Journal of neural engineering最新文献

Objective. machine learning has achieved significant success in electroencephalogram (EEG) based brain-computer interfaces (BCIs), with most existing research focusing on improving the decoding accuracy. However, recent studies have shown that EEG-based BCIs are vulnerable to adversarial attacks, where small perturbations added to the input can cause misclassification. Detecting adversarial examples is crucial for both understanding this phenomenon and developing effective defense strategies.Approach. this paper, for the first time, explores adversarial detection in EEG-based BCIs. We extend several popular adversarial detection approaches from computer vision to BCIs. Two new Mahalanobis distance based adversarial detection approaches, and three cosine distance based adversarial detection approaches, are also proposed, which showed promising performance in detecting three kinds of white-box attacks.Main results. we evaluated the performance of eight adversarial detection approaches on three EEG datasets, three neural networks, and four types of adversarial attacks. Our approach achieved an area under the curve score of up to 99.99% in detecting white-box attacks. Additionally, we assessed the transferability of different adversarial detectors to unknown attacks.Significance. through extensive experiments, we found that white-box attacks may be easily detected, and differences exist in the distributions of different types of adversarial examples. Our work should facilitate understanding the vulnerability of existing BCI models and developing more secure BCIs in the future.

Neurological disorders affecting speech production adversely impact quality of life for over 7 million individuals in the US. Traditional speech interfaces like eyetracking devices and P300 spellers are slow and unnatural for these patients. An alternative solution, speech Brain-Computer Interfaces (BCIs), directly decodes speech characteristics, offering a more natural communication mechanism. This research explores the feasibility of decoding speech features using non-invasive EEG. Nine neurologically intact participants were equipped with a 63-channel EEG system with additional sensors to eliminate eye artifacts. Participants read aloud sentences displayed on a screen selected for phonetic similarity to the English language. Deep learning models, including Convolutional Neural Networks and Recurrent Neural Networks with and without attention modules, were optimized with a focus on minimizing trainable parameters and utilizing small input window sizes for real-time application. These models were employed for discrete and continuous speech decoding tasks, achieving statistically significant participant-independent decoding performance for discrete classes and continuous characteristics of the produced audio signal. A frequency sub-band analysis highlighted the significance of certain frequency bands (delta, theta, and gamma) for decoding performance, and a perturbation analysis was used to identify crucial channels. Assessed channel selection methods did not significantly improve performance, suggesting a distributed representation of speech information encoded in the EEG signals. Leave-One-Out training demonstrated the feasibility of utilizing common speech neural correlates, reducing data collection requirements from individual participants.

Objetive. Decoding speech from brain activity can enable communication for individuals with speech disorders. Deep neural networks (DNNs) have shown great potential for speech decoding applications. However, the limited availability of large datasets containing neural recordings from speech-impaired subjects poses a challenge. Leveraging data from healthy participants can mitigate this limitation and expedite the development of speech neuroprostheses while minimizing the need for patient-specific training data.Approach. In this study, we collected a substantial dataset consisting of recordings from 56 healthy participants using 64 EEG channels. Multiple neural networks were trained to classify perceived sentences in the Spanish language using subject-independent, mixed-subjects, and fine-tuning approaches. The dataset has been made publicly available to foster further research in this area.Main results. Our results demonstrate a remarkable level of accuracy in distinguishing sentence identity across 30 classes, showcasing the feasibility of training DNNs to decode sentence identity from perceived speech using EEG. Notably, the subject-independent approach rendered accuracy comparable to the mixed-subjects approach, although with higher variability among subjects. Additionally, our fine-tuning approach yielded even higher accuracy, indicating an improved capability to adapt to individual subject characteristics, which enhances performance. This suggests that DNNs have effectively learned to decode universal features of brain activity across individuals while also being adaptable to specific participant data. Furthermore, our analyses indicate that EEGNet and DeepConvNet exhibit comparable performance, outperforming ShallowConvNet for sentence identity decoding. Finally, our Grad-CAM visualization analysis identifies key areas influencing the network's predictions, offering valuable insights into the neural processes underlying language perception and comprehension.Significance. These findings advance our understanding of EEG-based speech perception decoding and hold promise for the development of speech neuroprostheses, particularly in scenarios where subjects cannot provide their own training data.

Objective.To create highly immersive experiences in virtual reality (VR) it is important to not only include the visual sense but also to involve multimodal sensory input. To achieve optimal results, the temporal and spatial synchronization of these multimodal inputs is critical. It is therefore necessary to find methods to objectively evaluate the synchronization of VR experiences with a continuous tracking of the user.Approach.In this study a passive touch experience was incorporated in a visual-tactile VR setup using VR glasses and tactile sensations in mid-air. Inconsistencies of multimodal perception were intentionally integrated into a discrimination task. The participants' electroencephalogram (EEG) was recorded to obtain neural correlates of visual-tactile mismatch situations.Main results.The results showed significant differences in the event-related potentials (ERP) between match and mismatch situations. A biphasic ERP configuration consisting of a positivity at 120 ms and a later negativity at 370 ms was observed following a visual-tactile mismatch.Significance.This late negativity could be related to the N400 that is associated with semantic incongruency. These results provide a promising approach towards the objective evaluation of visual-tactile synchronization in virtual experiences.

Objective. Identifying the seizure occurrence period (SOP) in extended EEG recordings is crucial for neurologists to diagnose seizures effectively. However, many existing computer-aided diagnosis systems for epileptic seizure detection (ESD) primarily focus on distinguishing between ictal and interictal states in EEG recordings. This focus has limited their application in clinical settings, as these systems typically rely on supervised learning approaches that require labeled data.Approach. To address this, our study introduces an unsupervised learning framework for ESD using a 1D- cascaded convolutional autoencoder (1D-CasCAE). In this approach, EEG recordings from selected patients in the CHB-MIT datasets are first segmented into 5 s epochs. Eight informative channels are chosen based on the correlation coefficient and Shannon entropy. The 1D-CasCAE is designed to autonomously learn the characteristic patterns of interictal (non-seizure) segments through downsampling and upsampling processes. The integration of adaptive thresholding and a moving window significantly enhances the model's robustness, enabling it to accurately identify ictal segments in long EEG recordings.Main results. Experimental results demonstrate that the proposed 1D-CasCAE effectively learns normal EEG signal patterns and efficiently detects anomalies (ictal segments) using reconstruction errors. When compared with other leading methods in anomaly detection, our model exhibits superior performance, as evidenced by its average Gmean, sensitivity, specificity, precision, and false positive rate scores of 98.00% ± 3.51%, 94.94% ± 6.92%, 99.60% ± 0.30%, 79.92% ± 13.56% and 0.0044 ± 0.0030 h^-1respectively for a typical patient in CHB-MIT datasets.Significance. The developed model framework can be employed in clinical settings, replacing the manual inspection process of EEG signals by neurologists. Furthermore, the proposed automated system can adapt to each patient's SOP through the use of variable time windows for seizure detection.

Objective.Cortical beta (13-30 Hz) and gamma (30-60 Hz) oscillations are prominent in the motor cortex and are known to be transmitted to the muscles despite their limited direct impact on force modulation. However, we currently lack fundamental knowledge about the saliency of these oscillations at spinal level. Here, we developed an experimental approach to examine the modulations in high-frequency inputs to motoneurons under different motor states while maintaining a stable force, thus constraining behaviour.Approach.Specifically, we acquired brain and muscle activity during a 'GO'/'NO-GO' task. In this experiment, the effector muscle for the task (tibialis anterior) was kept tonically active during the trials, while participants (N= 12) reacted to sequences of auditory stimuli by either keeping the contraction unaltered ('NO-GO' trials), or by quickly performing a ballistic contraction ('GO' trials). Motor unit (MU) firing activity was extracted from high-density surface and intramuscular electromyographic signals, and the changes in its spectral contents in the 'NO-GO' trials were analysed.Main results.We observed an increase in beta and low-gamma (30-45 Hz) activity after the 'NO-GO' cue in the MU population activity. These results were in line with the brain activity changes measured with electroencephalography. These increases in power occur without relevant alterations in force, as behaviour was restricted to a stable force contraction.Significance.We show that modulations in motor cortical beta and gamma rhythms are also present in muscles when subjects cancel a prepared ballistic action while holding a stable contraction in a 'GO'/'NO-GO' task. This occurs while force levels produced by the task effector muscle remain largely unaltered. Our results suggest that muscle recordings are informative also about motor states that are not force-control signals. This opens up new potential use cases of peripheral neural interfaces.

Objective.Many psychiatric disorders involve excessive avoidant or defensive behavior, such as avoidance in anxiety and trauma disorders or defensive rituals in obsessive-compulsive disorders. Developing algorithms to predict these behaviors from local field potentials (LFPs) could serve as the foundational technology for closed-loop control of such disorders. A significant challenge is identifying the LFP features that encode these defensive behaviors.Approach.We analyzed LFP signals from the infralimbic cortex and basolateral amygdala of rats undergoing tone-shock conditioning and extinction, standard for investigating defensive behaviors. We utilized a comprehensive set of neuro-markers across spectral, temporal, and connectivity domains, employing SHapley Additive exPlanations for feature importance evaluation within Light Gradient-Boosting Machine models. Our goal was to decode three commonly studied avoidance/defensive behaviors: freezing, bar-press suppression, and motion (accelerometry), examining the impact of different features on decoding performance.Main results.Band power and band power ratio between channels emerged as optimal features across sessions. High-gamma (80-150 Hz) power, power ratios, and inter-regional correlations were more informative than other bands that are more classically linked to defensive behaviors. Focusing on highly informative features enhanced performance. Across 4 recording sessions with 16 subjects, we achieved an average coefficient of determination of 0.5357 and 0.3476, and Pearson correlation coefficients of 0.7579 and 0.6092 for accelerometry jerk and bar press rate, respectively. Utilizing only the most informative features revealed differential encoding between accelerometry and bar press rate, with the former primarily through local spectral power and the latter via inter-regional connectivity. Our methodology demonstrated remarkably low training/inference time and memory usage, requiring<310 ms for training,<0.051 ms for inference, and 16.6 kB of memory, using a single core of AMD Ryzen Threadripper PRO 5995WX CPU.Significance.Our results demonstrate the feasibility of accurately decoding defensive behaviors with minimal latency, using LFP features from neural circuits strongly linked to these behaviors. This methodology holds promise for real-time decoding to identify physiological targets in closed-loop psychiatric neuromodulation.

Objective. Temporal patterns in neuronal spiking encode stimulus uncertainty, and convey information about high-level functions such as memory and cognition. Estimating the associated information content and understanding how that evolves with time assume significance in the investigation of neuronal coding mechanisms and abnormal signaling. However, existing estimators of the entropy rate, a measure of information content, either ignore the inherent nonstationarity, or employ dictionary-based Lempel-Ziv (LZ) methods that converge too slowly for one to study temporal variations in sufficient detail. Against this backdrop, we seek estimates that handle nonstationarity, are fast converging, and hence allow meaningful temporal investigations.Approach. We proposed a homogeneous Markov model approximation of spike trains within windows of suitably chosen length and an entropy rate estimator based on empirical probabilities that converges quickly.Main results. We constructed mathematical families of nonstationary Markov processes with certain bi/multi-level properties (inspired by neuronal responses) with known entropy rates, and validated the proposed estimator against those. Further statistical validations were presented on data collected from hippocampal (and primary visual cortex) neuron populations in terms of single neuron behavior as well as population heterogeneity. Our estimator appears to be statistically more accurate and converges faster than existing LZ estimators, and hence well suited for temporal studies.Significance. The entropy rate analysis revealed not only informational and process memory heterogeneity among neurons, but distinct statistical patterns in neuronal populations (from two different brain regions) under basal and post-stimulus conditions. Taking inspiration, we envision future large-scale studies of different brain regions enabled by the proposed tool (estimator), potentially contributing to improved functional modeling of the brain and identification of statistical signatures of neurodegenerative diseases.

Objective.Stroke is a major cause of adult disability worldwide, resulting in motor impairments. To regain motor function, patients undergo rehabilitation, typically involving repetitive movement training. For those who lack volitional movement, novel technology-based approaches have emerged that directly involve the central nervous system, through neuromodulation techniques such as transcranial magnetic stimulation (TMS), and closed-loop neurofeedback like brain-computer interfaces (BCIs). This, can be augmented through proprioceptive feedback delivered many times by embodied virtual reality (VR). Nonetheless, despite a growing body of research demonstrating the individual efficacy of each technique, there is limited information on their combined effects.Approach.In this study, we analyzed the Electroencephalographic (EEG) signals acquired from 10 patients with more than 4 months since stroke during a longitudinal intervention with repetitive TMS followed by VR-BCI training. From the EEG, the event related desynchronization (ERD) and individual alpha frequency (IAF) were extracted, evaluated over time and correlated with clinical outcome.Main results.Every patient's clinical outcome improved after treatment, and ERD magnitude increased during simultaneous rTMS and VR-BCI. Additionally, IAF values showed a significant correlation with clinical outcome, nonetheless, no relationship was found between differences in ERD pre- post- intervention with the clinical improvement.Significance.This study furnishes empirical evidence supporting the efficacy of the joint action of rTMS and VR-BCI in enhancing patient recovery. It also suggests a relationship between IAF and rehabilitation outcomes, that could potentially serve as a retrievable biomarker for stroke recovery.

Objective.Decades ago, neurosurgeons used electrical impedance measurements in the brain for coarse intraoperative tissue differentiation. Over time, these techniques were largely replaced by more refined imaging and electrophysiological localization. Today, advanced methods of diffusion tensor imaging (DTI) and finite element method (FEM) modeling may permit non-invasive, high-resolution intracerebral impedance prediction. However, expectations for tissue-impedance relationships and experimentally verified parameters for impedance modeling in human brains are lacking. This study seeks to address this need.Approach.We used FEM to simulate high-resolution single- and dual-electrode impedance measurements along linear electrode trajectories through (1) canonical gray and white matter tissue models, and (2) selected anatomic structures within whole-brain patient DTI-based models. We then compared intraoperative impedance measurements taken at known locations along deep brain stimulation (DBS) surgical trajectories with model predictions to evaluate model accuracy and refine model parameters.Main results.In DTI-FEM models, single- and dual-electrode configurations performed similarly. While only dual-electrode configurations were sensitive to white matter fiber orientation, other influences on impedance, such as white matter density, enabled single-electrode impedance measurements to display significant spatial variation even within purely white matter structures. We compared 308 intraoperative single-electrode impedance measurements in five DBS patients to DTI-FEM predictions at one-to-one corresponding locations. After calibration of model coefficients to these data, predicted impedances reliably estimated intraoperative measurements in all patients (R=0.784±0.116,n=5). Through this study, we derived an updated value for the slope coefficient of the DTI conductance model published by Tuchet al,k=0.0649 S⋅smm-3 (originalk=0.844), for use specifically in humans at physiological frequencies.Significance.This is the first study to compare impedance estimates from imaging-based models of human brain tissue to experimental measurements at the same locationsin vivo. Accurate, non-invasive, imaging-based impedance prediction has numerous applications in functional neurosurgery, including tissue mapping, intraoperative electrode localization, and DBS.