Objective. Upper extremity (UE) motor function loss is one of the most impactful consequences of stroke. Recently, brain-computer interface (BCI) systems have been utilized in therapy programs to enhance UE motor recovery after stroke, widely attributed to neuroplasticity mechanisms. However, the effect that the BCI's closed-loop feedback can have in these programs is unclear. The aim of this study was to quantitatively assess and compare the neuroplasticity effects elicited in stroke patients by a UE motor rehabilitation BCI therapy and by its sham-BCI counterpart.Approach. Twenty patients were randomly assigned to either the experimental group (EG), who controlled the BCI system via UE motor intention, or the control group (CG), who received random feedback. The elicited neuroplasticity effects were quantified using asymmetry metrics derived from electroencephalography (EEG), functional magnetic resonance imaging (fMRI), and diffusion tensor imaging (DTI) data acquired before, at the middle, and at the end of the intervention, alongside UE sensorimotor function evaluations. These asymmetry indexes compare the affected and unaffected hemispheres and are robust to lesion location variability.Main results. Most patients from the EG presented brain activity lateralisation to one brain hemisphere, as described by EEG (8 patients) and fMRI (6 patients) metrics. Conversely, the CG showed less pronounced lateralisations, presenting primarily bilateral activity patterns. DTI metrics showed increased white matter integrity in half of the EG patients' unaffected hemisphere, and in all but 2 CG patients' affected hemisphere. Individual patient analysis suggested that lesion location was relevant since functional and structural lateralisations occurred towards different hemispheres depending on stroke site.Significance. This study shows that a BCI intervention can elicit more pronounced neuroplasticity-related lateralisations than a sham-BCI therapy. These findings could serve as future biomarkers, helping to better select patients and increasing the impact that a BCI intervention can achieve. Clinical trial: NCT04724824.

Objective. Seizure onset zone (SOZ) localization and SOZ resection outcome prediction are critical for the surgical treatment of drug-resistant epilepsy but have mainly relied on manual inspection of intracranial electroencephalography (iEEG) monitoring data, which can be both inaccurate and time-consuming. Therefore, automating SOZ localization and surgical outcome prediction by using appropriate iEEG neural features and machine learning models has become an emerging topic. However, current channel-wise local features, graph-theoretic network features, and system-theoretic network features cannot fully capture the spatial, temporal, and neural dynamical aspects of epilepsy, hindering accurate SOZ localization and surgical outcome prediction.Approach. Here, we develop a method for computing dynamical functional network controllability from multi-channel iEEG signals, which from a control-theoretic viewpoint, has the ability to simultaneously capture the spatial, temporal, functional, and dynamical aspects of epileptic brain networks. We then apply multiple machine learning models to use iEEG functional network controllability for localizing SOZ and predicting surgical outcomes in drug-resistant epilepsy patients and compare with existing neural features. We finally combine iEEG functional network controllability with representative local, graph-theoretic, and system-theoretic features to leverage complementary information for further improving performance.Main results. We find that iEEG functional network controllability at SOZ channels is significantly higher than that of other channels. We further show that machine learning models using iEEG functional network controllability successfully localize SOZ and predict surgical outcomes, significantly outperforming existing local, graph-theoretic, and system-theoretic features. We finally demonstrate that there exists complementary information among different types of neural features and fusing them further improves performance.Significance. Our results suggest that iEEG functional network controllability is an effective feature for automatic SOZ localization and surgical outcome prediction in epilepsy treatment.

Objective. Virtual reality (VR) serves as a modern and powerful tool to enrich neurofeedback (NF) and brain-computer interface (BCI) applications as well as to achieve higher user motivation and adherence to training. However, between 20%-80% of all the users develop symptoms of cybersickness (CS), namely nausea, oculomotor problems or disorientation during VR interaction, which influence user performance and behavior in VR. Hence, we investigated whether CS-inducing VR paradigms influence the success of a NF training task.Approach. We tested 39 healthy participants (20 female) in a single-session VR-based NF study. One half of the participants was presented with a high CS-inducing VR-environment where movement speed, field of view and camera angle were varied in a CS-inducing fashion throughout the session and the other half underwent NF training in a less CS-inducing VR environment, where those parameters were held constant. The NF training consisted of 6 runs of 3 min each, in which participants should increase their sensorimotor rhythm (SMR, 12-15 Hz) while keeping artifact control frequencies constant (Theta 4-7 Hz, Beta 16-30 Hz). Heart rate and subjectively experienced CS were also assessed.Main results. The high CS-inducing condition tended to lead to more subjectively experienced CS nausea symptoms than the low CS-inducing condition. Further, women experienced more CS, a higher heart rate and showed a worse NF performance compared to men. However, the SMR activity during the NF training was comparable between both the high and low CS-inducing groups. Both groups were able to increase their SMR across feedback runs, although, there was a tendency of higher SMR power for male participants in the low CS group.Significance. Hence, sickness symptoms in VR do not necessarily impair NF/BCI training success. This takes us one step further in evaluating the practicability of VR in BCI and NF applications. Nevertheless, inter-individual differences in CS susceptibility should be taken into account for VR-based NF applications.

Objective. Sharing computational models offers many benefits, including increased scientific rigor during project execution, readership of the associated paper, resource usage efficiency, replicability, and reusability. In recognition of the growing practice and requirement of sharing models, code, and data, herein, we provide guidance to facilitate sharing of computational models by providing an accessible resource for regular reference throughout a project's stages.Approach. We synthesized literature on good practices in scientific computing and on code and data sharing with our experience in developing, sharing, and using models of neural stimulation, although the guidance will also apply well to most other types of computational models.Main results. We first describe the '6 R' characteristics of shared models, leaning on prior scientific computing literature, which enforce accountability and enable advancement: re-runnability, repeatability, replicability, reproducibility, reusability, and readability. We then summarize action items associated with good practices in scientific computing, including selection of computational tools during project planning, code and documentation design during development, and user instructions for deployment. We provide a detailed checklist of the contents of shared models and associated materials, including the model itself, code for reproducing published figures, documentation, and supporting datasets. We describe code, model, and data repositories, including a list of characteristics to consider when selecting a platform for sharing. We describe intellectual property (IP) considerations to balance permissive, open-source licenses versus software patents and bespoke licenses that govern and incentivize commercialization. Finally, we exemplify these practices with our ASCENT pipeline for modeling peripheral nerve stimulation.Significance. We hope that this paper will serve as an important and actionable reference for scientists who develop models-from project planning through publication-as well as for model users, institutions, IP experts, journals, funding sources, and repository platform developers.

Objective.Functional brain network (FBN) methods are commonly integrated with deep learning (DL) models for EEG analysis. Typically, an FBN is constructed to extract features from EEG data, which are then fed into a DL model for further analysis. Beyond this two-step approach, there is potential to embed FBN construction directly within DL models as a feature extraction module, enabling the models to learn EEG representations end-to-end while incorporating insights from FBNs. However, a critical prerequisite is whether DL models can effectively learn the FBN construction process.Approach.To address this, we propose using DL models to learn FBN matrices derived from EEG data. The ability of DL models to accurately reproduce these matrices would validate their capacity to learn the FBN construction process. This approach is tested on two publicly available EEG datasets, utilizing seven DL models to learn four representative FBN matrices. Model performance is assessed through mean squared error (MSE), Pearson correlation coefficient (Corr), and concordance correlation coefficient (CCC) between predicted and actual matrices.Main results.The results show that DL models achieve low MSE and relatively high Corr and CCC values when learning the Coherence network. Visualizations of predicted and error matrices reveal that while DL models capture the general structure of all four FBNs, certain regions remain difficult to model accurately. Additionally, a pairedt-test comparing global efficiency and nodal degree between predicted and actual networks indicates that most predicted networks significantly differ from the actual networks (p<0.05).Significance.These findings suggest that while DL models can learn the connectivity relationships of certain FBNs, they struggle to capture the intrinsic topological structures. This highlights the irreplaceability of traditional FBN methods in EEG analysis and underscores the need for hybrid strategies that combine FBN methods with DL models for a more comprehensive analysis.

Objective. Epilepsy is a neurological disorder that affects millions of patients worldwide. Electroencephalogram-based seizure detection plays a crucial role in its timely diagnosis and effective monitoring. However, due to distribution shifts in patient data, existing seizure detection approaches are often patient-specific, which requires customized models for different patients. This paper considers privacy-preserving domain incremental learning (PP-DIL), where the model learns sequentially from each domain (patient) while only accessing the current domain data and previously trained models. This scenario has three main challenges: (1) catastrophic forgetting of previous domains, (2) privacy protection of previous domains, and (3) distribution shifts among domains.Approach. We propose a sparse knowledge sharing (SKS) approach. First, Euclidean alignment is employed to align data from different domains. Then, we propose an adaptive pruning approach for SKS to allocate subnet for each domain adaptively, allowing specific parameters to learn domain-specific knowledge while shared parameters to preserve knowledge from previous domains. Additionally, supervised contrastive learning is employed to enhance the model's ability to distinguish relevant features.Main Results. Experiments on two public seizure datasets demonstrated that SKS achieved superior performance in PP-DIL.Significance. SKS is a rehearsal-free privacy-preserving approach that effectively learns new domains while minimizing the impact on previously learned domains, achieving a better balance between plasticity and stability.

Objective.Improving the efficacy of vagus nerve (VN) stimulation therapy requires a detailed understanding of the anatomical and functional organization of nerve fiber bundles and their fascicles. Variousex-vivoimaging platforms have been optimized for this purpose. However, all existing tools with micrometer resolution require labeling to enhance the fascicle contrast, and this labeling is resource-intensive and time-consuming. Polarization-sensitive optical coherence tomography (PS-OCT) was previously used to perform high-speed, label-free small animal (rat) sciatic nerve imaging but has not been applied for imaging the full-thickness large animal VNs (>1 mm diameter thick) due to tissue-limited imaging depth. We developed a PS-OCT platform that circumvents this problem and demonstrate high-speed label-free imaging of full-depth, multiple centimeters-long mammalian VNs for the first time.Approach.We employed a custom-built PS-OCT system with a dual-surface scanning microscope to capture opposite sides of the sample in a single frame. A tailored post-processing algorithm maximized fascicle contrast and merged the two surfaces together. Multi-centimeter-long porcine VNs were imaged.Main Results.Our approach reconstructed fascicle information throughout the full-thickness of the VN when compressed to a 650μm thickness. Moreover, we cross-validated PS-OCT measurements of fascicular organization and retardance to assess myelination against pair histology from the same specimens, showing Spearman's rank correlation coefficient value of 0.69 (p-value < 0.001).Significance.We demonstrated a label-free optical imaging method for large-volume VN imaging. The time to image a 6.8 cm nerve was 680 s with 0.1 mm s-1longitudinal sample translation speed, which is more than two orders of magnitude faster than existing modalities that require labeling. With this gain in speed and the possibility of label-free quantification of a fascicle's myelination level, important studies on inter-sample variability in fascicle organization become feasible.

Objective.A motor imagery (MI)-based brain-computer interface (BCI) enables users to engage with external environments by capturing and decoding electroencephalography (EEG) signals associated with the imagined movement of specific limbs. Despite significant advancements in BCI technologies over the past 40 years, a notable challenge remains: many users lack BCI proficiency, unable to produce sufficiently distinct and reliable MI brain patterns, hence leading to low classification rates in their BCIs. The objective of this study is to enhance the online performance of MI-BCIs in a personalized, biomarker-driven approach using transcranial alternating current stimulation (tACS).Approach.Previous studies have identified that the peak power spectral density value in sensorimotor idling rhythms is a neural correlate of participants' upper limb MI-BCI performances. In this active-controlled, single-blind study, we applied 20 min of tACS at the participant-specific, peakµfrequency in resting-state sensorimotor rhythms (SMRs), with the goal of enhancing resting-stateµSMRs.Main results.After tACS, we observed significant improvements in event-related desynchronizations (ERDs) ofµSMRs, and in the performance of an online MI-BCI that decodes left versus right hand commands in healthy participants (N= 10)-but not in an active control-stimulation control group (N= 10). Lastly, we showed a significant correlation between the resting-stateµSMRs andµERD, offering a mechanistic interpretation behind the observed changes in online BCI performances.Significance.Our research lays the groundwork for future non-invasive interventions designed to enhance BCI performances, thereby improving the independence and interactions of individuals who rely on these systems.

Objective.Real-world implementation of brain-computer interfaces (BCIs) for continuous control of devices should ideally rely on fully asynchronous decoding approaches. That is, the decoding algorithm should continuously update its output by estimating the user's intended actions from real-time neural activity, without the need for any temporal alignment to an external cue. This kind of open-ended temporal flexibility is necessary to achieve naturalistic and intuitive control. However, the relation between cortical activity and behavior is not stationary: neural responses that appear related to a certain aspect of behavior (e.g. grasp force) in one context will exhibit a relationship to something else in another context (e.g. reach speed). This presents a challenge for generalizable decoding, since the applicability of a decoder for a given parameter changes over time.Approach.We developed a method to simplify the problem of continuous decoding that uses transient, end effector-specific neural responses to identify periods of relevant effector engagement. Specifically, we use transient responses in the population response observed at the onset and offset of all hand-related actions to signal the applicability of hand-related feature decoders (e.g. digit movement or force). By using this transient-based gating approach, specific feature decoding models can be simpler (owing to local linearities) and are less sensitive to interference from cross-effector interference such as combined reaching and grasping actions.Main results.The transient-based decoding approach enabled high-quality online decoding of grasp force and individual finger control in multiple behavioral paradigms. The benefits of the gated approach are most evident in tasks that require both hand and arm control, for which standard continuous decoding approaches exhibit high output variability.Significance.The approach proposed here addresses the challenge of decoder generalization across contexts. By limiting decoding to identified periods of effector engagement, this approach can support reliable BCI control in real-world applications.Clinical Trial ID: NCT01894802.

Objective.Non-invasive brain stimulation (NIBS) methodologies, such as transcranial electric stimulation (tES) are increasingly employed for therapeutic, diagnostic, or research purposes. The concurrent presence of active/passive implants can pose safety risks, affect the NIBS delivery, or generate confounding signals. A systematic investigation is required to understand the interaction mechanisms, quantify exposure, assess risks, and establish guidance for NIBS applications.Approach.We used measurements, simplified generic, and detailed anatomical modeling to: (i) systematically analyze exposure conditions with passive and active implants, considering local field enhancement, exposure dosimetry, tissue heating and neuromodulation, capacitive lead current injection, low-impedance pathways between electrode contacts, and insulation damage; (ii) identify risk metrics and efficient prediction strategies; (iii) quantify these metrics in relevant exposure cases and (iv) identify worst case conditions. Various aspects including implant design, positioning, scar tissue formation, anisotropy, and frequency were investigated.Main results.At typical tES frequencies, local enhancement of dosimetric exposure quantities can reach up to one order of magnitude for deep brain stimulation (DBS) and stereoelectroencephalography implants (more for elongated passive implants), potentially resulting in unwanted neuromodulation that can confound results but is still 2-3 orders of magnitude lower than active DBS. Under worst-case conditions, capacitive current injection in the active implants' lead can produce local exposures of similar magnitude as the passive field enhancement, while capacitive pathways between contacts are negligible. Above 10 kHz, applied current magnitudes increase, necessitating consideration of tissue heating. Furthermore, capacitive effects become more prominent, leading to current injection that can reach DBS-like levels. Adverse effects from abandoned/damaged leads in direct electrode vicinity cannot be excluded.Significance.Safety related concerns of tES application in the presence of implants are systematically identified and explored, resulting in specific and quantitative guidance and establishing basis for safety standards. Furthermore, several methods for reducing risks are suggested while acknowledging the limitations (see section4.5).