Denoising EEG Signals for Real-World BCI Applications Using GANs

As a measure of the brain's electrical activity, electroencephalography (EEG) is the primary signal of interest for brain-computer-interfaces (BCI). BCIs offer a communication pathway between a brain and an external device, translating thought into action with suitable processing. EEG data is the most common signal source for such technologies. However, artefacts induced in BCIs in the real-world context can severely degrade their performance relative to their in-laboratory performance. In most cases, the recorded signals are so heavily corrupted by noise that they are unusable and restrict BCI's broader applicability. To realise the use of portable BCIs capable of high-quality performance in a real-world setting, we use Generative Adversarial Networks (GANs) that can adopt both supervised and unsupervised learning approaches. Although our approach is supervised, the same model can be used for unsupervised tasks such as data augmentation/imputation in the low resource setting. Exploiting recent advancements in Generative Adversarial Networks (GAN), we construct a pipeline capable of denoising artefacts from EEG time series data. In the case of denoising data, it maps noisy EEG signals to clean EEG signals, given the nature of the respective artefact. We demonstrate the capability of our network on a toy dataset and a benchmark EEG dataset developed explicitly for deep learning denoising techniques. Our datasets consist of an artificially added mains noise (50/60 Hz) artefact dataset and an open-source EEG benchmark dataset with two artificially added artefacts. Artificially inducing myogenic and ocular artefacts for the benchmark dataset allows us to present qualitative and quantitative evidence of the GANs denoising capabilities and rank it among the current gold standard deep learning EEG denoising techniques. We show the power spectral density (PSD), signal-to-noise ratio (SNR), and other classical time series similarity measures for quantitative metrics and compare our model to those previously used in the literature. To our knowledge, this framework is the first example of a GAN capable of EEG artefact removal and generalisable to more than one artefact type. Our model has provided a competitive performance in advancing the state-of-the-art deep learning EEG denoising techniques. Furthermore, given the integration of AI into wearable technology, our method would allow for portable EEG devices with less noisy and more stable brain signals.

Benchmarking cEEGrid and Solid Gel-Based Electrodes to Classify Inattentional Deafness in a Flight Simulator

Transfer from experiments in the laboratory to real-life tasks is challenging due notably to the inability to reproduce the complexity of multitasking dynamic everyday life situations in a standardized lab condition and to the bulkiness and invasiveness of recording systems preventing participants from moving freely and disturbing the environment. In this study, we used a motion flight simulator to induce inattentional deafness to auditory alarms, a cognitive difficulty arising in complex environments. In addition, we assessed the possibility of two low-density EEG systems a solid gel-based electrode Enobio (Neuroelectrics, Barcelona, Spain) and a gel-based cEEGrid (TMSi, Oldenzaal, Netherlands) to record and classify brain activity associated with inattentional deafness (misses vs. hits to odd sounds) with a small pool of expert participants. In addition to inducing inattentional deafness (missing auditory alarms) at much higher rates than with usual lab tasks (34.7% compared to the usual 5%), we observed typical inattentional deafness-related activity in the time domain but also in the frequency and time-frequency domains with both systems. Finally, a classifier based on Riemannian Geometry principles allowed us to obtain more than 70% of single-trial classification accuracy for both mobile EEG, and up to 71.5% for the cEEGrid (TMSi, Oldenzaal, Netherlands). These results open promising avenues toward detecting cognitive failures in real-life situations, such as real flight.

Unsupervised Clustering of Individuals Sharing Selective Attentional Focus Using Physiological Synchrony

Research on brain signals as indicators of a certain attentional state is moving from laboratory environments to everyday settings. Uncovering the attentional focus of individuals in such settings is challenging because there is usually limited information about real-world events, as well as a lack of data from the real-world context at hand that is correctly labeled with respect to individuals' attentional state. In most approaches, such data is needed to train attention monitoring models. We here investigate whether unsupervised clustering can be combined with physiological synchrony in the electroencephalogram (EEG), electrodermal activity (EDA), and heart rate to automatically identify groups of individuals sharing attentional focus without using knowledge of the sensory stimuli or attentional focus of any of the individuals. We used data from an experiment in which 26 participants listened to an audiobook interspersed with emotional sounds and beeps. Thirteen participants were instructed to focus on the narrative of the audiobook and 13 participants were instructed to focus on the interspersed emotional sounds and beeps. We used a broad range of commonly applied dimensionality reduction ordination techniques—further referred to as mappings—in combination with unsupervised clustering algorithms to identify the two groups of individuals sharing attentional focus based on physiological synchrony. Analyses were performed using the three modalities EEG, EDA, and heart rate separately, and using all possible combinations of these modalities. The best unimodal results were obtained when applying clustering algorithms on physiological synchrony data in EEG, yielding a maximum clustering accuracy of 85%. Even though the use of EDA or heart rate by itself did not lead to accuracies significantly higher than chance level, combining EEG with these measures in a multimodal approach generally resulted in higher classification accuracies than when using only EEG. Additionally, classification results of multimodal data were found to be more consistent across algorithms than unimodal data, making algorithm choice less important. Our finding that unsupervised classification into attentional groups is possible is important to support studies on attentional engagement in everyday settings.

The Origins of Passive, Active, and Sleep-Related Fatigue

Driving is a safety-critical task that requires an alert and vigilant driver. Most research on the topic of vigilance has focused on its proximate causes, namely low arousal and resource expenditure. The present article aims to build upon previous work by discussing the ultimate causes, or the processes that tend to precede low arousal and resource expenditure. The authors review different aspects of fatigue that contribute to a loss of vigilance and how they tend to occur; specifically, the neurochemistry of passive fatigue, the electrophysiology of active fatigue, and the chronobiology of sleep-related fatigue.

Classification of Game Demand and the Presence of Experimental Pain Using Functional Near-Infrared Spectroscopy

Pain tolerance can be increased by the introduction of an active distraction, such as a computer game. This effect has been found to be moderated by game demand, i.e., increased game demand = higher pain tolerance. A study was performed to classify the level of game demand and the presence of pain using implicit measures from functional Near-InfraRed Spectroscopy (fNIRS) and heart rate features from an electrocardiogram (ECG). Twenty participants played a racing game that was configured to induce low (Easy) or high (Hard) levels of demand. Both Easy and Hard levels of game demand were played with or without the presence of experimental pain using the cold pressor test protocol. Eight channels of fNIRS data were recorded from a montage of frontal and central-parietal sites located on the midline. Features were generated from these data, a subset of which were selected for classification using the RELIEFF method. Classifiers for game demand (Easy vs. Hard) and pain (pain vs. no-pain) were developed using five methods: Support Vector Machine (SVM), k-Nearest Neighbour (kNN), Naive Bayes (NB) and Random Forest (RF). These models were validated using a ten fold cross-validation procedure. The SVM approach using features derived from fNIRS was the only method that classified game demand at higher than chance levels (accuracy = 0.66, F1 = 0.68). It was not possible to classify pain vs. no-pain at higher than chance level. The results demonstrate the viability of utilising fNIRS data to classify levels of game demand and the difficulty of classifying pain when another task is present.

A Methodological Framework to Capture Neuromuscular Fatigue Mechanisms Under Stress

Neuromuscular fatigue is exacerbated under stress and is characterized by shorter endurance time, greater perceived effort, lower force steadiness, and higher electromyographic activity. However, the underlying mechanisms of fatigue under stress are not well-understood. This review investigated existing methods of identifying central mechanisms of neuromuscular fatigue and the potential mechanisms of the influence of stress on neuromuscular fatigue. We found that the influence of stress on the activity of the prefrontal cortex, which are also involved in exercise regulation, may contribute to exacerbated fatigue under stress. We also found that the traditional methods involve the synchronized use of transcranial magnetic stimulation, peripheral nerve stimulation, and electromyography to identify the contribution of supraspinal fatigue, through measures such as voluntary activation, motor evoked potential, and silent period. However, these popular techniques are unable to provide information about neural alterations upstream of the descending drive that may contribute to supraspinal fatigue development. To address this gap, we propose that functional brain imaging techniques, which provide insights on activation and information flow between brain regions, need to be combined with the traditional measures of measuring central fatigue to fully understand the mechanisms behind the influence of stress on fatigue.

Sensitivity of Physiological Measures of Acute Driver Stress: A Meta-Analytic Review

Background: The link between driving performance impairment and driver stress is well-established. Identifying and understanding driver stress is therefore of major interest in terms of safety. Although many studies have examined various physiological measures to identify driver stress, none of these has as yet been definitively confirmed as offering definitive all-round validity in practice.Aims: Based on the data available in the literature, our main goal was to provide a quantitative assessment of the sensitivity of the physiological measures used to identify driver stress. The secondary goal was to assess the influence of individual factors (i.e., characteristics of the driver) and ambient factors (i.e., characteristics of the context) on driver stress. Age and gender were investigated as individual factors. Ambient factors were considered through the experimental apparatus (real-road vs. driving simulator), automation driving (manual driving vs. fully autonomous driving) and stressor exposure duration (short vs. long-term).Method: Nine meta-analyses were conducted to quantify the changes in each physiological measure during high-stress vs. low-stress driving. Meta-regressions and subgroup analyses were performed to assess the moderating effect of individual and ambient factors on driver stress.Results: Changes in stress responses suggest that several measures are sensitive to levels of driver stress, including heart rate, R-R intervals (RRI) and pupil diameter. No influence of individual and ambient factors was observed for heart rate.Applications and Perspective: These results provide an initial guide to researchers and practitioners when selecting physiological measures for quantifying driver stress. Based on the results, it is recommended that future research and practice use (i) multiple physiological measures, (ii) a triangulation-based methodology (combination of measurement modalities), and (iii) a multifactorial approach (analysis of the interaction of stressors and moderators).

Eye Tracking in Driver Attention Research—How Gaze Data Interpretations Influence What We Learn

Eye tracking (ET) has been used extensively in driver attention research. Amongst other findings, ET data have increased our knowledge about what drivers look at in different traffic environments and how they distribute their glances when interacting with non-driving related tasks. Eye tracking is also the go-to method when determining driver distraction via glance target classification. At the same time, eye trackers are limited in the sense that they can only objectively measure the gaze direction. To learn more about why drivers look where they do, what information they acquire foveally and peripherally, how the road environment and traffic situation affect their behavior, and how their own expertise influences their actions, it is necessary to go beyond counting the targets that the driver foveates. In this perspective paper, we suggest a glance analysis approach that classifies glances based on their purpose. The main idea is to consider not only the intention behind each glance, but to also account for what is relevant in the surrounding scene, regardless of whether the driver has looked there or not. In essence, the old approaches, unaware as they are of the larger context or motivation behind eye movements, have taken us as far as they can. We propose this more integrative approach to gain a better understanding of the complexity of drivers' informational needs and how they satisfy them in the moment.

Hybrid Systems to Boost EEG-Based Real-Time Action Decoding in Car Driving Scenarios

The complexity of concurrent cerebral processes underlying driving makes such human behavior one of the most studied real-world activities in neuroergonomics. Several attempts have been made to decode, both offline and online, cerebral activity during car driving with the ultimate goal to develop brain-based systems for assistive devices. Electroencephalography (EEG) is the cornerstone of these studies providing the highest temporal resolution to track those cerebral processes underlying overt behavior. Particularly when investigating real-world scenarios as driving, EEG is constrained by factors such as robustness, comfortability, and high data variability affecting the decoding performance. Hence, additional peripheral signals can be combined with EEG for increasing replicability and the overall performance of the brain-based action decoder. In this regard, hybrid systems have been proposed for the detection of braking and steering actions in driving scenarios to improve the predictive power of the single neurophysiological measurement. These recent results represent a proof of concept of the level of technological maturity. They may pave the way for increasing the predictive power of peripheral signals, such as electroculogram (EOG) and electromyography (EMG), collected in real-world scenarios when informed by EEG measurements, even if collected only offline in standard laboratory settings. The promising usability of such hybrid systems should be further investigated in other domains of neuroergonomics.

A Willingness-to-Pay Associated Right Prefrontal Activation During a Single, Real Use of Lipsticks as Assessed Using Functional Near-Infrared Spectroscopy

Understanding consumer preferences and behavior is a major goal of consumer-oriented companies. The application of neuroscience to this goal is a promising avenue for companies. Previously, we observed a positive correlation during actual cosmetic use between the right dorsolateral prefrontal cortex (dlPFC) activity, measured by functional near-infrared spectroscopy (fNIRS), and the associated willingness-to-pay (WTP) values. However, we were unable to find any consistent group differences in the right dlPFC between different powdery foundations. Thus, the main objective of this study was to replicate the previous study and in addition, we aimed to refine the method of the previous study to increase the chance that a difference in valuation between different products can be detected. Twenty-five frequent lipstick using females were asked to apply six different lipsticks to their lips and to record how much they were willing to pay. To maximize the variation of the subjective experience of the products and the associated brain activity, the most preferred color lipstick and a less preferred color lipstick were chosen for each participant, and each color of lipstick had three different textures (Lo, Mid, and Hi). The time series was analyzed with the general linear model (GLM) and the correlation between the right dlPFC beta scores for the lipsticks and their respective WTP values conducted for each participant. This revealed a significant positive correlation and replicated our previous study. Surprisingly, the lipstick color and the texture manipulations did not result in any consistent differences in WTP and similarly no consistent group differences in brain activations. This study replicates our previous study extending it to a different type of cosmetic. The right dlPFC activity during the use of cosmetics may be a potential brain-based personalization or product selection process biomarker.