Severe and enduring psychiatric illnesses, including melancholia and bipolar illness, show prolonged deviations in motivation and mood yet lack a unifying account of their long-term dynamics. Solomon's opponent-process theory provides a qualitative framework for short-timescale affective responses, composed of a fast stimulus-locked a-process opposed by a slower adaptive b-process. Here we evaluated whether these dynamics can be implemented as a computational homeostatic controller, and whether distinct clinical trajectories emerge as canonical failure modes of the same system. We formulated a tractable control-systems model with feedforward a- and b-processes, and examined its behaviour across minute-scale and month-scale regimes. Under "healthy" parameters, the model reproduced classical opponent-process responses. Altering only opponent-process gain and decay generated a prolonged downward drift, matching the clinical timescale and asymmetry of melancholia. Reducing damping within the same controller produced an endogenous underdamped oscillation with long period and phase asymmetry characteristic of bipolar illness. Together, these proof-of-principle simulations suggest that severe severe affective illnesses may be expressed as distinct dynamical regimes of a single motivational homeostat. This framework generates testable predictions and may facilitate experimental quantification of opponent-process recovery, damping, and gain as mechanistic markers of severe affective illnesses. The online version contains supplementary material available at 10.1007/s11571-026-10507-2.
Social communication (SC) relies on the integration of linguistic processing, pragmatic inference, and theory of mind (ToM), yet its neural architecture remains insufficiently characterized. We propose and empirically test a unified SC network by integrating information neural graph theory metrics, multi-domain cognitive modeling, and autistic traits. Forty-five neurotypical adults completed a cognitive battery assessing language, executive functions, social cognition, perceptual reasoning, and the autism spectrum quotient (AQ). Global efficiency, local efficiency, and clustering coefficients were computed for language, pragmatic, and ToM networks, and their combined architecture. A replication analysis for graph-metrics and its relationship with AQ was performed using an independent sample (n = 73, 31 autistic). Results revealed that language abilities were the strongest and most consistent predictors of network efficiency, particularly within temporal-parietal nodes implicated in semantic integration and contextual interpretation. Executive functions selectively predicted efficiency within frontal control regions, while perceptual reasoning was associated with global efficiency of the precuneus, associated with social and inferential processing. Importantly, autistic traits moderated multiple brain-behavior relationships, indicating that trait-level variability shapes how cognitive abilities map onto neural efficiency within neurotypical population. The replication analysis showed partial overlap with graph-metric and AQ results. These findings advance a network-level account of SC, demonstrating that communicative competence emerges from dynamic interactions among linguistic, executive, and inferential systems, whose neural organization is tuned by individual cognitive profiles and autistic traits. This dimensional framework provides a foundation for understanding variability in social-cognitive functioning and has implications for personalized models of communication. The online version contains supplementary material available at 10.1007/s11571-026-10481-9.
This study aims to address a theoretical gap in how people formulate cognitive maps in a social context. The concept of a cognitive map, a mental representation of spatial relations, has been widely used to explain individual navigation. When navigating with others, the exchange of spatial knowledge and the formation of cognitive maps in leaders and followers are natural and ubiquitous, yet their neural underpinnings remain insufficiently understood. We investigated these mechanisms by recording hyperscanning electroencephalography (EEG) from 70 participants engaged in dyadic route-planning and navigation tasks within a virtual reality environment. Intrabrain and interbrain couplings were analyzed across frequency bands using connectivity measures. We observed robust neural synchronization patterns associated with collaborative navigation performance and role. Intrabrain connectivity analyses showed increased delta coupling in both leaders and followers, whereas theta connectivity was particularly enhanced in followers. Alpha-band connectivity displayed divergent patterns between roles, suggesting distinct neural strategies for spatial processing. Interbrain analyses revealed increased delta causality between partners but decreased theta and gamma couplings from followers to leaders. Faster-performing dyads exhibited overall reduced interbrain coupling, especially in theta bands, indicating more efficient neural coordination. Collaborative navigation relies on frequency-specific, role-dependent neural coordination both within and between brains, and more efficient dyads may require less sustained interbrain coupling. These findings can inform the design of navigation aids, training protocols, and collaborative interfaces that better support leader-follower coordination and improve joint spatial decision making in real-world settings. The online version contains supplementary material available at 10.1007/s11571-026-10521-4.
This study aimed to characterize alterations in functional brain network organization in miners with approximately ten years of occupational exposure to extreme working conditions and shift work, using connectivity metrics derived from resting-state EEG recorded during both eyes-open (EO) and eyes-closed (EC) conditions. Directed Transfer Function (DTF), the imaginary part of Coherence, and the weighted Phase Lag Index were computed from non-overlapping 6-s epochs following two preprocessing pipelines: Independent Component Analysis and Artifact Subspace Reconstruction (ASR). Analyses were conducted for both miners (19 men, mean age 36.52 ± 5.08 years) and matched controls (19 men, mean age 35.42 ± 5.04 years). DTF demonstrated consistently excellent reliability (Intraclass Correlation Coefficients ≥ 0.75) and revealed significant group differences across all frequency bands when combined with ASR, as determined by linear mixed-effects models with false discovery rate correction (pc < 0.05), underscoring its high reproducibility. Specifically, miners exhibited reduced network segregation and integration, reflected by decreases in modularity (Q), global efficiency (GE), local efficiency (LE), clustering coefficient (CC), and transitivity (T) in the delta and theta bands, as well as reduced network resilience (R) at higher frequencies in the EC condition. Within the miner group, higher GE was associated with poorer executive function and slower processing speed, as measured by Trail Making Test subcomponents (0.51 ≤ r ≤ 0.71; 0.0008 ≤ pc ≤ 0.029). In addition, lower-frequency network metrics (CC, LE, T, and R) showed significant negative correlations with verbal recall performance (- 0.70 ≤ r ≤ - 0.54; 0.0009 ≤ pc ≤ 0.039). Collectively, these findings indicate that chronic occupational exposure disrupts the stability and large-scale organization of functional brain networks, resulting in reduced network efficiency and a decoupling between neural connectivity and cognitive performance. From a methodological perspective, the combination of DTF and ASR emerged as the most reliable approach for resting-state EEG connectivity analysis.
Numerous studies have made significant contributions to understanding resting-state brain networks, advancing the field of neuroscience. Studying dynamic functional connectivity is essential for capturing the temporal evolution of brain network organization. However, investigations of task-related functional networks, especially those assessed through dynamic connectivity during continuous cognitive tasks, remain relatively sparse. This study aims to investigate the temporal dynamics of functional prefrontal networks during continuous cognitive control through dynamic functional connectivity analysis. In contrast to conventional methods that primarily focus on identifying static spatial connectivity patterns, this study applies temporal group independent component analysis (TG-ICA) to functional near-infrared spectroscopy (fNIRS) data acquired during a continuous Stroop Color-Word task. This approach enables the identification of temporally evolving functional connectivity networks at the group level. The results revealed three distinct and interpretable prefrontal network components, including a left-lateralized system for early rule implementation and executive control, a right-dominant network for conflict monitoring and attentional reallocation, and a bilateral network supporting sustained goal maintenance and cognitive stability. These networks exhibited time-varying engagement aligned with different stages of cognitive control during continuous tasks. The findings highlight the utility of TG-ICA in capturing the spatiotemporal characteristics of functional brain networks and offer new insights into the dynamic organization of the prefrontal cortex during executive functioning. The online version contains supplementary material available at 10.1007/s11571-026-10467-7.
Understanding how cognition unfolds from neurophysiological signals presents a promising direction for cognitive science studies and wearable-enabled human-robot interaction applications. However, uncovering latent neurodynamic geometry and temporal progression remains challenging and underexplored due to the lack of observable temporal organization for annotation and, consequently, the difficulty of training models in a supervised manner. This study proposes a representational learning method for this segmentation problem that shifts the solution away from statistical change-point detection methods and Hidden Markov Models. Our method employs self-supervised learning to discover emergent properties of the underlying temporal organization directly from the neurodynamic data itself. Four objectives are introduced and jointly optimized, including within-stage temporal predictability, boundary contrast, cross-trial alignment, and sparse stage-specific feature weights. Population-based evolutionary search was adopted to explore the multiple-basins-of-attraction landscape, where mutation and crossover govern the convergence process. We validated the framework on EEG recordings collected from participants performing an embodied road-crossing decision-making task, which simulates a typical cognitive processing transition from perceptual assessment to risk evaluation and decision commitment. Results showed that our method achieves an order-of-magnitude improvement in boundary contrast of the discovered stages, indicating that the learning behavior fundamentally changes the working principle from seeking local statistical consistency to capturing higher-order global temporal organization. This inter-stage divergence serves as the driving force for latent regime discovery while preserving local temporal continuity and coherence. Ablation and sensitivity studies demonstrate that the model performance is robust in identifying cross-trial transferable state geometry and handling data variability introduced by subject and stimulus heterogeneity. The reconstructed cognitive stages are also behaviorally plausible, and the dimensions attended by the model are well aligned with the neurophysiological underpinnings governing critical cognitive activities underlying each stage.
As social beings, humans make decisions partly based on social interaction. Observing the behavior of others can lead to learning from and about them, potentially increasing trust and prompting trust-based behavioral changes. Observation-based decision making involves different neural structures. The orbitofrontal cortex (OFC) and lateral prefrontal cortex (LPFC) are known as neural structures mainly involved in processing emotional and cognitive decision values, respectively, while the anterior cingulate cortex (ACC) plays a pivotal role as a social hub, integrating the afferent expectancy signals from the OFC and LPFC. This paper presents a neurocomputational model of the interplay between observational learning and trust, as well as their role in individual decision making. Hence, our model provides a framework for investigating how emotional and rational responses may change when individuals observe the action-outcome associations of an alleged expert. We have modeled the neurodynamics of three cortical structures (OFC, LPFC, and ACC) and their interactions, where the neural oscillatory properties, modeled with Dynamic Bayesian Probability, represent the observer's attitude towards the expert and the decision options. As an example of an everyday behavioral situation related to climate change, we use the choice of transportation between home and work. The model generates EEG-like signals that show how patterns of neural activity change during observation-based decision making. The simulations suggest that higher levels of trust influence both emotional and rational evaluations when individuals observe the actions and outcomes of an expert. Overall, the proposed framework provides insight into how observational learning and trust work together to shape decision making. It highlights the dynamic interplay between emotional and cognitive processes and offers a mechanistic understanding of how social information can influence behavior.
Embodied and embedded cognition (EEC) theory proposes that language and cognitive development emerge from bodily interactions with the environment, yet empirical tests of this claim in clinical developmental populations remain rare. This mini-review synthesizes behavioral, electrophysiological, and structural neuroimaging evidence from children with serious early motor disorders-obstetric brachial plexus palsy and arthrogryposis multiplex congenita-which restrict upper limb movement from birth or before, providing a unique opportunity to test EEC predictions in a motor-restricted population. The results reveal a gradient of cognitive and linguistic alterations: from domain-specific deficits in action-verb semantics and verbal fluency, to broader impairments in memory, categorical reasoning, and naturalistic neural processing. Based on these multimodal findings, we propose that early sensorimotor restriction does not only affect motor systems but may shape the neurodevelopmental trajectory of language and other distributed cognitive architectures through mechanisms of embodied grounding.
Accurate identification of Alzheimer's disease (AD) and its early stages, namely subjective cognitive decline (SCD) and mild cognitive impairment (MCI), is crucial for timely intervention. Electroencephalography (EEG) is widely used for AD assessment due to its non-invasiveness and high temporal resolution; however, its non-stationarity, noise interference, and individual variability make classification more difficult. To address this, this paper proposes a self-supervised learning framework, TF-JointMAE (Temporal-Frequency Joint Masked Autoencoder), that jointly models temporal and time-frequency representations of EEG and incorporates age information as a conditional physiological prior within a unified embedding space. By performing self-supervised masked autoencoder pre-training on multiple public EEG datasets, the model learns consistent and robust EEG representations, thereby improving AD-spectrum classification under limited-label conditions. On the publicly available CAUEEG dataset (Normal, SCD, MCI, Dementia) and the olfactory EEG dataset (Normal, MCI, Dementia), TF-JointMAE achieved test accuracies of 77.97% and 96.43%, respectively, and demonstrated higher discriminative stability in the MCI category. Further occlusion-sensitivity analysis revealed that the model showed varying sensitivity to EEG channels and time-frequency regions across cognitive states. These results demonstrate that TF-JointMAE effectively improves the robustness of EEG representations, providing potential auxiliary support for AD-spectrum classification and clinical decision-making. All source code is publicly available to support reproducibility (https://github.com/redpig-zhu/TF-JointMAE). The online version contains supplementary material available at 10.1007/s11571-026-10501-8.
Mind-wandering (MW), a form of attentional disengagement during operational tasks, has emerged as a critical cognitive risk factor for unsafe behavior in high-risk industries. Moreover, the disruption of circadian rhythms induced by shift work further amplifies the risks associated with MW in operational contexts. However, existing research has largely relied on short duration experiments conducted under controlled laboratory conditions, leaving the neural mechanisms of MW in authentic high risk operational contexts insufficiently understood. Even fewer studies have empirically examined how shift work influences MW from the perspective of brain network organization. In response to this research gap, the present study employs Lab-in-the-Field experiments to analyze the brain network characteristics of workers in high-risk industries, aiming to elucidate the potential mechanisms by which MW and shift work jointly impact operational safety. The results demonstrated that MW significantly prolonged response times, and that shift work further exacerbated this negative effect. EEG based brain network analysis further showed that MW was accompanied by changes in brain network resource allocation, with a particularly evident decline in hub efficiency in the left prefrontal cortex. Moreover, shift work was found to affect cognitive states predominantly by altering the structural topology of brain networks, rather than by simply modulating the overall strength of functional connectivity. This study validates the brain network reorganization underlying MW and provides scientific evidence for developing cognitive state monitoring tools and targeted intervention strategies based on brain network topological features.
Parkinson's disease (PD) is a progressive neurodegenerative disorder that severely affects motor and cognitive functions, making early and accurate diagnosis crucial for effective clinical management. This research introduces the high-frequency substantia nigra and ventral tegmental area fusion network (HF-SNVTA-FusionNet), a robust EEG-based PD detection framework. The system employs independent component analysis (ICA) for artifact removal, multi-domain feature extraction (time, frequency, and time-frequency), and principal component analysis (PCA) for dimensionality reduction, followed by a CNN-BiGRU-multi-head self-attention (MHSA) classification pipeline. Specifically, CNN captures local spatial patterns, BiGRU models bidirectional temporal dependencies, and MHSA refines salient features, enabling improved discrimination between PD and healthy EEG signals. Experiments on three benchmark datasets (UI, PDG, and USDRS) demonstrate superior performance, achieving accuracies of 100%, 98.89%, and 98.15% within the 50-70 Hz band using 64 PCA features. Comparative evaluation confirms its advantage over existing state-of-the-art models. The proposed system holds strong potential for real-world PD screening and can be extended to cognitive task-based EEG analysis and low-power hardware deployment.
Visual-spatial attention (VSA) selects relevant sensory information and supports the preparation of responses to this information. Mental rotation (MR) is the ability to rotate an object seen from a certain perspective to a new orientation in space. Exercise stands out as a promising non-pharmacological treatment for cognitive functions. Balance control is known to be related to the visual system. Therefore, the aim was to investigate the effectiveness of video-based balance games and structured balance exercises on VSA and MR with EEG brain oscillations. 30 healthy participants were included in the study. Participants were divided into two groups (structured balance exercises group (SBEG) and video-based balance exercises group (VBBEG)) by randomization. Both groups received exercise sessions 2 days a week for a total of 6 weeks. The mentioned cognitive functions were evaluated by selecting tests previously used in the literature. For the VSA task, after 6 weeks of exercise, occipital theta (4-7 Hz) power decreased in the VBBEG group, while SBEG increased. In the MR task results, high alpha (11-13 Hz) power decreased in VBBEG and increased in SBEG when centroparietal areas were examined. In conclusion, it is thought that the two different exercise methods may affect visual-spatial attention and mental rotation skills in different ways. The online version contains supplementary material available at 10.1007/s11571-026-10494-4.
The human ability to smell functions as a critical cognitive function because it enables people to detect their surroundings while experiencing feelings and recalling memories and making choices. Researchers face difficulties when they use electroencephalography (EEG) to study how the brain responds to smells because olfactory brain signals produce low signal-to-noise ratios and different people show different response patterns and researchers lack established olfactory EEG databases for their studies. The study proposes a simulation-based framework which enables researchers to study olfactory EEG signals through power spectral density (PSD) analysis. The research team created a simulated olfactory EEG dataset which simulated the responses of fifty virtual participants who experienced two distinct odor categories of pleasant rose and unpleasant rotten at three different concentration levels of low medium and high to create six separate olfactory conditions. The simulated EEG signals included 45 channels which recorded data at a 256 Hz sampling rate. Welch's method estimated PSD features for five canonical EEG frequency bands which included delta theta alpha beta and gamma after the data underwent band-pass filtering at the 0.5-70 Hz range. The researchers used Stratified 10-fold cross-validation to evaluate the band's characteristics which they had developed as training data for their multiclass support vector machine (SVM) classification model. The PSD-based features demonstrated their ability to distinguish between different olfactory conditions in controlled tests which showed the system's classification accuracy of 99.67% and macro-averaged F1-score of 0.99. The research provides a methodological validation platform which enables scientists to conduct reproducible olfactory EEG studies through their complete pipeline of interpretation. The proposed framework establishes the essential foundations for subsequent research which will assess and develop these techniques through actual human olfactory EEG data in cognitive neuroscience studies.
Alzheimer's Disease (AD) is a disease of the brain that lowers quality of life due to cognitive impairment. A correct diagnosis is therefore essential for timely interventions and follow-up care for executives. However, existing deep learning methodologies for neuroimage-based diagnosis seriously lack reliability due to decreased accuracy and increased false positive rates, both of which can lead to misdiagnosis and suboptimal care planning. This study proposed an innovative Steerable Quantum Probabilistic Hamiltonian Generative Modeling with Adaptive Chaotic Satin Bowerbird Optimization (SQPHGM-ACSBO) framework for diagnosing AD using neuroimages within the ADNI dataset. A total of 5,154 neuroimages from the ADNI dataset were utilized in this study, comprising 2,590 MCI, 1,124 AD, and 1,440 cognitively normal (CN) samples. The process begins with image enhancement via an Adaptive Self-Guided Loop Filter, followed by precise brain region segmentation with GoogLeNet Inception-v3, and advanced feature extraction using a new Discrete Cosine-Krawtchouk-Tchebichef Transform (DCKTT). Classification is performed using the SQPHGM model, which combines the strengths of Steerable Transformers and Quantum-Probabilistic Hamiltonian Learning to model complex neuroimaging patterns, while the Adaptive Chaotic Satin Bowerbird Optimization (ACSBO) algorithm optimizes classification performance. With 99.9% accuracy and 99.8% precision, the suggested method outperforms current methods and provides a dependable, high-performance solution for AD diagnosis from neuroimaging data, according to experimental results. The main innovation of this work is the all-in-one optimization of steerable transformer-based directional feature learning, quantum-probabilistic Hamilton generator modeling, and adaptive-chaotic satin bowerbird optimization integrated within a single diagnostic pipeline, which allows for representing the complicated neuroimaging patterns better, and the convergence stability is also increased, compared to the previous deep learning-based Alzheimer diagnosis models.
Biological sensory systems achieve remarkable robustness through efficient cross-modal integration, yet replicating this in artificial spiking neural networks (SNNs) remains a challenge. Inspired by Mayer's multi-channel learning cognitive theory, we present CrossModal-Associated-SNN, a neuro-inspired framework that synergistically integrates visual and auditory information via Spike-Timing-Dependent Plasticity (STDP) clustering and associative learning. The architecture employs a multi-channel, multi-network design for modality-specific processing, followed by a cross-channel complementary strategy that refines decision-making through associative signals. Evaluated on small-sample benchmarks (MNIST3K and Spoken-MNIST3K), the model demonstrates superior generalization and robustness in multi-modal classification. Compared to the single-channel baseline (91% accuracy), the dual-channel dual-network improved visual and auditory recognition to 93% and 84%, respectively, with a fusion accuracy of 94%. The dual-channel triple-network architecture further maximized performance, attaining 96% (visual), 90% (auditory), and a peak 97% cross-modal fusion accuracy. These results suggest that cooperative shallow micro-networks, akin to biological small-neuron ensembles in superficial brain regions, offer a potentially energy-efficient alternative for multimodal tasks processing. CrossModal-Associated-SNN represents a critical step toward mimicking human sensory cognitive integration, offering a biologically plausible solution for energy-efficient, multi-modal intelligent systems.
Elucidating the language-brain relationship requires bridging the methodological gap between linguistics' abstract theoretical frameworks and neuroscience's empirical neural data. As an interdisciplinary cornerstone, computational neuroscience formalizes language's hierarchical and dynamic structures into testable neural representation models through modeling, simulation, and data analysis, enabling computational dialogue between linguistic hypotheses and neural mechanisms. Recent advances in deep learning, particularly large language models (LLMs), have further advanced this inquiry: their high-dimensional representational spaces provide a new scale for probing the neural basis of linguistic processing, the model-brain alignment framework offers a principled approach to evaluating the biological plausibility of language-related theories, provided that representational correspondence is interpreted together with behavioral, temporal, causal, and biological constraints. This review synthesizes interdisciplinary progress from a computational neuroscience perspective. First, it outlines the core connotations of major linguistic frameworks (generative grammar, functional linguistics, and cognitive linguistics), their cross-cultural and evolutionary characteristics, and key challenges for neural alignment, including limited quantitative mechanisms, poor accessibility of abstract constructs to neural measures, and insufficient treatment of dynamics and plasticity. Second, it introduces the methodological foundations of linguistics-neuroscience dialogue, focusing on four technical pillars: neural activity measurement (e.g., fMRI, EEG, MEG, fNIRS, ECoG, SEEG), linguistic numerical representation, the evolution of language models from statistical approaches to LLMs, and neural coding frameworks that link model representations to brain signals, illustrated with a model-brain alignment case study. Third, it summarizes major findings, ranging from early computational insights into predictability and structural processing to recent LLM-driven progress in cross-modal interaction, inter-brain coupling, hierarchical computation, learning strategy sensitivity, and language plasticity. Finally, the review discusses current limitations-including functional alignment without structural homology, constraints on real-time validation, biased research coverage, and narrow evaluation metrics-and proposes future directions, such as exploring whether spiking neural network-based language models can improve biological plausibility in settings requiring temporally precise and event-driven neural modeling, developing cognitive-level alignment frameworks integrating memory, causality, and metacognition, and extending clinical applications. In summary, this work aims to advance a comprehensive, mechanistic understanding of the language-brain relationship and promote computational neuroscience as a generative theoretical framework for testable neuro-computational accounts of language.
Oxytocin modulates social information processing by altering excitatory-inhibitory balance at the microcircuit level, but how such local modulation gives rise to selective processing at the level of distributed brain systems remains unclear. Here, we investigated the effects of oxytocin on large-scale neurodynamics across cortico-limbic network in the mouse brain using multisite local field potential recordings. Oxytocin selectively enhanced neural responses to infant calls in the auditory cortex (AC) and medial prefrontal cortex (mPFC). These enhancements occurred while baseline activity was reduced, indicating increased signal-to-noise ratio rather than a global increase in excitability. During auditory steady-state responses (ASSRs), oxytocin increased prefrontal phase coherence without altering ASSR power. During rest, oxytocin induced a transient, broadband reduction in spontaneous spectral power across regions. Despite this reduction in activity, analyses of interregional interactions revealed a selective increase in low-theta phase coupling and directional connectivity of AC→mPFC. Session-level analyses showed that stronger bottom-up AC→mPFC coupling was associated with lower prefrontal power, consistent with a gating or disinhibitory network regime favoring sensory-to-prefrontal information transfer. Multivariate analyses showed that oxytocin/saline conditions were reliably discriminable using supervised classification models, with specific contributions from spectral power, phase-locking, and Granger-causal connectivity features. Conversely, unsupervised dimensionality reduction did not identify a distinct low-dimensional manifold separating conditions, although a modest shift in the centroid of neural state space was observed. Together, these results indicate that oxytocin reduces background neural activity while selectively enhancing sensory-prefrontal network interactions, providing a systems-level account linking local inhibitory modulation to selective processing of socially salient infant cues.
Early and accurate detection of Mild Cognitive Impairment (MCI) is essential for preventing progression toward Alzheimer's disease (AD). In this cross-subject study, we investigate the effectiveness of entropy- and graph-based EEG features for distinguishing MCI from healthy controls (HC), using two modeling approaches: (1) a Transformer network applied to the engineered feature set, and (2) an EEGNet model trained on the same feature representation for comparison. The dataset consists of resting-state, eyes-closed EEG recordings from 183 participants (127 HC, 56 MCI), collected using a 20-channel STAT™ X24 wireless system and segmented into 3-second epochs. EEG data underwent standard preprocessing, including band-pass filtering, downsampling, normalization, and class-balancing augmentation applied to the minority class. From each channel, nonlinear dynamical measures (e.g., sample and fuzzy entropy, Higuchi fractal dimension, Lyapunov exponent) and graph-theoretic connectivity descriptors derived from coherence matrices across five frequency bands were extracted, yielding a structured 19[Formula: see text]77 feature representation. The feature-based Transformer achieved the best performance (97.04% ± 0.72), outperforming the feature-based EEGNet baseline and highlighting the benefits of combining rich handcrafted features with attention-based modeling. SHAP (SHapley Additive exPlanations) analysis provided global and local interpretability, revealing the most influential nonlinear and connectivity features as well as the EEG channels contributing most to classification. Overall, these results demonstrate the effectiveness of feature-Transformer integration and support the potential of interpretable feature-driven deep learning models for early MCI detection.
Paroxysmal kinesigenic dyskinesia is a rare neurological disorder characterized by brief, recurrent motor attacks that significantly impair quality of life. Prior studies have largely relied on unimodal data, which offer partial insights into neural regulation but are constrained by trade-offs between temporal and spatial resolution. To address this limitation, we developed a multimodal recognition and tracing framework integrating electroencephalography and functional magnetic resonance imaging. We propose GTBL-AF, a deep multimodal neural architecture that captures spatial connectivity and temporal dynamics of brain function through graph attention, Transformers, and bidirectional long short-term memory networks, with cross-attention enabling modality-level fusion. GTBL-AF achieved 94.2% classification accuracy in paroxysmal kinesigenic dyskinesia recognition, significantly outperforming unimodal methods. Incorporating dipole-based electroencephalography source localization and phase-locking value connectivity, we observed increased temporal complexity and reorganized functional connections in key cortical regions, including the prefrontal cortex, temporal pole, and parietal association areas. Whole-brain analyses using sample entropy and small-world metrics revealed greater dynamic uncertainty and enhanced small-world properties in paroxysmal kinesigenic dyskinesia patients, indicative of compensatory neural regulation. Furthermore, network-based statistics identified aberrant synchronous connectivity within circuits mediating cognitive control and motor initiation. This study presents a deep EEG-fMRI multimodal fusion framework for PKD and provides evidence of widespread network reorganization. These findings may contribute to a better understanding of PKD pathophysiology and provide a methodological reference for future multimodal-assisted diagnosis and individualized clinical assessment. The online version contains supplementary material available at 10.1007/s11571-026-10504-5.
The brain's higher cognitive functions rely on the dynamic integration of multimodal information, supported by coordinated neural oscillations. However, current neuromodulation approaches often fail to reliably control these dynamics due to high inter-individual variability and limited mechanistic grounding. To address this challenge, we introduce the Synergistic Entrainment Framework, which conceptualizes the brain as a dynamical system in which neural oscillations create periodic windows of excitability that regulate when information can be effectively processed. Within this framework, rhythmic inputs from different physical modalities (e.g., sensory streams and electromagnetic stimulation) can interact nonlinearly, leading to amplification or stabilization of neural responses, when their timing is aligned with ongoing cortical phase dynamics. We propose that this interaction operates across three coupled scales: (i) microscopic constructive interference that amplifies local neural responses, (ii) mesoscopic phase resetting that aligns regional excitability, and (iii) macroscopic network synchronization constrained by the brain's structural connectome. By explicitly linking these multiscale mechanisms, we propose the framework as a theoretical perspective and operational scaffold for formulating testable hypotheses about state-dependent, closed-loop neuromodulation. Rather than claiming empirical validation, it aims to organize operational criteria for future studies of how stimulation timing may be aligned with intrinsic brain dynamics.