Cognitive Neurodynamics最新文献

Vertigo, a prevalent neurovestibular disorder, arises from dysfunction in the vestibular system and often lacks precise, personalized treatments. This study proposes a bio-inspired spiking neural network (SNN) model that simulates vestibular dysfunction and adaptive recovery using Leaky Integrate-and-Fire (LIF) neurons with spike-timing-dependent plasticity (STDP). The architecture mimics the vestibular pathway through biologically plausible layers: hair cells, afferents, and cerebellar integrators, and models pathological states such as hair cell hypofunction and synaptic disruption. A reinforcement-based feedback mechanism enables the simulation of therapy-induced plasticity, resulting in a 48-62% drop and 38% recovery in cerebellar spike activity during adaptation epochs. The model demonstrates real-time feasibility, with an average simulation runtime of  4 s per epoch on standard hardware. Its design is scalable and well-suited for future deployment on neuromorphic platforms (e.g., Loihi, SpiNNaker). Its modular and interpretable design enables in silico testing of rehabilitation strategies, real-time monitoring of dysfunction, and future personalization using clinical datasets. This work establishes a computational foundation for AI-driven vestibular therapy that is adaptive, explainable, and hardware compatible.

Supplementary information: The online version contains supplementary material available at 10.1007/s11571-025-10368-1.

Accurate localization of the seizure onset zone (SOZ) is critical for successful surgery in drug-resistant epilepsy (DRE). To investigate the alterations of network characteristics between the SOZ and non-seizure onset zones (NSOZ) across different seizure stages, the intracranial electroencephalogram (iEEG) data based brain networks from 29 DRE patients have been constructed using the weighted phase lag index (WPLI) and phase transfer entropy (PTE), respectively. Then, graph theory metrics, such as eigenvector centrality, betweenness centrality, in-degree and out-degree, are calculated to compare network characteristics of SOZ and NSOZ nodes across interictal, pre-ictal, early-ictal and post-ictal periods in multiple frequency bands. Statistical analyses demonstrate that the SOZ exhibits significantly higher eigenvector centrality and betweenness centrality in the beta and gamma frequency bands, serving as network hubs and primary sources of information outflow. By contrast, the NSOZ shows elevated centrality only in the theta and alpha frequency bands during non-ictal states. Moreover, during the pre-ictal to early-ictal transition, the SOZ progressively evolves into hub nodes with enhanced outflow and reduced inflow, whereas the NSOZ shifts toward non-hub status with increased inflow. Importantly, the random forest model utilizing out-degree features of early-ictal gamma frequency band can effectively identify the SOZ, and achieve an area under the curve (AUC) of 0.82. Overall, these findings offer a novel network-based perspective on the state-dependent alterations of epileptic seizures in DRE and contribute to the treatment of epilepsy.

Supplementary information: The online version contains supplementary material available at 10.1007/s11571-025-10400-4.

Acupuncture modulates cognitive functions through acupoint stimulation and demonstrates significant regulatory effects on brain disorders. However, the underlying neurodynamic mechanisms of acupuncture remain unclear due to a lack of effective measures of brain activity. In this study, we developed an acupuncture-related potential (ARP) method based on Electroencephalogram (EEG) to elucidate the dynamic representation mechanisms underlying acupuncture stimulation. By analyzing ARP signal features and functional networks to capture stimulus-evoked brain activity, we derived spatiotemporal representations of neural manifolds and located across whole brain regions. It is exhibited that acupuncture induced significant four-phase event-related potentials (ERPs) waveforms predominantly in the parietal, frontal, central, and temporal lobes, with the parietal lobe exhibiting the highest amplitude at the P1 component (first positive peak). Latency gradients confirmed that the cortical neural activity originated in the parietal lobe and propagated through the central region to the frontal and temporal lobes. Dynamic network analysis revealed phase-specific reorganization: local frontal propagation (P1 component), global integration (P2 component), and novel topological pattern formation (P3 component). Neural manifold analysis uncovered a low-dimensional, ring-shaped representation encompassing the frontal, parietal, central, and temporal lobes. Acupuncture modulates brain function by activating key parietal lobe nodes, triggering distance-attenuated inter-regional signal transmission that dynamically reorganizes functional networks for multi-regional collaboration. The neural manifold representation revealed perception and integration of mechanisms of acupuncture information in the human brain. This ARP method provided a novel framework for investigating acupuncture-modulated spatiotemporal brain dynamics while enabling quantitative evaluation of its therapeutic effects.

Supplementary information: The online version contains supplementary material available at 10.1007/s11571-025-10408-w.

Parkinson's disease (PD) is among the two most prevalent neurodegenerative disorders (NDDs), affecting about 2-3% of individuals aged 65 and older. This NDD exhibits characteristic motor symptoms and several other non-motor features. Vocal deficits have been identified as one of the earliest quantifiable indicators of PD, which makes speech evaluation a viable, painless diagnostic instrument. We aim to apply machine learning (ML) models to vocal biomarkers for the early detection of PD, and use explainable artificial intelligence (XAI) techniques to interpret the predictions. The dataset is from Kaggle, a publicly reputable database, containing 1000 Parkinson's samples and 24 acoustic variables. We performed feature selection to identify the crucial vocal biological markers. Multiple machine learning (ML) models: Adaptive Boosting (AdaBoost), Random Forest (RF), Support Vector Machine (SVM), Multilayer Perceptron (MLP), Gaussian Naïve Bayes (GNB), Extreme Gradient Boosting (XGB), LightGBM (LGBM), CatBoost, Gradient Boosting (GB), Histogram-Based Gradient Boosting (HGB), and K-Nearest Neighbors (KNN) were employed. We also used SHAP (Shapley Additive exPlanations), LIME (Local Interpretable Model-agnostic Explanations), and Partial Dependence Plot (PDP) to explain the model performances. The HGB model ranked highest (1.00) based on accuracy, precision, recall, and F1-score, respectively. Also, the Confidence intervals (CI) (1.00,1.00) and p-value of < 0.001 of HGB were computed. XAI showed that jitter and shimmer-based biomarkers were the strongest contributors to the prediction of PD. In this study, the results showed that vocal base biomarker screening is not only economical but also an accessible diagnostic tool. In subsequent studies, we hope to include more varied datasets to improve both model and therapeutic relevance.

Graphical abstract:

Brain-controlled technology (BCT), centered on brain-computer interfaces (BCI), acquires and decodes neural signals to convert subjective intentions into control commands for external devices, establishing an intention output loop. In contrast, neuromodulation technology applies external physical stimuli to the central nervous system to regulate neuronal excitability and brain network states, achieving energy input for functional modulation and therapeutic purposes. The inherent differences in mechanisms and application goals determine that the ethical risk profiles and governance priorities of these two technologies cannot be conflated. Current public communication is characterized by terminology misuse and concept generalization, notably the misinterpretation of neuromodulation as controlling the brain. In response to the resulting ethical anxiety caused by capability extrapolation, this paper first clarifies the functional positioning of both technologies. Subsequently, a three-dimensional assessment model based on reality, reversibility, and technological dependence is constructed to map a stratified ethical risk landscape. The analysis reveals a significant asymmetry in risk distribution: risks of BCT are primarily concentrated on neural privacy leakage and responsibility attribution dilemmas within the intention decoding process, whereas risks of neuromodulation are deeply embedded in the potential erosion of personal identity and subject autonomy induced by external stimuli. To address institutional gaps in the current regulatory system regarding consumer-grade devices and long-term effects, this paper proposes a differentiated tiered governance strategy. It advocates establishing terminology demystification and conceptual rectification as the frontline defense for risk governance. On this basis, the strategy enforces physical defense mechanisms such as hardware fusing and parameter safety windows on the technical side, and strengthens data desensitization and algorithmic accountability on the data side. Ultimately, a multi-subject synergistic governance mechanism covering the full lifecycle from research and development and clinical trials to social application is constructed to provide institutional support for responsible innovation in neurotechnology.

Modern AI systems excel at pattern recognition and task execution, but they often fall short of replicating the layered, self-referential structure of human thought that unfolds over time. In this paper, we present a mathematically grounded and conceptually simple framework based on smoothed step functions-sigmoid approximations of Heaviside functions-to model the recursive development of mental activity. Each cognitive layer becomes active at a specific temporal threshold, with the abruptness or gradualness of activation governed by an impressiveness parameter [Formula: see text], which we interpret as a measure of emotional salience or situational impact. Small values of [Formula: see text] represent intense or traumatic experiences, producing sharp and impulsive responses, while large values correspond to persistent background stress, yielding slow but sustained cognitive activation. We formulate the recursive dynamics of these cognitive layers and demonstrate how they give rise to layered cognition, time-based attention, and adaptive memory reinforcement. Unlike conventional memory models, our approach captures thoughts and recall events through a recursive, impressiveness-sensitive pathway, leading to context-dependent memory traces. This recursive structure offers a new perspective on how awareness and memory evolve over time, and provides a promising foundation for designing artificial systems capable of simulating recursive, temporally grounded consciousness.

In biological neurons, synapses receive external stimuli to induce firing patterns. While the rapid generation of synapses regulates neural activity. In this paper, we use a magnetic-flux controlled memristor (MFCM) as a synapse to connect two functional neurons, establish the new coupled neurons, and study the synchronization characteristics. Firstly, we connect two neurons using memristive synapses, and derive the equations of the coupled neurons based on Kirchhoff's voltage law. Furthermore, we calculate the energy of the memristive coupling channels, and obtain the energy difference between the coupled neurons. Secondly, we propose a criterion for exponential growth controlled by energy difference. By setting higher coupling channel strength to establish synaptic connections, energy pumping can be effectively activated. Finally, for three modes, we analyze the energy evolution under the variations of memristive synapses, and find that the coupling channels are adaptively controlled by energy difference. The results show that when the coupling strength through synapses is enhanced, identical neurons can achieve complete synchronization, and different neurons can achieve phase locking. This study clarifies the underlying mechanisms of regulating coupled neurons via memristive synapses and explores how neurons achieve potential energy balance from the perspective of physical fields.

This paper designs two improved passive cosine-type ideal memristors and incorporates them into the Hopfield neural network, thereby proposing a novel cosine-type memristor-driven Hopfield neural network (CMDHNN). The model exhibits a planar equilibrium set and demonstrates extreme multistability, characterized by the coexistence of infinitely many attractors. The boundedness of the system is rigorously proven using the Lyapunov method. Nonlinear dynamics analysis tools, including bifurcation diagrams, Lyapunov exponent spectra, phase portraits, and time series plots, are employed to thoroughly investigate the model's complex chaotic dynamics. Leveraging the chaotic system of the proposed CMDHNN, an image encryption scheme is developed, in which chaotic sequences are utilized to generate diffusion and permutation key streams for encrypting the plaintext image. The results indicate that the encryption scheme based on this model exhibits excellent robustness and can effectively resist various common attacks.

Integrating structural magnetic resonance imaging (sMRI) with deep learning techniques is one of the important research directions for automated diagnosis of Alzheimer's disease (AD). Among these, Convolutional Neural Networks (CNNs) have been widely adopted as a mainstream approach due to their powerful feature extraction capabilities. However, existing convolutional neural network (CNN)-based voxel models with excellent performance are typically constrained to a single spatial scale. This limitation hinders the effective capture of complex, distributed brain atrophy features of AD and often results in insufficient model interpretability. To address these limitations, we propose BMSSnet, an interpretable AD recognition model based on a multi-scale multi-block attention mechanism. This model adopts a CNN-Transformer hybrid architecture. Specifically, it first captures local anatomical details using a 3D feature extraction network. Subsequently, it utilizes a dual-branch multi-scale attention mechanism to model patches of different sizes, enabling the Transformer to extract global long-range dependencies. Additionally, we devise a lightweight spatial gating unit to facilitate feature spatial interaction while maintaining computational efficiency. For interpretability, the model localizes decision-critical three-dimensional regions of interest (3D ROIs) using attention weights and aligns them with anatomical atlases to verify their pathological relevance. Finally, extensive experiments on the ADNI dataset demonstrate that BMSSnet not only achieves superior diagnostic performance but also accurately localizes AD-associated salient brain regions, offering reliable clinical interpretability.

Emotion recognition from electroencephalogram (EEG) signals remains a challenging problem due to the high dimensionality, nonlinearity, and complex spectral dependencies inherent in neural activity. Conventional deep learning approaches often treat EEG features independently, thereby limiting their ability to capture structured spectral relationships. In this work, we propose a graph-based representation learning framework that models frequency-domain EEG features as nodes within a structured graph and leverages a Graph Neural Network-Variational Autoencoder (GNN-VAE) to learn compact latent representations. Spectral adjacency is defined using k-ring neighborhood connectivity, enabling localized message passing across contiguous frequency bands. The learned latent embeddings are subsequently classified using recurrent and attention-based temporal models to capture sequential dependencies across spectral segments. Experiments conducted on an EEG emotion dataset comprising three affective states demonstrate that the proposed approach consistently outperforms traditional machine learning baselines and non-graph deep learning models, achieving an accuracy of [Formula: see text] 91% and F1-score of 0.903. Ablation analyses further confirm the contribution of graph-based encoding and variational regularization to improved generalization. While the current study focuses on fixed spectral connectivity and subject-dependent evaluation, the results highlight the potential of graph-structured latent modeling for EEG-based emotion recognition and provide a foundation for future extensions incorporating adaptive graph learning and explainable representations.