Journal of Neural Engineering

Objective. Neural decoding performance depends strongly on neural signal representation; however, how the informative components of broadband intracortical activity vary with electrode configuration and decoding strategy remains unclear. We systematically compared the neural features spanning multiple temporal scales to identify the factors underlying robust motor decoding. Approach. Neural recordings were obtained from the dorsal premotor cortex, primary motor cortex, and dorsolateral prefrontal cortex of rhesus monkeys performing delayed-reaching tasks toward static or moving targets. Signals were recorded using 64-channel linear S-probes or 96-channel planar Utah arrays. We compared spiking-band power (SBP), local motor potential (LMP), and local field potential (LFP) power across multiple frequency bands using linear and nonlinear decoders. We further characterized the spectral and temporal components contributing to LMP decoding and evaluated the channel redundancy. Main results. The decoding performance of the SBP and high-frequency LFP features scaled with the number of recorded single units, whereas the LMP performance was more strongly influenced by the electrode geometry. LMP achieved a higher decoding performance with planar Utah arrays than with linear S-probes, accompanied by lower inter-electrode correlations and greater variability in preferred directions. Spectral analysis showed that LMP decoding was dominated by low-frequency components overlapping the delta band and movement-related cortical potentials, with an optimal decoding range of approximately 0.1-8 Hz. Nonlinear temporal decoders, particularly the convolutional neural network and gated recurrent unit models, further improved LMP decoding. Significance. These findings demonstrate that the neural features most informative for motor decoding depend on both the spatial sampling configuration and the temporal structure of the recorded activity. The results highlight the importance of low-frequency population dynamics and provide principles for optimizing neural-signal representations and reducing redundant neural data for scalable and wireless brain-computer interfaces.

Background and Objectives: Code-modulated visual evoked potentials (c-VEPs) provide fast and reliable control signals for reactive brain-computer interfaces (rBCIs). However, deployment beyond laboratories remains limited by visually intrusive stimulation, reliance on wet electroencephalography (EEG), and false activations during non-control periods. We introduce a usability-oriented c-VEP rBCI combining low-salience textured stimulation, slow flashing codes, asynchronous evidence accumulation, and semi-dry EEG acquisition. Methods: Thirty participants evaluated the system on a standard screen using an 11-class keypad typing task. Two identical online typing sessions were separated by a 2-min video-viewing period serving as an explicit idle condition. Feasibility was then assessed in virtual reality (VR) through a proof-of-concept shell-game task involving head movements and immersive rendering constraints. Visual comfort, peripheral distraction, and fatigue were assessed subjectively in both conditions. Results: In the keypad task, the system achieved 99.8% accuracy with 33 s of calibration data and a mean decoding time of 4.3 s. No spurious commands occurred during the tested 2-min idle condition. In VR, online accuracy reached 100%, with a mean decoding time of 5.6 s. Participants reported high visual comfort and minimal peripheral distraction in both conditions, although fatigue was higher in VR. Conclusion: The proposed c-VEP rBCI combines short calibration, high accuracy, comfortable stimulation, semi-dry EEG, and asynchronous decision-making. Performance during the explicit idle condition and VR implementation support the feasibility of wearable and practical c-VEP systems for self-paced interaction beyond laboratory environments. These findings encourage evaluation in longer idle periods and real-world tasks.

Dexterous hand control is essential for daily activities and a top restoration priority for people with tetraplegia, yet neurophysiological datasets suitable for studying and decoding naturalistic hand behavior remain scarce. Most available recordings constrain movements to a small number of degrees of freedom or cover only a single stereotypical grasp-and-hold structure. Moreover, previous brain-computer interface (BCI) studies have largely relied on signals from a single cortical area, even though dexterous hand control is distributed across an interconnected frontoparietal network. To address these limitations, we designed a paradigm that elicits rich hand movements across 40 distinct hand-object interaction conditions. We recorded simultaneous frontoparietal activity with hand kinematics from a rhesus macaque performing them. Neural activity was recorded from the primary motor cortex, anterior intraparietal area, and ventral premotor cortex using microelectrode arrays spanning 1,920 channels, while a multi-camera markerless motion-capture system acquired video from which kinematics were extracted offline. The dataset comprises data from 20 sessions, including processed neural features, synchronized 3D coordinates of hand and elbow keypoints, and trial-level metadata including task conditions and event timing. We further provide technical validation, including a neural decoding benchmark with multiple algorithms for continuous keypoint coordination reconstruction. We found that cross-day and cross-task decoding performance varied among algorithms. Variation in channel availability leads to different performance degradation curves in different algorithms. This task-rich dataset offers a valuable resource for investigating the neural mechanisms of dexterous hand control and for advancing the performance of motor BCIs.

Electrocutaneous stimulation provides a means to warn workers of potential hazards. Our previous parameter studies were conducted at rest or under controlled external influences such as vibration, temperature, and humidity, whereas actual work tasks have not yet been addressed. As a step towards greater practical applicability, we conducted two studies in which a circumferential electrocutaneous warning signal was applied to the upper right arm while participants performed a reading task, screw-driving, and polishing. In study 1 (n = 32), the pulse interval was adjusted individually to evoke a vibrating sensation, and the intolerance threshold was determined. In study 2 (n = 29), a fixed pulse interval of 36 ms was applied to all participants, the warning threshold was determined in addition to the intolerance threshold, and muscle twitches were documented both by participant report and by an independent observer. Study 1 demonstrated that electrocutaneous warning during work tasks is feasible, with a median intolerance threshold higher during polishing (16.5 mA) than during reading or screw-driving (14 mA each). In study 2, the fixed parameter setting elicited the intended pulsating or vibrating sensation at rest in 89 % of the participants in both presentations. The median warning threshold was 8 mA and did not differ between the three work tasks, whereas the intolerance threshold was lower during reading (18 mA) than during screw-driving and polishing (20 mA each). Muscle twitches were largely absent at the warning threshold and became frequent towards the intolerance threshold, with a ventral dominance of their location in both studies. An individual parametrization is therefore not required for the majority of users when a binary warning is to be conveyed, and operating the system slightly above the warning threshold provides a usable amplitude range while minimizing muscle twitches. Future work will focus on electrode optimization and on the objective measurement of motor responses.

From a statistical point of view, clustering spike time-windows from multi-electrode arrays (MEAs) recordings is a challenging high-dimensional, unsupervised clustering task where stationarity, commonly assumed in standard time-series analysis, is often violated and for which a gold standard is currently unavailable. Here we aim at providing practical guidance on how to cluster this type of data by systematically comparing 108 clustering pipelines differing along three main dimensions: (i) the data feature space; (ii) the metrics used to quantify distance between data points; and (iii) the clustering algorithm. For our benchmark we used both labeled synthetic data, mimicking four physiologically-inspired classes of spike time-windows, and a set of real MEAs recordings from a two-dimensional in-vitro neural culture. The performance of the competing pipelines was evaluated in terms of balanced accuracy and computational time when analyzing the synthetic datasets, while the Silhouette score was used for the real dataset, where no ground-truth is available. Overall, our analysis shows that the best combination is formed by k-means with Euclidean distance applied after Principal Component Analysis (PCA) of the spike time-windows. Conversely, hierarchical clustering showed the highest computational burden, while Independent Component Analysis and kernel PCA provided less effective noise suppression.

Purpose: Electrical stimulation of the occipital cortex evokes visual percepts known as phosphenes. To date, most human studies have examined phosphenes elicited by large surface or depth electrodes. These differ considerably from the small penetrating intracortical microelectrodes capable of stimulating highly localized cortical populations that are the basis of current next-generation devices. Here, we provide a detailed characterization of phosphenes evoked by such intracortical microelectrodes. Methods: As part of the CORTIVIS clinical trial, a Utah Electrode Array with 100 penetrating electrodes was implanted in the early visual cortex of four blind volunteers. Participants completed drawing and description tasks in response to single-electrode stimulation. Two participants also performed forced-choice tasks to determine whether phosphenes elicited by different electrodes could be reliably discriminated. Results: Single-electrode stimulation reliably evoked phosphenes in all participants. Most percepts appeared in the contralateral visual field, although two participants reported near-midline ipsilateral percepts. Phosphenes ranged from simple dots to elongated, textured, circular, or otherwise idiosyncratic forms. One participant reported distinguishable phosphene colors, which depended on electrode and amplitude structure and varied across sessions. The two participants tested on forced-choice discrimination could successfully identify electrode-specific phosphenes across multiple sessions. Conclusions: Intracortical microstimulation of the human visual cortex produces phosphenes with diverse, electrode-specific appearances that are recognizable across repeated trials. These findings indicate that penetrating microelectrodes often evoke more structured percepts than simple points or blobs, providing foundational evidence that electrode-specific encoding strategies will be necessary in future visual prosthetic devices.

Closed-loop brain stimulation offers a promising approach for modulating neural oscillations, but the mechanisms of phase-dependent neuromodulation remain poorly understood. We developed a real-time MEG-based closed-loop system delivering median nerve stimulation using a phase-modulated Poisson process. Sixteen healthy participants (11 female, 5 male) underwent open-loop characterization followed by closed-loop testing of eight phase-offset conditions (0 degrees-315 degrees), each paired with a statistically matched replay control. Closed-loop stimulation induced robust phase-dependent modulation (14.21 +/- 8.24%) versus replay (7.03 +/- 2.95%; p = 0.0003), with consistent phase specificity across participants (circular mean: 70.2 degrees, referenced to each individual's dominant frequency). At the group level, alpha oscillations (8-12 Hz) showed bidirectional modulation (mean boost: +21.0 +/- 19.9% at 0 degrees-135 degrees; mean suppression: -13.3 +/- 10.8% at 180 degrees-315 degrees), while beta oscillations (13-30 Hz) showed predominantly unidirectional suppression (-9.3 +/- 9.6%). Successful fixed-frequency targeting required alignment with each individual's natural peak. Misalignment led to different phase shifts producing power at different frequencies, a phenomenon we term phase-frequency coupling. Dynamic frequency tracking (8-30 Hz) automatically identified and targeted each individual's natural peak, achieving robust bidirectional modulation without prior knowledge of peak location. These findings demonstrate that real-time phase-locked peripheral nerve stimulation selectively modulates sensorimotor oscillations beyond statistically matched open-loop stimulation. The distinct phase-excitability relationships of alpha and beta oscillations, together with the importance of aligning stimulation with individual natural peaks, establish principled guidelines for personalized closed-loop neuromodulation.

With the rapid advancement of clinical neuroscience and Brain-Computer Interfaces (BCIs), there is an increasing demand for convenient, user-friendly EEG recording methods suitable for diverse environments, including mobile and home-based settings. Traditional gel-based EEG systems, while reliable, are inconvenient and time-consuming, whereas dry electrode systems tend to suffer from elevated noise levels. In this study, we investigated a novel methodological manipulation aimed at enhancing dry electrode signal quality: wetting the hair directly with tap water to improve scalp electrode conductivity. To this end, we recruited 22 healthy participants and compared their resting-state (RS) EEG activity across three experimental conditions (dry, semi-dry, and gel) within a single-session design. Specifically, we analyzed electrode impedance, spectral power (SP), and functional connectivity (FC). Our results show that basic hair-wetting substantially improved mean electrode impedance by 75.7% relative to the dry condition, reduced the proportion of bad channels from approximately 40% to 15%, and increased spectral-power similarity with gel recordings by 7.7% during eyes-closed recordings, and improved FC similarity by approximately 56-70% across frequency bands. Although this method does not fully match the signal quality of traditional gel-based systems, it represents a promising compromise, enhancing EEG data quality under suboptimal conditions. This approach offers a practical and non-invasive means to improve EEG signal quality, ultimately expanding real-world applications of EEG.

Tracking individual motor units (MUs) across multiple testing visits is critical for understanding neuromuscular adaptations and disease progression. However, existing methods fail to account reliably for spatiotemporal variations in motor unit action potentials (MUAPs) and lack validation for longitudinal tracking. Here, we present a robust computational framework for continuous MU tracking that integrates automatic estimation of electrode grid displacements and compensation for MUAP shape variations. Our algorithm demonstrated superior tracking accuracy, capturing 63% and 75% of MUs in the tibialis anterior and medial gastrocnemius, respectively, substantially outperforming state-of-the-art methods. By applying this approach to investigate estradiol-driven neuromuscular plasticity as a model system, we reveal previously undetectable modulations in recruitment thresholds and firing rates, linking hormonal fluctuations to motor unit behavior. Beyond advancing fundamental neuromuscular research, our approach holds promise for enhancing biomarker sensitivity in clinical trials, particularly in conditions such as amyotrophic lateral sclerosis, where reliable MU tracking is paramount. These findings establish a new standard for non-invasive motor unit tracking, unlocking opportunities for longitudinal neurophysiological assessment and evaluation of therapeutics.

Intracortical microstimulation (ICMS) directly modulates cortical activity, providing an artificial sensory stream to guide accurate control of prosthetic limbs. Yet, only a fraction of reported sensations evoked by ICMS are proprioceptive (i.e., describing the position and movement of the body). Taking a learning-based approach to encoding an artificial proprioceptive signal bypasses this limitation. For example, animals trained with multi-channel ICMS in association with natural vision learn to both decode the ICMS signal to localize an invisible target and learn to integrate artificial sensation with natural sensation. The limitation of a learning-based approach, however, is that it requires training. We hypothesized that changing the algorithm used to encode artificial sensory information could both reduce learning time and improve plateau performance on ICMS-guided navigation. To test this idea, we trained eight mice on a sensory-guided navigation task: mice were required to locate a target in a training cage, the position of which was encoded by a red circle (vision), by multi-channel ICMS (ICMS trials), or by both vision and ICMS (multimodal). ICMS was encoded using one of two algorithms: sparse (fewer simultaneously stimulating electrodes) or redundant (more simultaneously stimulating electrodes). We found that redundant encoding sped learning of the ICMS signal relative to sparse encoding. Further, mice guided by redundant ICMS ran faster and completed trials in less time than when guided by sparse ICMS. Having redundant encoding also facilitated multisensory integration of ICMS with natural vision, improving success rates, path efficiency, movement speed, and movement time. We conclude that optimized ICMS encoding algorithms could overcome the current limitations of learning-based sensory encoding, facilitating learning and integration of artificial sensory information into existing sensorimotor neural circuits, paving the way to restoring both the sensory and motor streams of information flow in damaged sensorimotor systems.

Electrical stimulation of brain activity has emerged as a key therapeutic intervention for neurological diseases. The success of these approaches relies on precisely tuned stimulation parameters. A major challenge is that these parameters should ideally be adjusted in real-time based on changes in brain activity. Closed-loop controllers address this challenge by monitoring brain activity and adapting stimulation according to ongoing changes in neural dynamics. While machine learning has been helpful in this context, closed-loop controllers typically do not adapt in real-time to fluctuations in brain activity. Here, we propose a novel framework based on reservoir computing to train a closed-loop, real-time neural controller. This approach, termed CLEAR-FORCE, trains the output weights of a recurrent network based on a combination of target signals and feedback from brain activity. Two series of simulations were performed, where the controller was successful at (1) inducing neural oscillations in a population of spiking neurons and (2) suppressing pathological oscillations. This framework provides a promising avenue for the implementation of neural controllers in low-powered, real-time, and adaptive hardware that harnesses machine learning for the personalized treatment of brain disorders.

Recurrent brain-computer interfaces (rBCIs) deliver targeted stimulation to one brain region based on neural activity recorded and processed in real-time at another. In a volitionally controlled rBCI, the user must also learn to intentionally produce the triggering neural activity, a process typically facilitated by an external reward. However, reliance on an external reward introduces a delay between the triggering event and reinforcement that can slow association between the two and, ultimately, control of the rBCI (particularly for neural activity not directly related to an observable behavior such as a goal-directed movement). The constraint to an external reward also limits studies of neural dynamics during natural, free behavior. In contrast, delivering activity-dependent stimulation directly to the brain's own reward centers - i.e., covert reinforcement - would bypass the need for external feedback, enabling precise temporal coupling between triggering events and reward and allowing for unrestrained rBCI use. Here, we show that rats can gain volitional control of an rBCI using activity-dependent electrical stimulation of the medial forebrain bundle as a sole reinforcer. Animals successfully learned to modulate motor unit discharge rates as well as specific electrocorticographic frequency bands (beta and gamma) to self-trigger stimulation. Furthermore, rats demonstrated flexible control by independently modulating distinct spectral features from the same cortical site. These findings suggest that covert reinforcement of endogenous reward networks can drive operant conditioning of specific neural signals during unrestrained behavior, opening new avenues for adaptive BCIs and closed-loop psychiatric therapies.

Black participants remain underrepresented in electroencephalography (EEG) research, partly because current EEG devices work poorly with dense, coily Type 4 Afro-textured hair. Research-grade EEG typically wires electrodes on a cap, and therefore the electrodes often fail to reach the scalp through dense, Afro-textured hair. Protocols that ask Black participants to unbraid or restyle protective hairstyles add burdens that discourage enrollment in research studies. To address this, we designed and tested 3D-printed attachments that fit onto an EEG electrode and anchored it directly in the hair, using the hairstyle itself as the mounting structure. We fabricated three design configurations for natural Afro-textured hair, cornrows/braids, and twists/microlocs. Seventeen healthy adults with Type 4a-4c hair used electrode attachments that matched their hairstyle. During a single visit, EEG was recorded with 16 conventional cap electrodes defined by the international 10-20 system and with 16 electrodes using novel electrode attachments anchored directly to their hair. Participants completed resting and walking trials with each system. Electrode-to-scalp distance and gel volume were measured during setup. Electrode disconnection, impedance, signal quality (i.e. peak alpha power and aperiodic (1/f) component), and comfort level were measured throughout the data collection. Compared with the conventional cap, the novel attachments reduced electrode-to-scalp distance, gel use, and impedance. Twenty percent of the conventional cap electrodes failed to connect to the scalp compared with 0.1% of novel electrodes. Impedance rose across trials for the conventional cap, whereas the novel attachments showed a smaller increase and maintained lower impedance throughout. Signal quality was significantly better using the novel attachments, while comfort level did not differ between systems. In conclusion, hair-anchored electrode attachment may present a feasible alternative to conventional cap-based EEG for people with Type 4 hair, yielding more usable channels and better signal quality, and supporting more inclusive EEG acquisition.

Severe hearing impairment affects approximately 5% of the global population, with a prevalence projected to increase over the coming decades. Cochlear implant devices pose a neuroprosthetic treatment option to allow sound perception and speech comprehension. However, users are limited in the spectrum of auditory information they can regain because electrical stimulation of the auditory nerve generates a broad spread of current and limits the transmission of sound frequency information. Recently, optogenetic stimulation of the cochlear nerve has emerged as a promising alternative for hearing restoration. Because light can be precisely confined in space, optogenetic stimulation of the cochlea permits to activate auditory neurons within smaller tonotopic regions, thereby improving the encoding of sound frequencies compared to electrical stimulation. However, this approach is limited in terms of dynamic range and ability to follow rapidly varying inputs, thus preventing faithful transmission of auditory information. Here, we developed an alternative approach targeting directly sensory inner hair cells with optogenetics. We demonstrate that this approach decreases the amount of light energy necessary to evoke auditory responses. Direct stimulation of hair cells increases the dynamic range of intensities encoded by neurons in subcortical auditory areas and improves the temporal fidelity of optogenetic responses. Finally, a biophysical model of the auditory system suggests that preserving the synapses between inner hair cells and auditory neurons greatly improves the encoding performance of an optical cochlear implant. Together, our results suggest that targeting sensory inner hair cells is a promising approach for hearing restoration through an optical implant.

Deep brain stimulation (DBS) is a well-established therapy for various neurological disorders. Clinical outcomes of DBS depend on electrode placement and stimulation programming, with growing evidence suggesting that targeting specific sites and networks can improve therapeutic efficacy. This targeting can be facilitated using high-accuracy, multi-modal biophysical models. In this article, we present the open-source platform OSS-DBS v2 that enables researchers and clinicians to build such biophys- ical models. Building upon previous work, the new version combines automated volume conductor generation, accurate and efficient electric field modelling and stimulation-driven neural simulations. OSS-DBS v2 provides an extensive library of intracranial electrodes, which can be modelled simul- taneously to test complex stimulation paradigms. Its application programming interface facilitates integration of external data workflows, such as impedance measurements. Furthermore, an integration with the Lead-DBS toolbox offered a straightforward graphical user interface for stimulation volume and pathway activation modelling, further complemented by an optimization framework. OSS-DBS v2 relies entirely on open-source software, supports all major operating systems and is designed for efficient execution on consumer-grade hardware.

Independent component analysis (ICA) is widely used to remove artefacts from EEG and TMS-EEG recordings, but the components that correspond to artefacts are usually identified by manual inspection, which is subjective and difficult to reproduce. Existing automated classifiers are either designed for ordinary EEG and do not target the artefacts specific to concurrent transcranial magnetic stimulation, or are built on EEGLAB, so none is available for FieldTrip-based pipelines. We present TEFAR, a lightweight, configurable framework for semi-automatic artefact-component classification that applies the same detectors to ordinary EEG and TMS-EEG. TEFAR is built on FieldTrip and requires no additional MATLAB toolboxes. Each artefact class (line noise, blinks, lateral eye movements, cranial muscle, cardiac activity, and the TMS decay and recharge artefacts) is identified from the spectral, spatial, or temporal signature that its components are known to express. Detection thresholds use robust statistics (the median and the median absolute deviation rather than the mean and standard deviation), so that a single dominant artefact component cannot raise the threshold that is meant to detect it. In 30 ground-truth simulations, TEFAR reached a specificity of 1.000 and a sensitivity of 0.96-0.97 for both profiles. An application of TEFAR to real EEG and TMS-EEG recordings is the next step and will be reported separately.

Photovoltaic subretinal prostheses replace the lost photoreceptors in patients blinded by atrophic age-related macular degeneration (AMD) and provide central vision by electrically stimulating the inner retinal neurons. Subretinal debris in geographic atrophy separate the inner nuclear layer from the implant by tens of micrometers, increasing the stimulation threshold and reducing the pattern contrast. Absence of this separating layer in animal models diminishes the accuracy of preclinical testing of the next-generation implants, especially with smaller pixels. We developed three-dimensional biomimetic scaffolds nanoprinted onto photovoltaic arrays to provide a precisely controlled separation between the implants and target neurons in animal models of retinal degeneration. Gap sizes of the scaffolds were adjusted to meet the following requirements: they should allow light penetration to the implant, provide unobstructed flow of electric current, but preclude migration of cells into their 3-D structure. Gaps of 3 m were found to be optimal as they restricted cellular migration and maintained the inner nuclear layer above both planar and pillar electrode arrays. Scaffolds did not measurably alter the photovoltaic response and accurately reproduced the distance dependence of stimulation threshold. Following implantation, the initially fluid-filled intra-scaffold space was progressively filled by cellular matter matching the density of the subretinal debris. This transition is expected to increase resistivity of the scaffold to that of tissue, which was evidenced by concurrently improved grating acuity. These results establish a model of the controlled separation between the implant and the target neurons reproducing the features of subretinal debris in geographic atrophy of human patients, thereby providing a platform for accurate preclinical evaluation of the subretinal implants.

Introduction: Safety parameters for non-invasive neuromodulation (NINM) remain scattered across consensus documents that differ in format, units, and scope, so pre-session parameter screening stays error-prone and difficult to audit. Objectives: To describe the NINM Safety Parameter Explorer, a browser-based tool that consolidates safety thresholds for six NINM modalities (e.g., TMS, tDCS, peripheral magnetic stimulation, TENS, transcranial ultrasound, and transcutaneous vagus nerve stimulation) into a single rule engine in which every numeric boundary carries an explicit, auditable provenance classification as literature-derived, literature-anchored, or operational default. Methods and Materials: We encoded the tool's rule corpus from published international consensus guidelines and peer-reviewed literature (82 modality threshold rules: 29 literature-derived, 36 literature-anchored, 17 operational defaults; complemented by 5 target-site caps for 87 total evaluative rules, alongside 15 informational reference values), then evaluated it in a simulated, automated internal verification suite of 38 example protocols spanning all six modalities and three expected verdicts, with automated static integrity assertions, regression guards, and behavioural and accessibility unit checks, all under a single pinned software version (v1.1.0-beta). Results: All 38 examples agreed with their declared verdicts (19 safe, 8 caution, 11 unsafe), and all static guards and supporting checks passed. Conclusion: The Explorer shows complete internal self-consistency between its rule corpus, engine, and interface, and its per-rule provenance labelling renders safety reasoning transparent and auditable for research and educational use. The tool is an in-silico protocol planning aid; it is not a medical device and is not intended for clinical decision-making or use in humans.

Background: High-definition transcranial alternating current stimulation (HD-tACS) is favored by the neurostimulation community for its precision and ability to influence neuronal dynamics. Yet, the exact mechanism by which the underlying brain structures are being affected remains unclear. We believe that the investigation of the aperiodic nature of the electroencephalograph (EEG) could shed light on the modulatory effects of HD-tACS. Methods: We analyzed the EEG of 9 participants during a compensatory tracking task (CTT) in two sessions, each with different HD-tACS protocols. Every session consisted of an initial period of no stimulation, followed by 30 Hz HD-tACS in the left motor (M30) or frontal (F30) cortex. We then isolated the aperiodic component of the EEG and calculated its spectral slope {beta}. Results and Discussion: {beta} decreased during F30 mainly in the right frontal cortex, indicating a shift towards higher frequencies and an increase of the excitatory/inhibitory balance. Additionally, we found that despite the long monotonus task the accuracy of the participants did not decrease, which might be attributed to the ability of both M30 and F30 to sustain attention for prolonged time. Finally, the change of CTT accuracy during the stimulation correlated with the {beta} of specific channels before the stimulation. This indicates the potential of {beta} to be used as a screening biomarker in future studies. In conclusion, we showed the ability of HD-tACS to alter EEG aperiodic dynamics and paved the way for future exploration of such dynamics in the field.

Deep learning models for electroencephalography (EEG) pathology detection achieve competitive accuracy but offer limited interpretability. In this paper, we introduce concept bottleneck models (CBMs) for EEG pathology detection which use clinical EEG biomarkers computed analytically from the signal as a bottleneck to be used for prediction. Those concepts include band power, spectral ratios, alpha peak frequency and amplitude, hemispheric asymmetry and phase locking values. With the backbone based on shallow convolutional neural network (CNN), we compare a black box baseline with two variants of CBMs. The hybrid variant is a classifier receives a residual projection of the backbone features. The pure variant is a CBM where the classifier receives only the predicted concepts. Models are evaluated on abnormality detection and dementia classification followed by concept prediction quality and leakage analysis. Both CBMs approach the baseline on TUAB (82.7-82.9% vs 83.6%) and are 2-5 percentage points lower on CAUEEG (57.9-60.5% vs 62.7%). The pure CBM exhibits concept fidelity concern where the predicted concepts deviate from their ground truth values. Replacing them with their true values causes severe cost on accuracy by 26 and 22 points on TUAB and CAUEEG. The concept fidelity concern is mitigated by increasing the weight of the concept loss to match the ground truth concepts. the mitigation comes with small accuracy cost on TUAB but a relatively larger cost on CAUEEG. The code of the experiments and full logs are available at https://www.github.com/mhmdrdwn/cbm.