiment, TrafficFlowNet outperformed sixteen state-of-the-art methods across real-world highway traffic sensor benchmarks (PEMS03, PEMS04, PEMS07, and PEMS08, each including >150 sensors, >15 000 time steps). On PEMS03 data, TrafficFlowNet achieves a mean absolute error (MAE) of 13.314, representing up to 8.43% improvements over prior best-performing model (14.541). These results highlight the potential and foundational insight of our physics-grounded framework for traffic prediction.

Emerging neuroimaging evidence shows that pathological tau proteins build up along specific brain networks, suggesting that large-scale network architecture plays a key role in the progression of Alzheimer’s disease (AD). However, how the interaction between structural (SC) and functional connectivity (FC) is associated with tau propagation remains largely unexplored. Using longitudinal neuroimaging data, we develop a multi-layer graph diffusion model to investigate this interplay. We identify a regionally asymmetric pattern, FC dominates tau accumulation in subcortical, insular, and frontotemporal regions, whereas SC prevails in occipital, parietal, and limbic areas. The principal conduit of tau propagation shifts from FC-mediated spread in early stages to SC-constrained diffusion as the disease advances. Furthermore, these propagation patterns correlate with regional expression of AD-associated genes (e.g., CHUK, TMEM106B, MCL1, NOTCH1, TH) and are selectively modulated by risk factors like APOE genotype and biological sex. By disentangling these dynamic pathways, our framework enables personalized trajectory prediction and refined patient stratification. These findings offer mechanistic insights into selective regional vulnerability, providing a quantitative basis for stage-specific therapeutic targeting and the optimization of clinical trial designs through network-based biomarkers. Our results are validated in an independent cohort, yielding consistent findings.

Longitudinal brain imaging data facilitate the monitoring of structural and functional alterations in individual brains across time, offering essential understanding of dynamic neurobiological mechanisms. Such data improve sensitivity for detecting early biomarkers of disease progression and enhance the evaluation of intervention effects. While recent matrix-response regression models can relate static brain networks to external predictors, there remain few statistical methods for longitudinal brain networks, especially those derived from high-dimensional imaging data. We introduce a matrix-response generalized linear mixed model that accommodates longitudinal brain networks and identifies edges whose connectivity is influenced by external predictors. An efficient Monte Carlo Expectation-Maximization algorithm is developed for parameter estimation. Extensive simulations demonstrate effective identification of covariate-related network components and accurate parameter estimation. We further demonstrate the usage of the proposed method through applications to diffusion tensor imaging (DTI) and functional MRI (fMRI) datasets.

Mutations in the chromatin remodeler CHD8 are associated with autism and macrocephaly. While mouse models of Chd8 haploinsufficiency recapitulate brain overgrowth, the specific cellular mechanisms and developmental timing that lead to these anatomical abnormalities remain poorly understood. Here, we conducted 3D imaging of Chd8 V986*/+ mouse brains using magnetic resonance imaging followed by tissue clearing and cellular resolution light-sheet microscopy across embryonic and postnatal developmental stages. We found that brain overgrowth occurs postnatally, driven by an expansion of oligodendrocytes and microglia. Unexpectedly, we identified prevalent molecular layer heterotopias within the frontal cortex of Chd8 V986*/+ mice appearing during embryonic development and persisting throughout life. Molecular layer heterotopias were previously identified in post-mortem brains from individuals with autism and other neurodevelopmental disorders, suggesting functional significance in human patients.

Despite the popular of multimodal statistical models, there lacks rigorous statistical inference tools for inferring the significance of a single modality within a multimodal model, especially in high-dimensional models. For high-dimensional multimodal generalized linear models, we propose a novel entropy-based metric, called the expected relative entropy, to quantify the information gain of one modality in addition to all other modalities in the model. We propose a deviance-based statistic to estimate the expected relative entropy, prove that it is consistent and its asymptotic distribution can be approximated by a non-central chi-squared distribution. That enables the calculation of confidence intervals and p-values to assess the significance of the expected relative entropy for a given modality. We numerically evaluate the empirical performance of our proposed inference tool by simulations and apply it to a multimodal neuroimaging dataset to demonstrate its good performance on various high-dimensional multimodal generalized linear models.