OBJECTIVE:Speech Imagery (SI) has emerged as a promising paradigm for Brain-Computer Interface (BCI) control, attracting growing interest due to its intuitive nature--allowing users to interact with the system by internally saying a command. In this study, we investigate the replicability and reproducibility of SI decoding methods. These two aspects are critical in BCI research, where prior literature has highlighted that many studies suffer from incomplete methodological reporting or flawed evaluation procedures, making reproduction difficult. The inherent variability of brain signals further complicates the replication of results. Evaluating the reproducibility of SI decoding approaches is therefore essential to assess the true feasibility of SI as a viable BCI paradigm. APPROACH:To assess reproducibility, we selected two of the most widely used open-access SI datasets and attempted to reproduce four published decoding pipelines for each dataset. We followed each implementation step-by-step, documented missing or ambiguous information, detailed how we addressed it, and compared our decoding results to those originally reported. To assess replicability, we applied standard decoding pipelines across different timefrequency configurations to three open SI datasets and our own collected dataset. For context and validation, we conducted the same procedure on four publicly available and widely used motor imagery (MI) datasets. MAIN RESULTS:All evaluated SI studies contained some form of missing methodological detail, and others did not include cross-validation procedures. Our reproduction attempts consistently yielded lower classification accuracies than originally reported, with discrepancies ranging from 2 to 39% (x = 11.25 ± 12.41%). In the replication analysis, we found no consistent time-frequency patterns across SI datasets. Furthermore, only 36% of SI participants achieved classification accuracies above statistical significance thresholds, compared to 91% of the participants in MI datasets. Significance . This is the first comprehensive assessment of both reproducibility and replicability in SI decoding. Our findings raise important concerns about the reliability of current SI research and suggest that the feasibility of SI as a practical BCI paradigm may have been overestimated.
OBJECTIVE:Postoperative delirium (POD) is a common perioperative complication involving central nervous system dysfunction, particularly among critically ill and elderly patients, yet its rapid and objective detection remains challenging in clinical settings. To develop and evaluate a lightweight, interpretable electroencephalography (EEG)-based deep learning framework for POD detection that remains accurate under sparse-electrode acquisition and practical for bedside or portable deployment.
Approach: We propose Helix Fusion HarmonyNet (HFHN), a transform-domain architecture that explicitly separates EEG representations into amplitude and phase pathways. HFHN integrates multi-head self-attention, convolutional local feature extraction, learnable sinusoidal positional encoding, and controlled cross-pathway fusion to jointly model spectral intensity, temporal synchrony, local patterns, and long-range dependencies. The framework was evaluated against representative CNN-, Transformer-, and EEG-specific deep learning models across signal domains, fusion strategies, electrode configurations, ablation settings, and deployment-oriented efficiency tests.
Main results: In Fourier-domain POD detection, HFHN achieved 100% accuracy, precision, specificity, F1-score, and sensitivity. Ablation analysis confirmed the importance of amplitude-phase interaction, as removing cross-pathway fusion reduced accuracy from 100% to 95.38%. Under sparse-channel conditions, HFHN maintained 99.42% accuracy and a 99.33% F1-score with 8 electrodes, and 94.08% accuracy and a 93.26% F1-score with only 4 high-attention electrodes. Reducing the number of electrodes from 60 to 4 decreased FLOPs from 16.95 G to 0.99 G, while inference latency remained below 3.09 ms.
Significance: HFHN provides an accurate, interpretable, and computationally efficient solution for EEG-based POD detection. Its robustness under sparse-channel acquisition and low-latency performance support potential integration into portable or bedside EEG systems for rapid postoperative monitoring in wards and intensive care units.
Objective.Dual-task standing challenges postural regulation in older adults, but the associated cortical network reorganizations remain incompletely characterized. This study investigated how a visuomotor dual-task modulates postural dynamics and cortical network during unstable standing.Approach.Twenty-six community-dwelling older adults performed foam-surface standing under single-task and visuomotor dual-task conditions. Postural dynamics were quantified using center-of-pressure (COP) measures and stabilogram diffusion analysis (SDA). Cortical network organization was assessed using electroencephalography (EEG) functional connectivity (phase-lag index) and minimum spanning tree (MST) topology. To improve characterization of distributed cortical adaptation, a novel fused MST framework integrating theta, alpha, and beta-band connectivity was evaluated together with conventional band-specific MST.Main results.Dual-task standing significantly increased COP displacement and velocity in both anterior-posterior and medial-lateral directions, accompanied by altered SDA parameters across directional domains, indicating substantial changes in postural control dynamics. EEG analyses revealed task-related cortical network reorganization, with increased theta and alpha band overlap ratios and significant topology changes in the fused MST representation. Compared with band-specific MST representations, the fused MST framework provided an integrated characterization of task-related cortical network adaptation across frequency bands.Significance.Visuomotor dual-task standing induces substantial reorganization of both postural dynamics and cortical network topology in older adults under unstable conditions. These findings suggest that the fused MST framework may provide a useful quantitative approach for assessing cortical adaptation to dual-task balance challenges in older adults.
OBJECTIVE:The fidelity of neural representations learned by large EEG foundation models depends on how raw brain signals are tokenized. Existing methods suffer from arbitrary temporal boundaries misaligned with neural state transitions, neglecting inter-channel spatial information, and fixed segmentation criteria that fail to generalize across heterogeneous EEG paradigms. APPROACH:This study proposes the SE-DAGAF Adaptive Tokenizer (SEDAT), a hybrid framework integrating squeeze and- excitation (SE)-based spatial aggregation, data-adaptive Gaussian average filtering (DAGAF)- based signal decomposition, instantaneous-frequency-guided adaptive segmentation, and Fourier domain resampling into a single computationally efficient pipeline. SEDAT is evaluated across 10 heterogeneous EEG datasets spanning motor imagery, mental imagery, P300, slow cortical potentials, sleep staging, and epilepsy paradigms, using four large foundation models: LaBraM, EEGFormer, EEGPT, and NeuroGPT. It is benchmarked against five competitive baselines: fixed length windowing, CTXSEG, LiPCoT, TFM-Tokenizer, and SiS. MAIN RESULTS:SEDAT achieves classification improvements of up to 15.3% over fixed-length windowing and 1.2-4.6% over the second-best method, with all comparisons reaching p < 0.001 after Benjamini-Hochberg correction. Token quality analysis confirms substantially improved feature separability, with Silhouette scores of 0.81-0.85 versus 0.33-0.48 for rigid baselines. SIGNIFICANCE:With O(CN + KN logN) complexity, SEDAT explores new applications for SE and DAGAF as tokenizers and provides a physiologically grounded and computationally practical tokenization solution for large-scale EEG foundation models.
OBJECTIVE:Advances in neural recording technology enable simultaneous measurements across multiple brain regions, providing new opportunities to study inter-regional interactions. However, several challenges arise when developing nonlinear dynamical models of cross-region interactions. In particular, we identify three key desired properties for a nonlinear framework. First, it should prioritize extraction of cross-region dynamics to avoid confounds from within-region dynamics. Second, it should enable localization of nonlinear structure within the model to better interpret cross-regional interaction models. Third, it should support both causal and non-causal inference of shared dynamics and do so using source region activity alone. Current cross-regional models do not satisfy all these properties. APPROACH:Here, we address these challenges by developing cross-population prioritized dynamical nonlinear interaction model (CroP-DYNO), a nonlinear dynamical framework that prioritizes the learning of shared cross-population dynamics to avoid confounds from within-population activity. Further, CroP-DYNO allows individual components of the dynamical model to be independently specified as nonlinear or linear. Finally, it supports both causal and non-causal inference of latent states, using only source region activity. RESULTS:We validate our method on datasets across species and distinct brain regions. We find that both the prioritized learning and the nonlinear modeling in CroP-DYNO are important for accurately extracting cross-population dynamics. As such, CroP-DYNO outperforms baseline nonprioritized and linear prioritized models in predicting target neural population activity from source activity. Further, CroP-DYNO enables systematic localization of nonlinear structure and quantifies dominant interaction pathways between brain regions, with interaction strengths that align with known circuit anatomy. SIGNIFICANCE:Overall, these results establish CroP-DYNO as a flexible nonlinear framework for studying interactions between neural populations and brain regions, enabling both accurate modeling and interpretable dissection of nonlinear structure in cross-regional communication models.
OBJECTIVE:Handwriting imagery (HI) based on electroencephalography (EEG) offers a non-invasive route to text input and complex intention expression for brain-computer interfaces (BCIs). However, HI EEG decoding is challenged by low signal-to-noise ratios, non-stationarity, cross-session distribution shifts, and the coexistence of continuous temporal trends and local high-response patterns. APPROACH:We propose a Dual-View Representation Decoupling Network (DRDNet) for within-subject crosssession HI EEG classification. DRDNet first constructs two complementary temporal views from spatial EEG features using average and max pooling, corresponding to smooth trend-oriented and salient response-oriented representations. These views are then modeled by a bidirectional Mamba encoder and a Transformer encoder, respectively, and adaptively integrated through a time-step-level dynamic fusion mechanism followed by long short-term memory based temporal aggregation. The method is evaluated on two tasks from a public HI EEG dataset: Chinese character stroke handwriting imagery (CCSHI) and pinyin single-vowel handwriting imagery (SVHI). MAIN RESULTS:DRDNet achieved average accuracies of 67.74% and 62.51%, with Cohen's kappa scores of 0.5968 and 0.5502, on CCSHI and SVHI, respectively. It outperformed seven representative EEG decoding baselines under the same crosssession protocol. Confusion matrices, feature visualization, ablation studies, structural variant analysis, and complexity evaluation further showed improved feature separability and a favorable balance between decoding performance and computational efficiency. SIGNIFICANCE:The results indicate that decoupling HI EEG into complementary temporal views and matching them with heterogeneous temporal encoders provides an effective representation learning strategy for robust non-invasive handwriting BCI decoding.
Objective.Seizures are often characterized by epileptiform discharges that appear to be highly synchronous across electroencephalogram (EEG) channels, yet closer analysis reveals small time delays indicative of fast traveling waves. Microelectrode studies suggest that a possible source of these waves is the ictal wavefront, a slowly advancing boundary emitting opposing waves; attempts to extend this analysis to macroelectrodes produced conflicting theories, suggesting both static and moving radial sources. Because variability in EEG measurement and analysis may contribute to conflicting results, we systematically evaluated how signal characteristics and methodological choices affect wave direction estimates. We applied the resulting framework to human intracranial EEG (iEEG) to validate the results.Approach.We simulated iEEG data using a second-order autoregressive model, modeling coherence as a distance-weighted sum of signals. We then applied one of eight different time-delay propagation patterns at a range of wave speeds, and accuracy was assessed for multiple referencing schemes, sampling rates, and spatial resolutions. Human iEEG recordings were then analyzed to validate methodological recommendations and characterize ictal waves.Main Results.Simulations revealed that wave patterns could be measured most accurately for wave speeds <1000 mm s-1and average iEEG coherence levels >0.4. Under these conditions, a corner electrode reference outperformed other referencing schemes. Wave estimation accuracy was highest for high sampling rates and low electrode spacing. Accurate results were obtained at 3 mm electrode spacing, but not at 9 mm spacing, which approximates standard clinical subdural grids (p< 0.01). Human iEEG exhibited a wide range of propagation patterns, including spirals, sources, and sinks which have not been previously reported.Significance.Detailed analyses of simulated and human iEEG data highlight critical methodological choices that must be considered when characterizing complex seizure wave patterns. Our findings provide a validated framework to increase the rigor of future studies.
Objective.Deep learning has shown significant potential in electroencephalogram (EEG)-based seizure prediction. However, translating these advances into practical applications faces critical challenges: the prohibitive time costs of reviewing accumulated historical data during frequent model updates, the need for rapid adaptation to diverse model architectures across hardware platforms, and the substantial burden of massive data storage. To this end, we introduce the first dataset distillation study tailored for seizure prediction, which condenses the original dataset into a compact, information-dense synthetic dataset. This tiny-scale proxy allows models to match the performance of those trained on the full-scale dataset.Approach.Methodologically, we formulate the distillation as a min-max optimization task, introducing an EEG-aware alignment objective that jointly preserves latent feature distributions, temporal dynamics in the time-frequency domain, and inter-channel correlation structures. Through signal-compatible amplitude and phase decomposition, we optimize the frequency parameter to maximize the CF discrepancy via a sampling network, while simultaneously tuning the synthetic data to minimize this gap. Additionally, we integrate hard labels to refine soft labels to enhance the generalization of synthetic data.Main results.Extensive validation across 37 public and 35 clinical subjects demonstrate the method's robustness across diverse scenarios and model architectures. With merely 1% of the original data volume, our approach retains roughly 95% of the original performance while reducing training duration by approximately 98%.Significance.This method addresses clinical translation bottlenecks by distilling massive EEG data into compact synthetic sets, offering a promising solution for efficient model maintenance, flexible model switching, and optimized storage.
OBJECTIVE:Characterizing the relationship between structural connectivity (SC) and functional connectivity (FC) is a central problem in multimodal neuroimaging. SC-to-FC prediction accuracy is often used to summarize this relationship, but a high score may largely reflect population-level regularities shared across subjects and does not reveal whether regional, network-level, or subject-specific correspondence is preserved. We therefore developed a framework that makes these different levels of SC-FC correspondence explicit and directly testable.
Approach. For each subject, we constructed separate spectral representations of SC and FC and estimated an explicit linear map between them. Signal transfer was evaluated against matched controls that disrupted node identity, within-network organization, or subject pairing, allowing aggregate transfer to be distinguished from more specific forms of correspondence. A training-derived reference map was further used to define the functional component expected under a reference SC-FC relationship and the deviation from that expectation for downstream aging and disease-related analyses.
Main results. Across four independent datasets, all primary matched-control gaps were positive after false-discovery-rate correction, although their magnitudes varied across cohorts. Strong aggregate transfer did not necessarily imply strong subject-level correspondence. Compared with external baselines, the proposed framework more consistently preserved regional identity, within-network organization, and correct subject pairing. In aging analyses, reduced alignment with a young-reference SC-FC relationship was robust in CamCAN but not significant in NKI, whereas empirical FC magnitude showed no comparable decline. In ADNI, reference-conditioned markers showed exploratory disease-related signals, but the associated uncertainty did not support definitive diagnostic or incremental-value claims.
Significance. The proposed framework separates the ability to transfer SC-derived signals into FC space from the specificity of the correspondence being preserved. It provides an interpretable basis for testing population-level, network-level, and subject-sensitive SC-FC relationships and for constructing reference-conditioned individual deviation measures without treating prediction accuracy as direct evidence of biological constraint.
Objective. The imbalance in interhemispheric functional connectivity following stroke fundamentally impedes motor recovery. Although transcranial direct current stimulation (tDCS) effectively modulates neuroplasticity, the acute effects of distinct stimulation montages on lateralized brain networks and how these network shifts drive behavioral improvements warrant further investigation.Approach. We employed a single-blind, randomized crossover design involving 28 patients with chronic subcortical stroke. Participants received anodal, cathodal, bilateral, or sham tDCS targeting the primary motor cortex. Resting-state electroencephalography data were acquired immediately before and after each intervention. To quantify network lateralization, we computed dynamic functional connectivity (dFC) using mutual information and constructed an asymmetry index matrix. Key connectivity features were isolated via robust feature selection algorithms, classifying stimulation states and evaluating correlations with acute motor gains on the Jebsen-Taylor Hand Function Test (JTT).Main results. Network dynamics features successfully discriminated pre- versus post-intervention states with high accuracy (AUC: 0.87-0.98). All three active tDCS protocols exhibited montage-specific directional modulation. Crucially, anodal tDCS uniquely reversed contralesional network dominance, facilitating ipsilesional connectivity and shifting the interhemispheric balance toward the affected hemisphere. Conversely, cathodal and bilateral tDCS predominantly induced contralesional network inhibition, whereas sham stimulation exhibited no directional bias. Furthermore, the functional inhibition of the contralesional network (specifically at the P2-P6 connection) induced by bilateral tDCS significantly correlated with acute JTT improvements (r= 0.55,p= 0.04).Significance. This study demonstrates that distinct tDCS montages modulate stroke-induced network imbalances through dissociable lateralization mechanisms. While anodal tDCS drives ipsilesional facilitation to reverse contralesional dominance, bilateral and cathodal montages operate primarily via contralesional inhibition. These findings provide critical mechanistic insights into interhemispheric network dynamics.
Objective.Electrical stimulation and neural recording underpin neural prostheses for restoring function and treating neurological disorders, but clinical adoption is limited by the invasiveness of implantation. The Endovascular Neural Interface offers an alternative by accessing intracranial targets through the cerebral vasculature. This work presents the first strength-duration characterization of cortical evoked potentials, elicited by endovascular stimulation, adjacent to the cerebellum.Approach.A polymer-based stent-electrode array was deployed into the left transverse sinus of an ovine model. Biphasic current pulses targeting the cerebellum were delivered via the stent electrodes. Cortical responses were recorded using a subdural electrocorticography grid.Main results.Endovascular stimulation consistently evoked time-locked cortical potentials with early and late components at approximately 40 ms and 100 ms post-stimulation. Electrode functionality and stability were confirmed through impedance monitoring throughout the experiments. Strength-duration analysis revealed rheobase and chronaxie values, providing a quantitative basis for parameter selection and comparison with established intracranial stimulation modalities.Significance.These results demonstrate that endovascular electrodes may access non-superficial brain structures and evoke reproducible cortical responses without open neurosurgery. This work helps establish a foundational framework for endovascular neuromodulation and supports further investigation of its potential for future closed-loop and network-level neuromodulation research.
Objective. Central vertigo (CV) lacks objective electrophysiological measures for severity assessment and rehabilitation monitoring. We aimed to characterize multiscale resting-state EEG alterations and identify clinically interpretable candidate features in stroke-related CV.Approach. Resting-state EEG was analyzed in 50 patients with stroke-related CV (31 moderate, 19 severe) and 31 age-matched healthy controls. The framework integrated relative spectral power, cross-frequency coupling, PLV-based sensor-level phase synchrony, graph metrics, machine-learning feature ranking, and associations with balance confidence and dizziness severity.Main results. Severe CV showed widespread relative delta-power reductions of 28.7%-29.4% versus controls. Post hoc analysis showed lower global absolute delta power in severe CV than controls (Tukeyp= 0.0227; rank-based false-discovery-rate (FDR)q= 0.0459), although the absolute-power effect was less spatially extensive. Both patient groups showed reduced delta-theta and delta-beta amplitude-amplitude coupling (AAC), enhanced delta-alpha PPC, and theta-band increases in PLV-derived node degree, clustering, and global efficiency; local efficiency increased only in SV. Delta-beta AAC ranked highest across machine-learning methods and correlated moderately with balance confidence (ρ= 0.470,p= 0.001) and dizziness severity (ρ= - 0.472,p= 0.001). Zero-lag-robust measures showed the same theta ordering but were nonsignificant after FDR correction and did not establish volume-conduction-independent topology, supporting cautious PLV interpretation.Significance. Stroke-related CV involves coordinated alterations across oscillatory, cross-frequency, and sensor-level network measures. This interpretable framework identifies candidate EEG features for objective characterization that require external and longitudinal validation before clinical use.
Objective.Current clinical differentiation of Alzheimer's disease (AD) and frontotemporal dementia (FTD) suffers from a misdiagnosis rate exceeding 40% due to overlapping symptomatology, and existing Electroencephalography (EEG)-based tools inadequately integrate multidimensional features or lack adaptive optimization. We aimed to develop a framework combining periodic and aperiodic EEG features with adaptive optimization for high-precision differential diagnosis.Approach.We proposed a dual-branch neural network integrating multi-dimensional electroencephalogram (MDEEG) features with an improved Marine Predator Algorithm (IMPA). The MDEEG-IMPA model combines convolutional neural network-based power spectral density feature extraction with a fully connected branch for aperiodic parameters (1/foffset and exponent), enhanced by IMPA for adaptive feature weighting. The model was evaluated on resting-state EEG from 88 subjects (36 AD, 23 FTD, 29 healthy controls (HC)).Main results.MDEEG-IMPA achieved 99.20% accuracy in discriminating AD from FTD (Recall = 98.82%,F1-score = 99.02%), substantially outperforming comparative methods including STEADYNet (84.59%) and SVM (93.5%). The MDEEG-IMPA model demonstrated robust performance in discriminating between AD and HC (Accuracy = 98.11%) and between FTD and HC (Accuracy = 98.95%). Ablation studies confirmed that 96.3% of the improvement in performance originated from the synergy between features and algorithms.Significance.The MDEEG-IMPA framework provides a reliable, high-precision computer-assisted diagnostic solution for neurodegenerative diseases with overlapping clinical presentations. By integrating periodic and aperiodic EEG features within an optimized multi-branch architecture, this work demonstrates that multidimensional electrophysiological characterization combined with adaptive optimization can substantially improve differential diagnostic accuracy, offering significant potential for clinical translation in early and accurate identification of AD and FTD.
Objective.The widespread application of surface electromyography (sEMG) pattern recognition in artificial limbs largely relies on a robust decoding system that continuously and accurately identifies user hand movement intention. However, the muscle contraction variability significantly distorts the time-frequency components in sEMG signals, thereby weakening the performance of hand movement decoding. This paper aims to propose a novel framework to achieve robust decoding of hand movement for mitigating muscle contraction variability from sEMG signals.Approach.We collect two-channel sEMG signals from eight healthy subjects performing six hand movements under three muscle contraction levels (normal level, medium level, high level). Fifteen popular sEMG features are extracted, and five classic classifiers are employed for decoding hand movements. We analyze how muscle contraction variability degrades hand movement decoding performance and propose a simulation-driven feature mapping framework. It maps abnormal sEMG features from medium and high levels back to normal levels by Feature Denoising Network using simulated sEMG features.Main results.Without the proposed framework, decoding accuracy drops markedly when muscle contraction intensity is switched from normal level to medium and high levels, resulting in accuracy losses of at least 32.32% for observed muscle contraction levels and 37.50% for predicted muscle contraction levels. With the proposed framework, decoding accuracy is significantly improved, with accuracy improvements of at least 31.82% for observed muscle contraction levels and 26.99% for predicted muscle contraction levels.Significance.The paper provides a feasible solution to challenge of muscle contraction variability for robust hand movement decoding from sEMG signals, which holds potential reference value in the development of myoelectric hand prosthesis.
Objective.Static postural control is a dynamic feedback process in which the integration of visual and somatosensory inputs within the central nervous system determines the activation of lower kinetic chain muscles, resulting in postural oscillations, to maintain balance during quiet standing. This study explored the association between cortical and muscular activity in relation to postural responses.Approach.Electroencephalography (EEG) is commonly used to assess neural oscillations. Network graph theory (NGT) applied to EEG-derived connectivity outlines the functional connectome. Surface electromyography (sEMG), combined with stabilogram analysis, enables the characterization of muscle activation patterns during postural control. By simultaneously acquiring EEG, sEMG and stabilometric signals, EEG-derived NGT metrics, sEMG-derived time- and frequency-domain features and sway parameters were correlated in twenty-one healthy volunteers (11 males, 29.52 ± 3.70 years) performing static postural control tasks for 30 s, standing upright on a stable surface, with eyes-closed (EC) and eyes-open (EO). Partial Spearman's rank correlation analyses were used to assess multimodal associations between NGT metrics and electromyographic and biomechanical measures.Main Results.During EC sessions, changes in two postural sway measures (jerk, path length) were negatively correlated to changes in the resilience of the EEG connectome, as reconstructed from broadband EEG signals (P< 0.05, corrected for multiple comparisons). During both EC and EO sessions, changes in muscle activation, fatigue-related and postural sway measures were weakly correlated to changes in multiple topological parameters of the EEG connectome, as reconstructed from either broadband (EC) or delta (EO), theta (EO), alpha (EO), beta (EO) and gamma (EC) EEG signals.Significance.This study highlights possible frequency-dependent associations between the global topological organization of the EEG functional connectome and muscular and postural responses during balance maintenance. Albeit preliminary, these exploratory findings may provide a novel perspective for the multimodal investigation of cortico-muscular interactions subserving postural control in health and disease.
Objective.Traumatic spinal cord injury causes functional impairments in large part due to the limited ability of damaged axons to regenerate. During development, the extracellular matrix (ECM) plays a critical role in providing molecular cues that promote and direct axon growth. Biomaterials that mimic the physicochemical properties of the ECM may be promising for promoting axon growth in the injured spinal cord. Glycosaminoglycans (GAGs) in the ECM during development are known to direct axon growth and guidance. We previously reported that GAG-mimetic cellulose sulfate, which can be synthesized with varying degrees of sulfation, partially and fully sulfated cellulose (pCelS and fCelS, respectively), promoted neurite extension over native GAGs.Approach.The present study investigated the use of GAG-mimetic containing scaffolds as a platform for Schwann cell transplants to promote axon growth. The aligned fibrous scaffolds consisted of gelatin or gelatin blended with polycaprolactone (PCL), in order to improve hydrolytic stability, with 0.25 wt.% GAG-mimetics.Main results.Our results demonstrated that the fCelS-containing scaffolds with SCs promoted more axon growth than the pCelS-containing scaffolds and the greatest myelination over all other scaffolds. Also, in the presence of astrocytes, fCelS-containing scaffolds supported axon growth. Furthermore, the fCelS-containing scaffolds were capable of binding and retaining a greater amount of the neurotrophins brain-derived neurotrophic factor and neurotrophin-3 as compared to all other scaffolds.Significance.Aligned fibrous GAG-mimetic scaffolds containing fully sulfated cellulose may hold promise to repair damaged spinal cord tissue by providing physiochemical axon growth-promoting cues and a transplantation platform for SCs to support axon growth and myelination.
Objective.Anterior nucleus of the thalamus deep brain stimulation (ANT-DBS) is an effective therapeutic option for drug-resistant epilepsy (DRE); however, substantial inter-individual variability in treatment response limits its clinical optimization. This study aimed to develop an objective and explainable preoperative prediction model for ANT-DBS outcomes using nonlinear dynamical features derived from preoperative N2 sleep electroencephalography (EEG).Approach.Artifact-free N2 sleep EEG segments were retrospectively collected from 26 patients with DRE who underwent ANT-DBS and were classified as responders or non-responders according to postoperative seizure reduction. Nonlinear dynamical features, including conditional entropy, robust permutation entropy (RPE), and their multiscale variants, were extracted across six frequency bands (δ, θ, α, σ, β, and low-γ). Statistically significant features were identified using the Mann-WhitneyUtest with false discovery rate correction (q< 0.05). These features were subsequently integrated into a multidimensional feature space to construct a support vector machine (SVM) classifier. Model interpretability was further evaluated using the SHapley additive explanations (SHAP) framework.Main results.Non-responders exhibited significantly higher complexity in theδandσbands (q< 0.05), indicating impaired thalamocortical rhythmic fidelity and disrupted synchronization stability. In contrast, responders demonstrated significantly higherα-band RPE (q< 0.01), reflecting greater cortical functional flexibility and a richer dynamical repertoire. The SVM classifier achieved an area under the curve of 0.933 on an independent validation set, with 100% sensitivity for responder identification. SHAP-based attribution further revealed that the most influential features were strongly associated with physiologically meaningful alterations in thalamocortical network dynamics.Significance.These findings demonstrate that preoperative N2 sleep EEG complexity may serve as a non-invasive and interpretable biomarker for predicting ANT-DBS outcomes in DRE. This framework provides a clinically accessible decision-support tool for personalized patient selection and precision neuromodulation.
Objective.Emotion decoding is a growing field of electroencephalography (EEG) research with numerous applications in areas such as healthcare, particularly when coupled with mobile EEG. However, research into mobile emotion decoding has been somewhat limited, especially regarding emotions elicited by music and how data splitting affects performance. The present study focuses on music emotion decoding with mobile (around-the-ear) EEG and the impact of the data splitting technique used.Approach.We collected the decoding auditory attention and musical emotions with ear-EEG dataset (DAAMEE), featuring both scalp EEG data (DAAMEE-s), and around-the-ear cEEGrid data (DAAMEE-c). DAAMEE featured several tasks, in one of which subjects listened to music and recorded their emotional state. We used the emotion task data, as well as the scalp EEG data from the existing DEAP dataset, for performance tests with various data splitting techniques and deep learning models. The tests focused on binary valence decoding, but the best-performing models were also applied for arousal, dominance, and three-dimensional valence-arousal-dominance classification.Main results.Performance with DAAMEE-c was generally similar to performance with DAAMEE-s and DEAP, showing the viability of cEEGrid for emotion decoding. The results also demonstrate that using certain data splitting techniques may lead to performance inflation due to mechanisms such as temporal correlations between samples being exploited.Significance.While these results motivate further research on mobile emotion decoding, future studies (for emotion decoding, and EEG classification generally), should take care to choose a data splitting paradigm that avoids overfitting.
Objective.Future motor brain-computer interfaces (BCIs) are expected to benefit from integrating neural signals from multiple motor-related brain regions. While decoding studies have largely focused on lateral sensorimotor cortex, the medial wall of the cerebral hemisphere remains relatively underexplored. Here, we investigate the contribution of medial wall regions to finger movement decoding using human electrocorticography (ECoG) recordings.Approach.We analyzed ECoG data from four subjects performing finger movements. Single- and multi-channel decoding analyses were applied to medial wall electrodes, examining the contribution of time-domain and frequency-domain features, including local motor potentials (LMP) and oscillatory power in the(8-12 Hz) and(12-34 Hz) bands. Decoding performance was assessed for movement detection and finger discrimination.Main results.Significantly above-chance finger movement detection was observed across multiple medial wall subregions. LMP and-band power contributed most strongly to decoding performance. Feature dynamics shared key properties with primary motor cortex, including pre-movement-desynchronization, while also exhibiting region-specific patterns such as anatomically dependent positive or negative LMP modulations. Although movement detection was the dominant outcome, medial wall channels in two subjects enabled significant differentiation between individual fingers. In one subject, both contralateral and ipsilateral finger movements could be decoded with some generalization across hands; however, this observation is based on a single case and should be interpreted as preliminary.Significance.These findings identify the medial wall as a viable source of motor-related signals for finger movement decoding, with potential for future invasive motor BCI applications, while underscoring the need for further studies to confirm generalizability across individuals.
Objective.Electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) provide complementary temporal and spatial information for brain-computer interfaces (BCIs). However, effectively exploiting this complementarity remains challenging due to the heterogeneous characteristics of electrophysiological and hemodynamic signals.Approach.In this study, we propose Bi-modal Guidance and Spatio-Temporal Fusion Network (BiGSTF-Net), a multimodal architecture designed to improve cognitive state decoding by leveraging the complementary properties of EEG and fNIRS signals. The proposed framework first employs heterogeneous spatio-temporal feature extractors to capture temporal-oriented and spatial-oriented representations from each modality. To facilitate cross-modal interaction, a Modal Residual Interaction Unit is introduced to enable bidirectional inter-modal guidance while preserving modality-specific characteristics. Subsequently, a Spatio-Temporal Gated Unit performs intra-modal feature integration to produce compact and discriminative representations.Main results.Experiments conducted under cross-session evaluation on multiple BCI datasets demonstrate that BiGSTF-Net consistently outperforms representative multimodal fusion baselines. Ablation studies further verify the effectiveness of the proposed architectural components, while visualization analyses reveal activation patterns that align with known neurophysiological characteristics of EEG and fNIRS signals.Significance.These results indicate that the proposed framework provides an effective approach for multimodal neural signal decoding.
