The development of increasingly complex implantable neuromodulation devices requires extensive benchtop validation and testing prior to in vivo implementation. Our goal is to enable this in vitro testing by creating a benchtop system that is accessible, easy-to-use, and functions reliably across academic and industry environments, reducing discordant results that arise from discordant approaches. Furthermore, such a system paves the way for standardizing benchtop characterization of future implantable systems by establishing a repeatable and comparable methodology for validating these devices and their capabilities. Approach. We describe the NeuroTest Board (NTB), a low-cost, open-source device that combines a low-noise current source waveform generator with a high-precision data acquisition system. We provide information on its architecture, operation, and performance as an initial benchmark of the potential capabilities of the system. Main results. We demonstrate the use of the NTB to assess electrophysiology systems in two separate applications. In one case, we assess a commonly used laboratory system (the RHD2132 headstage and Open Ephys data acquisition system), while the second case characterizes an investigational neuromodulation device in development for human use (the CorTec Brain InterChange). Following this initial sensing characterization, we explore stimulation-related artifacts that would hamper in vivo data collection. With the NTB, we develop and evaluate artifact mitigation strategies for each system, thus demonstrating an example application of this benchtop testbed. Significance. The ability to test neuromodulation devices against a common benchmark will enable faster development of novel therapeutics by increasing inter-institutional reliability and decreasing troubleshooting time.

EEG source localization is an ill-posed inverse problem complicated by measurement noise, electrode dropout from sensor miscalibration, and variability in the number of active sources. Classical methods such as eLORETA and LCMV beamforming impose regularization assumptions that degrade under realistic noise conditions and do not naturally extend to multi-source scenarios. Recent neural network approaches improve localization accuracy but are typically trained and evaluated under idealized conditions that limit their applicability to real-world EEG acquisition. This work aims to develop and evaluate a source localization model that generalizes across the noise levels, source counts, and electrode configurations encountered in clinical and experimental EEG settings. Approach: We present a transformer-based model for single- and multi-source EEG localization. Hundreds of thousands of simulated EEG measurements are generated from known source locations, spanning a range of noise levels, source counts, and electrode dropout configurations, to train and evaluate the model. We establish a systematic evaluation framework that explicitly factors in source region density, signal-to-noise ratio, and electrode availability, and benchmark the transformer against classical (eLORETA, LCMV beamforming) and deep learning baselines across these conditions. We additionally conduct a thorough ablation study on network architecture and training pipeline design. Main results: Across all tested conditions, the transformer consistently outperforms classical and deep learning approaches, with improvements as high as 12\,mm in mean localization error (MLE) for multi-source scenarios under high noise. Additionally, when number of sources is unknown, the transformer can correctly predict the number and location of sources and drastically out perform the threshold-based approach (e.g an F1 Score of 68\% vs 12\% when simulated at 25\% noise). The multiembedding transformers maintains reasonable performance under conditions such as electrode dropout and increased noise. Significance: By explicitly modeling the noise, electrode dropout, and multi-source variability that characterize real EEG acquisition, this work addresses a key gap between benchmark performance and experimental utility in EEG source localization. These results lay the groundwork for more robust localization techniques applicable to real EEG data, with potential relevance to applications such as brain-computer interfaces. The data and code for this work will be made publicly available to support further development and benchmarking in the field.

We evaluated temporal neural features for motor decoding using a minimally-invasive endovascular brain-computer interfaces (BCI). We hypothesised that relatively long compound movements generate distinct patterns of neural activity that are encoded by temporal features. Approach. A single participant implanted with the Stentrode endovascular BCI attempted to hand write English characters, which were then classified offline and online based on the recorded brain activity. The input to the model was a one-dimensional trajectory of high-gamma power, summed across all channels to isolate the effect of temporal information on decoding performance. Main results. A small number of classes across multiple months was reliably decoded, showing that temporal decoding could identify stable patterns of brain activity associated with each handwriting movement. Across four online sessions, where each session contained 20-30 trials/character and 80-180 rest trials, the best performing (3-class) model achieved a mean rest recall of 87% as well as character recall and precision of 65% and 59%, respectively. These results showed that attempted handwriting can be used in an online multi-class selection BCI and that temporal neural features capture the internal user context. Linear warping of trials to minimise temporal variability, where all trials were modified to have the same duration, led to significantly increased decoding performance compared to classifying raw trials. Significance. The results from our participant showed that decoding brain activity based on the differences between feature trajectories across time can be an effective strategy for discriminating different movements. Therefore, BCIs may be able to use temporal information to reduce their spatial footprint, likely making them less invasive. Overall, our study shows that further investigation of temporal decoding across multiple participants and other recording modalities is warranted. Strategies for automatic temporal alignment are likely to further improve the performance of these systems.

Semantic Brain-Computer Interfaces (sBCIs) aim to decode conceptual information directly from neural activity, enabling more efficient and natural communication for individuals with severe motor impairments. However, existing approaches are often confounded by stimulus-driven perceptual processes and lack a reliable means of tagging the moment of concept formation. We introduce the Graded Inventory of Semantic Triggers (GIST), a novel electroencephalography (EEG) paradigm designed to elicit semantic retrieval from graded verbal clues with precise temporal alignment. Approach: EEG data were recorded from 14 participants performing the GIST task across three semantic categories (food, musical instruments, and body parts). Target neural activity was aligned to each participant's self-reported moment of semantic retrieval (spacebar press), enabling response-locked analysis. Here, we propose a staged framework where the first comprises a Conformer-based encoder trained on pooled data using a hybrid contrastive objective that combines instance-level consistency and cross-participant class alignment with domain-adversarial learning to reduce participant-specific information. In the second stage, the frozen encoder is transferred to a held-out participant whose data played no part in Stage 1, and a support vector machine is calibrated on that participant's own labelled items under a nested leave-one-item-out protocol. Results: Above-chance semantic decoding was achieved in all three pairwise tasks. Mean accuracies were 73.4% (Body vs. Food), 70.7% (Body vs. Music), and 68.2% (Food vs. Music). Participant-level permutation tests were significant in 9/11 participants for Body vs. Food and 8/11 for each of the other tasks (p < 0.05). The pre-response interval at [-1.5s, -0.5s] was the most discriminative. Significance: The GIST paradigm enables controlled elicitation and temporal alignment of semantic processing while reducing perceptual confounds. Combined with cross-participant representation pretraining and participant-specific calibration, the results demonstrate above-chance decoding of semantic categories from EEG in individuals excluded from representation learning, supporting the feasibility of sBCIs.

Cross-subject gesture recognition in wearable brain-computer interfaces is constrained by four structural deficiencies: symmetric cross-modal attention discarding cortical precedence over muscular execution, priors confined to post hoc analysis, invariance anchored on unimodal features, and adaptation requiring labelled target-domain data. This study closes these gaps by embedding corticomuscular coupling directly into the network architecture rather than recovering it from data. Approach: ACF-Net is a four-stage pipeline embedding neurophysiological priors directly into fusion. The corticomuscular coherence prior injection layer (CMC-Prior) encodes time-lagged coherence and phase-locking values for every electroencephalography (EEG)-electromyography (EMG) channel pair as a learnable bias tensor. Granger-causality-inspired directed cross-modal attention (GC-Attn) uses a causal mask allowing only EEG-to-EMG influence within 30-120 ms. The fused embedding is projected onto a symmetric positive definite manifold with log-Euclidean triplet loss. Physiology-consistent test-time adaptation (PC-TTA) updates only batch-normalisation affine parameters, using EEG-precedes-EMG as unlabelled self-supervision, eliminating labelled target data. Leave-one-subject-out validation on WAY-EEG-GAL (12 participants) and WEE-Ges (20 participants, three days) used fold-wise priors and Holm-corrected Wilcoxon tests. Main results: ACF-Net reached 87.6% on WAY-EEG-GAL and 89.2% on WEE-Ges, exceeding the strongest baseline by 9.4 and 9.1 percentage points with all comparisons significant after correction. Ablation attributed gains of 6.3, 4.9, 3.4 and 1.8 percentage points to directed attention, prior injection, manifold embedding and test-time adaptation. Accuracy remained 77.5% under 20 mm simulated electrode displacement, adaptation reduced the cross-day loss from 6.8 to 2.1 percentage points, and a forward pass on an NVIDIA Jetson Nano took 32.4 ms within a 178 ms end-to-end response at 30.9 Hz. Significance: Learned coupling weights recovered central-to-forearm-extensor projection, and delays matched corticospinal conduction with motor-unit recruitment, suggesting physiological structure, not dataset-specific statistics. Yet healthy young adults, gel-based/semi-dry wired hardware, and simulated displacement make this methodological; clinical and wireless dry-electrode transfer remains unestablished.

Objective. Speech Brain-computer interfaces (BCIs) have emerged as a pivotal technology in facilitating communication for individuals with speech impairments. Utilizing electroencephalography (EEG) for noninvasive speech BCIs offers an accessible and affordable solution, potentially benefiting a broader audience. However, EEG-based speech decoding remains controversial especially for overt speech, due to difficulties in separating speech-related neural activities from myoelectric potential artifacts generated during articulation. Moreover, because paradigms for collecting covert speech data are inherently less scalable than those for overt speech, whether the neural dynamics underlying EEG-based decoding of overt speech are also present during covert speech remains an open and practically important question. Approach. Here we aim to delineate the extent of the neural contributions to EEG-based speech decoding, and the degree to which the underlying neural dynamics are shared across overt and covert speech, by employing Explainable AI techniques to a convolutional neural network predicting spoken words based on signals obtained by ultra-high-density (uhd)-EEG. Main results. We found that electrode-wise contributions to the decoding cannot be explained by their mutual information with electromyography (EMG). Furthermore, contributing periods of speech to EEG-based decoding are distinct from those to decoding solely relying on EMG. In contrast, there are significant overlaps in signal timings contributing to EEG-based decoding, regardless of vocal conditions such as overt or covert speech. Notably, the denoising process successfully enhanced the decoding contribution from electrodes within speech-related brain areas for all speech conditions. Significance. Altogether, our findings support the idea that, with appropriate preprocessing, EEG becomes a valuable tool for decoding spoken words based on underlying neural activities. .

Objective A calibration-derived score has been related to online P300-speller accuracy in the same session, always within the cohort measured. We evaluated whether a mapping from that score to expected accuracy transports to withheld cohorts. Approach Retrospective secondary analysis of BigP3BCI 1.0.0, the largest evaluation of this relationship to date: 18 of 20 studies contributed 271 participants, 739 session-condition records, and 19,611 character selections. The predictor was a decoder's cross-validated discriminability on calibration epochs, one study withheld from development. Between-cohort variation was summarised by the random-effects standard deviation tau. Main Results The association was positive in all 18 cohorts but varied widely (r 0.190 to 0.928; participant-level r = 0.714, p < 0.001), with 12 CIs excluding zero. Pooled estimation error was 0.098 (95% CI 0.091 to 0.107) against a benchmark of 0.146. Calibration did not transport: intercept tau 0.87 (unrepresented-cohort interval -1.97 to 1.85, log-odds); cohort-specific slopes varied more than tenfold (tau 0.43, interval 0.111 to 2.005). Neither alignment (signal- or score-space) nor either nonlinear boundary meaningfully reduced this heterogeneity. Local recalibration with up to 16 participants did not reliably improve on transporting; refitting both intercept and slope was significantly worse than transporting at the smallest sizes. Median time per selection, a stopping-rule proxy, accounted for 66% of between-cohort slope variance in meta-regression. The four ALS cohorts gave an uncorrected slope spread of 0.223, against 0.560 overall. Significance The evaluated score is a classifier's cross-validated discriminability on each session's calibration data rather than the source studies' undocumented online decoder. That score reproducibly signals within-session accuracy, but its mapping to accuracy is cohort-specific and supports ranking sessions after local validation rather than reporting accuracy elsewhere. The four-cohort ALS subgroup appeared more homogeneous than the full archive, illustrating how conclusions about transportability from a small cohort set can be fragile. .

Scoring well on a task-demand label does not show that a decoder measures arousal. We ask whether peripheral physiology decodes demand better than EEG, whether the index behaves like graded arousal rather than a clock, and what a controller could do with it. Approach. We re-analysed a public boundary-avoidance flight task in which sixteen participants received EEG neurofeedback, half-sham feedback or silence. A decoder trained on peripheral signals (cardiac, respiratory, electrodermal, pupillometric) from an open-loop calibration block was compared with a retrained EEG decoder, filter-bank common spatial patterns, and motor-only and shuffled-label controls. All analyses are offline; the decoder was never in the loop. Main results. The peripheral decoder reached 93.0% within-subject AUC against 85.2% for EEG. Ring size shrinks on a fixed schedule, so accuracy alone cannot separate arousal from trial position. Three tests did: 48.6% of hard-course trials ended below the scale midpoint; the index separated the courses at matched elapsed time (p= 1.4 ×10-31), where EEG did not; and it tracked the autonomic channels. Only the peripheral index recovered an inverted-U with performance, robust in a duration-safe hazard analysis (p= 3.5 ×10-10). Arousal rose 14.1 points in the 10 s before failure; sustained between-condition differences above 7.5 points are excluded. Subjects varied reliably in how strongly excess arousal predicted failure, predictably from pre-loop calibration. Personalised control bands gave the best held-out discrimination, though at N = 16 indistinguishable from a retuned universal band. Significance. Arousal is a state of the whole body, and here the body reads it better than the scalp. The results separate three often-conflated requirements: accuracy on a demand label, validity as an arousal signal, and usefulness for control. The signal that added information over a fixed rule was distance above a band set by each subject's recent variability.

Despite decades of advances in neural interface technology, the long-term instability of implantable neuroprostheses remains a major barrier to achieving reliable lifetime performance. Chronic foreign-body reactions and fibrotic encapsulation progressively degrade the electrode-tissue interface, leading to signal deterioration and eventual device failure. Existing strategies that focus on improving material biocompatibility, reducing mechanical mismatch, or modifying implant surfaces have produced important but largely incremental improvements, while leaving the fundamental challenge unresolved. This perspective paper introduces a fundamentally different paradigm: Cybertissue Neuroprosthetics (CN). CN moves beyond the traditional electrode‑based neural interface by treating the implant itself as a functional tissue device-composed of living cells and optionally augmented with minimal nanoelectronics, nanomaterials, or genetic modifications-designed to integrate structurally and functionally with the host nervous system to achieve long‑term neuromodulation and information exchange. By minimizing or eliminating foreign materials and leveraging the brain's own cellular signaling mechanisms, CN promises to overcome the inherent limitations of current inorganic neuroprosthetic devices and seeks to achieve seamless lifetime integration between neuroprosthetic devices and the nervous system. This paper introduces the CN concept, unifies a range of emerging research directions under this umbrella, and articulates the core opportunities-from engineered cybercells to transplantable living devices-that collectively define a new interdisciplinary frontier. We argue that CN represents a foundational shift in neuroprosthetics, with the potential to deliver truly lifetime neural implants that are biologically integrated, self‑sustaining, and clinically transformative.

Peripheral neurons respond to structural, biochemical, and electrical cues to regenerate following injury. However, how these cues contribute individually and in combination to promoting neurite outgrowth remains incompletely understood. Here, we developed an aligned collagen fiber-based scaffold to examine how combinations of extracellular matrix proteins contribute to neurite outgrowth with and without electrical stimulus. Type I collagen fibers were fabricated by contact-drawing, a dry spinning process capable of producing collagen fibers that can be twisted into multifilaments that support anisotropic cell growth. A custom frame enabled immobilization of the collagen multifilaments and quantification of longest neurite length. Aligned collagen scaffolds significantly enhanced neurite outgrowth and alignment compared to isotropic controls.Incorporation of fibronectin or collagen IV increased neurite elongation in a concentrationdependent manner, with the longest neurites observed at intermediate ligand densities. Laminin I did not significantly alter neurite length. Combinatorial presentation of fibronectin and collagen IV further demonstrated non-linear responses, with low-concentration combinations enhancing neurite extension and high-concentration combinations providing no additional benefit. Electrical stimulation increased neurite length on collagen-only scaffolds but exhibited diminished or negative effects when applied to ECM-enriched multifilaments, suggesting context-dependent integration of biochemical cues and electrical stimulus. Together, these findings demonstrate that scaffold anisotropy establishes directional guidance, whereas ECM composition and electrical stimulation modulate neurite extension in a non-additive manner. The collagen multifilament system functions as a tunable in vitro platform for systematically interrogating multimodal regenerative cues and defining design principles for next-generation nerve guidance biomaterials.

The question of the mechanisms by which cellular interactions in the brain are converted into integrative behavioral phenomena is closely related to functional localization. Electrophysiology is usually limited in its ability to display information about the functioning of neural circuits with sufficient spatial resolution, although the temporal resolution is almost always very high. Other modern methods for recording the functional characteristics of neural activity in the brain vary considerably in spatial and temporal resolution. Voltammetry is one of the key instrumental methods used in neurobiology to measure neurotransmitter release and dynamics in vivo. It is similar to electrophysiology in terms of temporal resolution, but spatial resolution remains low. This method records changes in the concentration of chemicals, typically neurotransmitters, in the brain's extracellular environment with sufficient temporal resolution to capture release peaks. However, as this monitoring is performed at a single location or target, the technique does not provide information about the spatial distribution of the recorded signal characteristics. Various techniques for functional optical imaging of brain structure and activity offer relatively high spatial resolution, often accompanied by adequate temporal resolution. Some of these methods are suitable for visualizing neural activity in both 2D and 3D. However, while functional brain imaging allows areas of neural activity to be identified, it typically does not provide quantitative information about changes in the concentration levels of specific molecules. A significant challenge would be to combine the key advantages of voltammetry and functional optical brain imaging, i.e., to obtain continuous quantitative information on the concentration of neurochemically active molecules across many areas of the investigated brain region with sufficiently high temporal and spatial resolution. This review discusses the current state of voltammetry and functional brain imaging techniques, as well as the hypothetical possibilities for combining them based on the experimental data obtained to date. .

Periodic discharges (PDs) and rhythmic delta activity (RDA) are common EEG patterns in critically ill patients that require characterization of lateralization, spatial localization, and frequency. Manual characterization is subjective and time-consuming. We present an automated system for all four major subtypes: lateralized and generalized PDs (LPD, GPD) and lateralized and generalized RDA (LRDA, GRDA). Approach. The PD-Profiler combines a per-channel convolutional neural network, hemisphere-specific learned evidence traces, dynamic programming with an approximately periodic prior, and discharge-locked topographic localization. The RDA-Profiler uses iterative narrowband Hilbert refinement for frequency and lateralization, with phase-locking-value analysis for spatial extent. Both pipelines were trained and evaluated on 12,425 EEG segments from 11,729 patients. Main results. Four electroencephalographers independently scored 200 stratified segments per subtype. The algorithm exceeded expert-expert (EE) agreement on LPD frequency (expert-algorithm (EA) ICC 0.931 versus EE 0.897; p<0.001) and was statistically indistinguishable from EE on GPD frequency (EA 0.980 versus EE 0.973; p=0.061), GRDA frequency (p=0.056), LRDA frequency (p=0.435), LPD laterality (EA kappa 0.945 versus EE 0.974; p=0.079), and LRDA laterality (EA kappa 0.961 versus EE 0.993; p=0.269). No attribute fell significantly below the EE ceiling. PD discharge timing achieved an F1 of 0.817 at single-sample precision (mean absolute timing error 3.1 ms). PD spatial localization reached 97.3% of expert inter-rater Jaccard agreement, and RDA spatial extent matched EE agreement. Significance. GROND reaches expert-level reliability across lateralization, spatial localization, discharge timing, and frequency for PDs and RDA. In discordant cases, experts judged its frequency estimates more accurate than the original labels in 94% of cases, supporting reproducible characterization at scale.

Objective.Motor imagery recruits a distributed network whose components and large-scale organization are well characterized. The temporal order in which those components are engaged is not: the latency at which each region begins to desynchronize within a trial has not been compared statistically in the healthy brain, and the interval in which movement-specific information appears remains unidentified. Early post-cue classification peaks have been attributed to visually induced orienting rather than to imagery, an inference drawn from a condition contrast rather than direct measurement.Approach.Two independent public datasets were analysed under a matched pipeline of causal filtering, ocular regression and current source density: Weibo2014 as the internal dataset and BCI Competition IV dataset 2a as the external one. Onset was compared across regions under nine definition-by-threshold combinations, and the time course of class-discriminative information was obtained by time-resolved decoding and modelled with and without a transition.Main results.Mu and beta desynchronization onset was ordered across cortex, from occipital through sensorimotor and midline pre-supplementary motor area to lateral prefrontal. The ordering held under all nine onset criteria examined, under channel-count matching and within-trial pairing, and after removing the sensorimotor component. Class-discriminative information rose most steeply between 0.60 and 0.75 s and saturated within a further 0.3 s. This transition lay closer to sensorimotor onset than to cue presentation in every participant of both datasets and fell at or after it in nearly all, although the two were not shown equivalent at the resolution of the decoding grid.Significance.Windows centred on each region's own onset classified equivalently to reassigned windows, so the ordering serves as a temporal landmark rather than a decodable feature, locating the interval in which that information becomes available and placing the analysis interval for imagery decoding on a measured rather than a conventional basis.