Cognitive Neurodynamics最新文献

Sleep staging is a critical indicator for assessing sleep quality and sleep disorders. Although significant progress has been made in sleep staging research, the representation of prominent waveforms and the capture of dynamic transitions between sleep stages still pose challenges. To address these issues, we propose MCTSleepNet, an Sleep staging Network containing Multiscale waveform representation, Composite attention and Time dependency learning modules based on single-channel electroencephalography (EEG). Firstly, multiscale waveform representation is learned from EEG signals using a dual-scale convolutional neural network (CNN). Then, a Composite Attention module is employed to enhance signal feature representation by considering both local and global contextual dependencies, thereby more effectively capturing prominent waveform features. Finally, a Bidirectional Gated Recurrent Unit (Bi-GRU) is used to learn the time dependent feature between EEG signals, enabling MCTSleepNet to model dynamic transitions between different sleep stages. Furthermore, considering the data imbalance between different sleep stages, this paper introduces an adaptive cross-entropy polynomial loss function to adjust the weights of different classes, thereby enhancing the model's attention to minority classes. Evaluation results on the publicly available Sleep-EDF-20 and Sleep-EDF-78 datasets demonstrate that MCTSleepNet performs exceptionally well in the sleep staging task.

Numerous physiological observations have shown that the brain operates at the edge of a critical state between order and disorder. Meanwhile, brain structures at different scales, from cortical columns to the entire brain, are organized in a modular manner. However, whether modular brain networks represent the optimized structure shaped for criticality and in what ways, have not been fully answered. In this study, a modular network with dense intra-module links but sparse inter-module links is established, and the behavior of each neuron is governed by the Kinouchi-Copelli model. Moreover, randomized surrogate networks with identical degree distribution are introduced to illustrate the significance of modular structures for criticality. Results suggest that the modular network requires fewer synaptic resources and lower firing costs to achieve criticality. More importantly, smaller avalanches indicate that the modular structures can enhance network resilience, facilitating rapid recovery from perturbations. Furthermore, by testing the sensitivity of the network state to local excitatory-inhibitory fluctuations, it is found that the efficiency of excitatory and inhibitory regulation is closely related to the 2-level excitatory input density. In addition, inhibitory regulation targeting modules with larger maximum real eigenvalues can more effectively suppress hyperexcitatory activities to achieve balance. When local excitation is greatly enhanced, even if the modular network is adjusted to the critical state, the size-to-duration ratio of module-level avalanches can effectively capture abnormalities. The properties also manifest in clinical recordings from patients with temporal lobe epilepsy, which may provide a promising method for epileptogenic zone localization.

Detecting seizures automatically is crucial for diagnosing and treating epilepsy, substantially benefiting affected patients. Various deep learning models and methods have been developed to automatically extract features from electroencephalogram (EEG) data for detecting seizures, but may often fail to adequately capture the significant periodic and semi-periodic dynamics in EEG signals, thus incompletely representing the extracted features. To address this challenge, we here introduced a novel EEG feature learning framework named ContrLF. This framework combines a contrastive learning framework and the Floss method to improve EEG feature learning for epileptic seizure detection. In our methodology, initially, both strong and weak augmentation are applied to transform the original EEG data into two distinct yet correlated views. Then, Floss is employed to automatically detect and learn the primary periodic dynamics within the augmented EEG data, capturing meaningful periodic representations that are essential for understanding seizure patterns in EEG signals. In parallel, the augmented EEG data were sequentially processed through temporal and contextual contrasting modules, which are designed to learn robust feature representations of the EEG signals. Finally, a Support Vector Machine (SVM) classifier was used to evaluate the effectiveness of the EEG features extracted using our proposed framework. Experimental results generated using both scalp and intracranial electroencephalogram (iEEG) datasets revealed that the proposed framework achieves over 90% accuracy, sensitivity, and specificity in detecting seizures. The framework outperforms other state-of-the-art methods, demonstrating its superiority in both cross-patient and specific-patient seizure detection.

Although electroencephalography (EEG) offers significant advantages in terms of high temporal resolution and cost-effectiveness, its application is often constrained by limited spatial resolution. This limitation makes it challenging to accurately localize and characterize activity within specific target regions of the brain. To address this, we propose a computational model for brain-network analysis based on independent component analysis (ICA) and source-space clustering. First, repetitive ICA decomposition is performed on a trial-by-trial basis, followed by clustering to extract stable independent components and their corresponding spatial mapping vectors. Subsequently, standardized low-resolution brain electromagnetic tomography (sLORETA) is employed for source localization. The resulting source locations are then clustered across trials to define network nodes, which are utilized to construct a source-level brain network for the investigation of neural mechanisms. The efficacy of this algorithm was validated using two datasets: the international Brain-Computer Interface (BCI) competition dataset involving motor imagery, and a self-collected dataset recorded during the preparatory phase of pistol shooting. Analysis of the motor-imagery dataset demonstrated that the proposed method identified active brain regions consistent with those observed in previous functional magnetic resonance imaging (fMRI) studies. Regarding the pistol-shooting preparation dataset, the method revealed heightened activity in the frontal, occipital, and bilateral temporal lobes. Furthermore, the intensity of information interaction among multiple brain regions exhibited a significant correlation with shooting performance. These findings not only corroborate prior research but also uncover novel features regarding source-level functional connectivity. Consequently, this novel framework achieves precise source localization and network analysis using EEG, significantly enhancing spatial resolution and providing a more accurate elucidation of target brain activities and information-interaction mechanisms during motor tasks.

Supplementary information: The online version contains supplementary material available at 10.1007/s11571-025-10405-z.

This study investigated the electrophysiological correlates of anticipatory and reactive processing and behavior associated with a visuomotor discrimination response task of professional dancers to test the effect of dance practice on their cognitive functions. To control the physical activity practice effects, dancers were compared with non-dancers matched for physical activity level. Considering the intrinsic features of the training routine to which professional dancers are constantly exposed - characterized by high temporal anticipation, continuous spatial monitoring and complex sensorimotor integration - we hypothesized differences in attentional control mechanisms and anticipatory processes compared to physically active controls in a discrimination response task. Behavioral data showed that dancers were more accurate than controls, and they had comparable response times. This effect was paralleled by the analysis of event-related potential (ERP), showing dancers compared to controls larger cognitive preparation in the prefrontal cortex (PFC), indexed by the prefrontal negativity (pN) ERP component. This may indicate a more intense top-down attentional control of the upcoming task. Dancers also showed reduced early sensory processing (P1 component) and less intense stimulus-response mapping (pP2 component), suggesting more efficient reactive processing in early sensory processing and associative brain areas. In contrast, the pP1 component was enhanced in dancers, likely reflecting superior sensory-motor integration, a pivotal function in choreographic demands. No difference emerged in the P3, signaling a similar workload load for the two groups. The results outline a peculiar neurofunctional profile of professional dancers, relying on intense cognitive anticipatory control and optimized proactive processing, allowing them superior response precision in sensory-motor performance. Further studies are needed to fully understand the specific trajectories of brain plasticity found here associated with dance practice.

We introduce a brain-constrained neurocomputational model designed to simulate higher cognitive functions of the human brain, implemented using NEST, a widely used open-source simulator optimised for high-performance spiking neural network simulations. Previously implemented in the custom-built C-based Felix simulation library, transitioning the model to NEST enhances accessibility, reproducibility, and computational efficiency. At the cellular level, the model comprises spiking excitatory neurons and local inhibitory neurons, whereas at the network level, it replicates the structural and functional organisation of 12 cortical regions spanning frontal, temporal, and occipital cortices, along with their associated inter-area connectivity. Additionally, global inhibition mechanisms and neuronal noise are integrated. Learning in the model follows biologically plausible Hebbian plasticity principles, incorporating both long-term potentiation and long-term depression. To validate the NEST implementation, we replicated previous simulation findings obtained with the Felix-based model. The new implementation successfully reproduced the same topographical distribution of cell assemblies following associative learning of object and action words within action and perception systems, replicating a range of previous neuroimaging results. Although the NEST model produced larger cell assemblies than Felix, the overall topographical patterns remained similar, indicating preservation of fundamental network characteristics. Moreover, the transition to NEST significantly enhanced computational efficiency, reducing simulation runtime nearly sixfold compared to Felix. This improvement in computational speed is crucial for expanding the model to include additional cortical regions, such as extending to the right hemisphere, which necessitates increased computational resources.

Supplementary information: The online version contains supplementary material available at 10.1007/s11571-026-10415-5.

Multi-label depressive emotion classification remains challenging due to the co-occurrence of multiple mental health-related emotions, implicit linguistic expressions, long-range contextual dependencies, and the presence of both active and passive depressive signals. This paper presents a comprehensive study on multi-label depressive emotion detection using the DepressionEmo dataset, which contains textual instances annotated with eight clinically relevant depression emotions: Anger, Cognitive Dysfunction, Emptiness, Hopelessness, Loneliness, Sadness, Suicide Intent and Worthlessness. The objective is to develop an effective and computationally efficient multi-label classification framework capable of accurately identifying both explicit active and latent passive depressive emotions from text. To address this problem, a broad spectrum of transformer-based and hybrid architectures is evaluated, including BERT, RoBERTa, DistilBERT, T5, BART, and DeBERTa with BiLSTM integration, as well as seq2seq BART and seq2seq RoBERTa-BART models. The proposed DeBERTa-BiLSTM architecture integrates disentangled self-attention for rich contextual representation with a BiLSTM layer for sequential dependency learning and history-state fusion, enabling effective modeling of long-range depressive cues. Experimental results demonstrate that the proposed DeBERTa-BiLSTM model consistently outperforms baseline seq2seq BART, BERT, T5, GAN-BERT, and all other developed variants, achieving an F1-Micro score of 0.83 and an F1-Macro score of 0.80, along with the lowest Hamming Loss (0.15) and the highest Jaccard Index (0.71). The model further achieves micro-precision of 0.81 and micro-recall of 0.85 indicating robust detection of both frequent and minority emotion labels. Runtime analysis shows notable inference efficiency, reducing time per sample by 26.32% at batch size 4 and 21.39% at batch size 32 compared to seq2seq BART. Despite these advantages, the model remains computationally heavier than lightweight transformers, is influenced by the dataset's modest size, and requires further validation across broader mental health domains.

Epilepsy is a neurological disorder characterized by recurrent, unprovoked seizures. Altered connectivity within the default mode network (DMN) has been associated with epilepsy, highlighting its role in seizure propagation. In this study, we investigate the temporal patterns of DMN connectivity in epilepsy patients compared to healthy controls using data-driven models of dynamic functional connectivity (dFC). Specifically, we employ one Hidden Markov Model (HMM) and two Hidden Semi-Markov Models (HSMMs) with Gamma and Poisson sojourn distributions to capture latent brain state transitions, as well as hidden connectivity states and their temporal properties. Dynamic metrics (i.e., fractional occupancy, switching rate, and mean lifetime) were derived for each subject, revealing prolonged dwell times in low-connectivity states and reduced flexibility in state transitions, particularly in low-connectivity DMN states. HSMMs, especially the Gamma variant, demonstrated superior sensitivity in capturing these alterations compared to the standard HMM, highlighting the importance of flexible sojourn modeling in dynamic functional connectivity analysis. Additionally, group-specific transition patterns suggested disrupted temporal progression of DMN state transitions. Our findings highlight the potential of HSMMs in capturing alterations in functional brain states and provide new insights into the dynamic reorganization of the DMN in epilepsy.

Supplementary information: The online version contains supplementary material available at 10.1007/s11571-025-10382-3.

Detecting Psychogenic Nonepileptic Seizures (PNES) is vital because PNES mimics epileptic seizures but has psychological-not electrical-origins, leading to frequent misdiagnosis and ineffective treatment. Electroencephalography (EEG) provides a non-invasive view of brain activity for distinguishing PNES from true epilepsy. Current PNES detection methods remain limited. This study introduces a curated PNES EEG dataset and a novel explainable feature-engineering (XFE) model. Expert neurologists annotated three classes: Normal, PNES with Verbal Suggestion Provocation (VSP+), and PNES without VSP (VSP -). The introduced explainable feature engineering (XFE) framework includes four components: (i) Distance Counter Pattern (DCPat) for channel-pair feature extraction (190 features for 20 channels), (ii) Cumulative Weight-based Neighborhood Component Analysis (CWNCA) for feature selection (threshold = 0.99), (iii) t-algorithm k-Nearest Neighbors (tkNN) ensemble classifier with Iterative Majority Voting (IMV) and greedy optimization, and (iv) Directed Lobish (DLob) for symbolic interpretation and cortical connectome mapping. For this research, we curated an EEG dataset and four cases are created using the curated dataset. These four cases are: Case 1 (Normal vs. PNES VSP+), Case 2 (Normal vs. PNES VSP-), Case 3 (PNES VSP + vs. PNES VSP-), and Case 4 (all three classes).). The introduced DCPat XFE framework reached accuracy above 96.5% in all four cases; Case 2 attained the best overall value (99.11%). DLob strings and connectome diagrams provided clear symbolic explanations of PNES-related patterns. The DCPat-based XFE framework yields high accuracy and interpretable outputs for PNES detection on EEG. These results support its use as a reliable, explainable tool for clinical decision support.

EEG microstates sequence analysis gained a lot of attention in recent years and different sequence analysis methods have been applied to study microstates sequence randomness, complexity, speed, periodicity, and long-range memory. Although several studies have reported on the reliability of temporal parameters, the stability of sequence-based metrics within subjects has not yet been systematically examined. In this study, we analysed EEG recordings from 60 healthy young adults and assessed short-term (90 min) and long-term (30 days) test-retest reliability and agreement of sequence measures: long-range memory (Hurst exponent), complexity (two Lempel-Ziv algorithms), and randomness (entropy and entropy rate). Across metrics, short-term reliability was consistently good to excellent (ICC = 0.831-0.902), and long-term reliability was moderate to good (ICC = 0.651-0.793). Entropy and entropy rate emerged as the most stable measures across both intervals, confirmed by minimal bias and strong agreement. These findings demonstrate that EEG microstate sequence dynamics represent a stable trait of neural activity, providing a solid methodological foundation for future studies that aim to embed these metrics into computational models and explore their translational value as neurophysiological biomarkers.

Supplementary information: The online version contains supplementary material available at 10.1007/s11571-025-10391-2.