Journal of neural engineering最新文献

Objective.This review paper focuses on how both direct current (DC) stimulation and alternating current (AC) stimulation affects the central nervous system's (CNSs) cells and its potential as a neurotherapeutic. Furthermore, addressing the promise of combinatorial approaches that utilize other treatments alongside electrical stimulation (ES) and how ES has shaped clinical approaches as a new rehabilitation treatment.Approach.Authors conducted this review to bridge the gap between basic research and clinical translation; 124 manuscripts were identified through Google Scholar for insights into ES effects on neurons and glia in bothin vitroandin vivomodels.Main results.The review summarizes findings from DC and AC stimulation paradigms applied toin vitroorin vivopreclinical models and summarizes the promise of ES when applied clinically. Generally, DC stimulation promotes axonal extension towards the cathode, while axons retract at the anode, limiting regeneration. AC stimulation alternates electrode polarity, enabling axonal extension in both directions. The intensity and duration of ES significantly affects the extent of neurite outgrowth. For astrocytes and microglia, ES-whether AC or DC-downregulates pro-inflammatory cytokine production and upregulates anti-inflammatory cytokine production, promoting A2 or M2 reactive states conducive to regeneration, respectively. Regarding oligodendrocyte precursor cells (OPCs), both DC and AC stimulation enhance OPC differentiation into oligodendrocytes, increasing myelin content and supporting axonal myelination. ES, when combined with stem cell treatments, drug delivery approaches, or with electroactive biomaterials, facilitate greater efficacy of these approaches. Clinically, short-single sessions of ES have shown long-term improvement. More specifically, preliminary efforts have been implemented to restore gait, hand tremors, and speech in spinal cord injuries, Parkinson's Disease, and stroke patients, respectively.Significance.ES is an evolving neurotherapeutic strategy for CNS related disease or injuries. Understanding how ES modulates neurons and glia is critical for optimizing its application in the clinic.

Objective.Conductivity estimation exploiting evoked potentials and fields is a promising method to reduce the uncertainty of electroencephalography (EEG) and combined EEG/magnetoencephalography (MEG) source analysis due to inter-individual variations of tissue conductivities. Approaches for skull conductivity estimation based on evoked potentials and fields have been proposed in several studies, but the current knowledge about their sensitivity towards uncertainties of other tissue conductivities and the effects on source analysis accuracy is insufficient. In this study, we analyze this sensitivity for EEG and EEG/MEG skull conductivity estimation and to what extent skull conductivity estimation improves the EEG, MEG, and combined EEG/MEG source analysis of interictal epileptic discharges (IEDs).Approach.We simulated EEG and MEG measurements of evoked brain activity and IEDs for randomly assigned tissue conductivities and performed EEG and EEG/MEG skull conductivity estimation for the simulated measurements. Following, we performed EEG, MEG, and combined EEG/MEG source analysis of the simulated IEDs and compared the results with and without using the individually estimated skull conductivities.Main results.We find that EEG/MEG skull conductivity estimation is more accurate than EEG skull conductivity estimation, especially when considering realistic noise levels, whereas the type of the evoked brain activity only had a minor influence on the accuracy of the conductivity estimation. Both EEG and EEG/MEG skull conductivity estimation clearly improve source analysis accuracy for EEG and combined EEG/MEG source analysis, reducing the uncertainty of the source localization from a few centimeters to less than one centimeter for most sources. However, we find that the effect of the conductivity estimation is less pronounced for sources at the base of the brain.Significance.EEG and EEG/MEG conductivity estimation exploiting evoked potentials and fields has the potential to become a valuable tool to reduce uncertainty in source analysis of IEDs, while it only requires little additional measurement effort.

Objective.Single-trial measurement of median nerve somatosensory evoked potentials (SEPs) with noninvasive electroencephalography (EEG) is challenging due to low signal-to-noise ratio (SNR), limiting its use in real-time neurorehabilitation applications. We describe and evaluate methodological optimizations for eliciting reliable median nerve SEPs measurable in real time, with reduced reliance on post-processing.Methods.In twelve healthy participants, two sessions each, SEPs were assessed at three pulse widths (0.1, 0.5, 1 ms), at a low-frequency stimulation (0.5 Hz ± 10%), and at an intensity sufficient to evoke consistent and robust sensory nerve action potentials and compound muscle action potentials. The evoked potential operant conditioning system platform was used to monitor responses in real time. Feasibility was also evaluated in a participant with incomplete spinal cord injury (iSCI).Results.SEP P50 and N70 were reliably elicited in healthy participants, and in individual with iSCI, across all tested pulse widths with minimal discomfort. N70 amplitude increased significantly with pulse width (χ2= 17.64,p= 0.0001,w= 0.80), while P50 amplitude remained unchanged. SNR showed a significant pulse width-dependent increase (χ2= 7.82,p= 0.02,w= 0.35) with improvements of 40% and 52% at 0.5 and 1 ms, respectively. N70 single-trial separability significantly improved at 1 ms (AUC of 0.83,χ2= 8.17,p= 0.017), including the iSCI participant (0.84-less impaired hand, 0.79-more impaired hand). Test-retest reliability (intraclass correlation coefficient = 0.70-0.84,p< 0.05) was highest at 0.5 ms, indicating more consistent N70 and P50 measurements across sessions at a longer pulse width.Significance.Robust median nerve SEPs can be measured at single trials with methodological optimizations such as a longer pulse width (0.5-1 ms), low frequency (0.5 Hz), a consistent afferent excitation guided by nerve and muscle responses, and a robust EEG acquisition system. This setup can be useful for real time SEP-based brain computer interface applications for rehabilitation.

Objective.Closed-loop responsive neurostimulators, such as the NeuroPace responsive neurostimulation (RNS) system, continuously monitor brain activity and deliver electrical stimulation in response to abnormal electrographic activity in patients with drug-resistant epilepsy. Practical technical constraints limit the temporal resolution of these devices, reducing the quality of EEG recordings.Approach.In this work, we introduce a novel technique to convert high-resolution intracranial electroencephalography (iEEG) obtained from inpatient monitoring into the same format, parameters, and resolution produced by the RNS system, allowing direct comparison of iEEG with RNS system data. We validated this technique using data from patients who had both iEEG and RNS. Electrodes from the iEEG and RNS system were co-registered onto the same 3D coordinate grid, and vector math was applied to determine the iEEG electrodes closest to the operational RNS electrodes.Main results.Through spectral analysis, we derived a transfer function that accounts for all filtering and data processing produced by the RNS system. Comparison of the recorded data using visual and spectral analysis from iEEG and RNS confirmed that EEG characteristics were correctly transformed by the filtering function, allowing analysis of how iEEG signals would appear within the RNS system. We demonstrate two examples from the extreme edges of the spectra, showing how DC shifts and high frequency oscillations would be transformed by the RNS. We provide a tutorial to tune this method to local device parameters, a process that can be applied to other devices as well.Significance.This tool allows researchers and clinicians to extract EEG biomarkers from high-resolution iEEG and determine if/how they can be detected in lower-resolution RNS. This provides an opportunity to develop patient-specific seizure detection parameters and investigate the long-term effects of neurostimulation therapy.

Objective.Foundation models have demonstrated transformative potential in medical artificial intelligence but remain underexplored in functional neuroimaging, particularly magnetoencephalography (MEG). This study aims to develop a domain-specific, self-supervised MEG clinical foundation model tailored for headache disorders to address the challenges of high-dimensional data and limited labeled datasets in clinical research.Approach. We developed a transformer-based model pretrained on a large-scale dataset comprising multi-state MEG recordings (resting-state, auditory, and somatosensory stimulation) from 416 participants (362 headache patients and 54 healthy controls). The model utilized a self-supervised masked-signal reconstruction strategy to learn latent spatiotemporal representations of neural activity. We evaluated the model's performance through signal reconstruction, visualization of attention weights, and downstream classification tasks comparing model-derived features against original MEG signals for migraine diagnosis.Main results. The pretrained model successfully reconstructed both continuous MEG signals and stimulus-specific evoked responses, effectively capturing intrinsic spatiotemporal brain dynamics. Visualization of the model's attention weights demonstrated spatial alignment with corresponding sensory brain regions, confirming its neurophysiological interpretability. Furthermore, classifiers trained on features extracted from the pretrained model significantly outperformed those using original MEG signals in identifying migraine patients, revealing distinct neural response patterns.Significance. This study introduces a scalable, data-efficient framework for clinical MEG analysis that significantly reduces reliance on manual feature extraction and labeled data. It demonstrates the efficacy of foundation models in decoding complex neural dynamics, offering promising implications for understanding neuropathology and facilitating precision diagnostics in neurology.

Objective.The relationship between swallowing motor imagery (MI) and actual swallowing remains unclear, leading to a lack of physiological basis for the application of swallowing imagery-based brain-computer interface (BCI) paradigms in rehabilitation. This research explored the link between swallowing execution and imagery, aiming to optimize BCI applications for swallowing rehabilitation in patients with dysphagia.Approach.Thirty healthy participants performed swallowing MI and saliva swallowing tasks under video cues, and electroencephalography (EEG) signals from 64 channels and electromyographic (EMG) signals from the suprahyoid muscles were recorded. This study investigates swallowing onset detection using EMG, and explores neural dynamics during swallowing imagery and execution through EEG-based time-frequency analysis, functional connectivity analysis, and nonlinear dynamic analysis (sample entropy (SampEn)).Main Results.The results revealed event-related desynchronization (ERD) in the central region (CPz, CP1-CP4) and parietal region (Pz, P1-P4) for both swallowing MI and actual swallowing. Pearson's correlation analysis indicated a weak but significant correlation (P= 0.0102). The ERD phenomenon during swallowing imagery was more similar to that during the pharyngeal stage, with a weak but significant correlation (P= 0.0139). Functional connectivity analysis revealed greater activation of the central region during swallowing imagery than during actual swallowing. In terms of SampEn, swallowing motor execution exhibited higher signal complexity and dynamic characteristics compared to imagery.Significance.This study highlights the similarity in neural activation between swallowing imagery and execution, particularly in the central and parietal regions, supporting the application of the swallowing imagery paradigm in these regions for rehabilitation. Further research is required to enhance BCI applications in swallowing disorders.

Objective. Epilepsy affects approximately 70 million individuals worldwide. Vagus nerve activity is known to be modulated by seizures; however, the types of fibers that are activated during seizures remain unknown. This work compares the electrical activity of the vagus nerve before, during, and after seizures in epileptic rats.Approach. Six rats experiencing pentylenetetrazol-induced epilepsy seizures and two rats under saline solution were investigated. Action potentials (AP) identified by template matching were sorted according to the fiber type they are deemed to originate from. AP templates were based on a 3D COMSOL simplified model of the vagus nerve. Model templates were established for fibers of different diameters based on histology. Correspondences are thus established based on specific fiber diameters.Main results. During seizures, an increase in the percentage of occurrence of APs was observed for 2μm and 3μm fibers, while a decrease was observed for 4µm, 5-6µm, and 7-11µm fibers. This was not observed in the rat group under saline solution. The increase in smaller diameter sizes is believed to be linked to an increase in autonomic activity.Significance. These findings contribute to a better understanding of vagus nerve dynamics during epileptic seizures and highlight the potential of vagus nerve activity as a physiological marker for seizure detection and monitoring. This would be of particular interest in vagus nerve stimulation to control any closed-loop form of therapy. This work provides a foundation for developing novel diagnostic and therapeutic approaches in epilepsy management.

Objective.To determine whether the perceptual intensity of speech signals-manipulated via loudness and dynamically adjusted through a brain state-dependent stimulation (BSDS) paradigm-modulates neural speech tracking and short-term memory.Approach.We implemented an EEG brain state-dependent design in which real-time variations in alpha power were used to modulate the loudness of pre-recorded digits during a task modelled on the digit span test. Speech tracking was quantified using lagged Gaussian copula mutual information (2-10 Hz), and behavioural performance was assessed through recall accuracy.Main results.Contrary to our initial hypothesis that higher loudness would enhance speech tracking and memory via bottom-up attention, digit recall accuracy was stable across loudness conditions. Speech tracking revealed an unexpected pattern: louder stimuli presented during high alpha power (low attention) elicited reduced tracking magnitudes and shorter peak latencies, whereas quieter stimuli delivered during low alpha power (high attention) produced stronger and more temporally extended tracking responses.Significance.These findings may suggest that internal attentional state, rather than external stimulus salience, plays a dominant role in shaping speech encoding. The study provides proof-of-concept evidence for BSDS in auditory paradigms, showing the importance of attentional fluctuations and stimulus loudness in determining the strength and timing of neural speech tracking, with implications for the design of adaptive speech-enhancement strategies.

Objective.The human brain organizes conceptual knowledge into semantic categories; however, the extent to which these categories share common or distinct neural representations remains unclear. This study aims to clarify this organizational structure by identifying consistent, modality-controlled activation patterns across several widely used and frequently investigated semantic domains in functional magnetic resonance imaging (fMRI) research. By quantifying the distinctiveness and overlap among these patterns, we provide a more precise foundation for understanding the brain's semantic architecture, as well as for applications such as semantic brain-computer interfaces (BCI).Approach.Following PRISMA guidelines, we conducted a systematic review and meta-analysis of 75 fMRI studies covering six semantic categories: animals, tools, food, music, body parts, and pain. Using activation likelihood estimation, we identified convergent activation patterns for each category while controlling for stimulus modality (visual, auditory, tactile, and written). Subsequently, Jaccard-based overlap analyses were performed to quantify the degree of neural commonality and separability across concept-modality pairs, thereby revealing the underlying structure of representational similarity.Main results.Distinct yet partially overlapping activation networks were identified for each semantic category. Tools and animals showed shared activity in the lateral occipital and ventral temporal regions, reflecting common object-based visual processing. In contrast, food-related stimuli primarily recruited limbic and subcortical structures associated with affective and motivational processing. Music and animal sounds overlapped within the superior temporal and insular cortices, whereas body parts and pain engaged occipito-parietal and cingulo-insular networks, respectively. Together, these findings reveal a hierarchically organized and modality-dependent semantic architecture in the human brain.Significance.This meta-analysis offers a quantitative and integrative characterization of how semantic knowledge is distributed and differentiated across cortical systems. By demonstrating how conceptual content and sensory modality jointly shape neural organization, the study refines theoretical models of semantic cognition and provides a methodological basis for evaluating conceptual separability. These insights have direct implications for semantic neural decoding and for the development of BCI systems grounded in meaning-based neural representations.

Objective. Stereoelectroencephalography (sEEG) is a mesoscale intracranial monitoring technique that records from the brain volumetrically with depth electrodes. sEEG is typically used for monitoring of epileptic foci, but can also serve as a useful tool to study distributed brain dynamics. Herein, we detail the implementation of sEEG-based brain-computer interfacing (BCI) across a diverse and large patient cohort.Approach. Across 27 subjects (15 female, 31 total feedback experiments), we identified channels with increases in broadband during hand, tongue, or foot movements using a simple block-design screening task. Subsequently, broadband power in these channels was coupled to continuous movement of a cursor on a screen during both overt movement and kinesthetic imagery.Main results. 26 subjects (29 out of 31 feedback conditions) established successful control, defined as more than 80 percent accuracy, during the overt movement BCI task, while only 12 (of the same 31 conditions) achieved control during the motor imagery BCI task. In successful imagery BCI, broadband power in the reinforced control channel(s) in the two target conditions separated into distinct subpopulations. Outside of the control channel(s), we demonstrate that imagery BCI engages unique subnetworks of the motor system compared to cued movement or kinesthetic imagery alone.Significance. Pericentral sEEG-based motor BCI utilizing overt movement and kinesthetic imagery is robust across a diverse patient cohort with inconsistent accuracy during imagined movement. Cued movement, kinesthetic imagery, and feedback engage the motor network uniquely, providing the opportunity to understand the network dynamics underlying BCI control and improve future BCIs.