Here, we present the first online BCI based on deep learning (DL). Application of DL has so far focused on computer-vision tasks. Recently, the BCI community is also developing an increasing interest for DL. We have previously shown DL using convolutional networks (ConvNets) to be competitive with the current state-of-the-art algorithms for motor-execution offline decoding. The present work extends the application of ConvNets to online decoding. We acquired EEG in five healthy subjects (S1-5) each performing five mental tasks: right hand finger tapping, both feet toe movement, rotation of an L shape, generation of words starting with the same letter, and rest. We performed training and feedback with a menu-based graphical user interface (GUI). During offline training, actions with the GUI were controlled by the paradigm. We then used the raw offline EEG data to train hybrid networks combining a deep ConvNet with a shallower ConvNet. Test evaluation on the last two runs (each 10min) of the EEG data yielded class averaged offline decoding accuracies of 70.7%, 49.2%, 73.1%, 58.8% and 32.6% for each subject, respectively. There were no significant differences between offline deep ConvNets and Filter Bank Common Spatial Patterns (FBCSP) results, confirming that deep ConvNets are competitive with FBCSP. Using self-developed visualization techniques, we show that the ConvNets extracted neuronal features in the delta, alpha and beta bands. During online feedback, all subjects successfully navigated the menu structure with 5 actions (select, down, back, up, wait) using the trained ConvNets, demonstrating the feasibility of DL-based online BCI control.

Based on recent methodological and technical progress, as well as on an increasing knowledge about the neural correlates of behavior and cognition, brain-computer interfaces (BCIs) are attracting growing interest in both the scientific and medical communities. According to the survey from several articles, a portable BCIs is needed for practical application. Now, most of the BCI is based on Personal Computer (PC), which is bulky and not portable. So in this paper, a BCIs, which is based on embedded processor, is designed.

The main content of this paper include embedded system building, the design of USB driver and BCI application program. And the embedded system building includes the cutting of embedded Linux kernel, file system’s making and the installation of the QT library. The function of the USB driver is to transfer EEG data from EEG amplifier to embedded system. The BCI application program includes a waveform display program and a BCI program. And the BCI program implements the function of visual stimulation, EEG data acquisition, feature extraction and pattern classification, etc.

Finally, in order to verify the performance of the embedded BCI system, we tested the whole system and its core module. The average accuracy of the BCI system is 82.8%, and the average information transfer rate is 19.9 bits per minute. The test results show that the BCI system can meet the demand of basic medical application.

Generating and monitoring motor actions are key functions of the human nervous system. Brain disorders can lead to motor impairments that limit affected individuals in their ability to communicate and interact with their environment. Brain-computer interfaces (BCIs) can be used to replace lost or improve impaired functions. Despite of the usefulness of current non-invasive BCIs, there is still room for improvement in respect to their performance and specificity. To provide more advanced features for BCIs, we are using electroencephalographic (EEG) source imaging for investigating cortical dynamics linked to motor functions.

We have studied repetitive movements [i.e. gait (N=10) and finger movements (N=18) ], distinguishing movement state-related from movement phase-related activities. Movement state-related activities are reflected by sustained suppression of mid beta (18–24 Hz) and enhancement of high gamma (60-80 Hz) oscillations during movement. These activities are suggested to represent upregulation of cortical excitability in regions representing the limb that is moved. Movement phase-related activities appear as dynamic amplitude modulations, most pronounced at high beta (24-40 Hz) frequencies in prefrontal and bilateral sensorimotor areas. These patterns are significantly related to the time course of repetitive movements. Because we identified the frequency spectra and spatial sources of movement state- and movement phase-related activities to be different, we suggest that they represent different functional large-scale networks providing independent information. In our latest work, we utilize these two network specific EEG source features to improve BCIs in decoding brisk movements.

Functional brain connectivity (FC) is a marker of brain state. In last decade there is a growing interest in examination of its dynamics which is connected to the synchronization and desynchronization of neuronal populations activity. To better understand these processes we analysed resting state simultaneous fMRI/EEG data. The aim was to answer whether dynamics of FC (as seen by fMRI) is reflected in EEG power fluctuations.

We measured 50 healthy controls with 1.5T MRI and 30 channel MR compatible EEG during resting state. Preprocessing was done in the standard way in SPM8 and BrainVision Analyzer software. BOLD data were decomposed by Group ICA into 20 stable components (tested for inter/intra class stability). Dynamic functional connectivity (DFC) was computed on component's time series using sliding rectangular window and Pearson correlation coefficients. DFC state vector was then extracted for each component pair based on values of correlation coefficients (3 states - synchronous, asynchronous and no communication).

Simultaneously, EEG data were analysed by Parallel Factor Analysis (PARAFAC), where 3D spectrogram (3-way array with modes of electrodes, time and frequency) was decomposed. Resulting components contain 3 signatures - spatial, temporal and spectral.

Finally, ANOVA tests were performed to assess relations between DFC state vectors and fluctuations of EEG spectral patterns.

Our previous findings revealed relation between fluctuations of EEG spectral patterns and hemodynamics of large scale brain networks (LSBN). We present results, which show that the relation exists also at the level of DFC among LSBN.

Long-term exposure of humans to unusual sensory spaces is effective to see the adaptive strategy for an environment. Because little had been examined about adaptation to left-right reversed audition, we constructed a left-right reversed stereophonic system using wearable devices and examined the adaptation effects on audiovisual spatial integration, as reported in BaCI 2015. However, the adaptation effects on auditory-motor spatial coordination still remains unknown. In a way similar to our previous study, we asked two participants to wear the system for 4 weeks, and tested the effects with MEG every week. The MEG responses were measured under the selective reaction time task, where they immediately distinguished between sounds delivered to either the left or the right ear with the index finger on the compatible or incompatible side. The N1m intensities for the response-compatible sounds tended to be larger than those for the response-incompatible sounds until the third week but decreased on the fourth week, which correlated with the initially shorter and longer reaction times for the compatible and incompatible conditions, respectively. Moreover, Granger causality analysis showed disruption of the auditory-motor connectivity in the second week, with the largest N1m intensities and the longest reaction times, irrespective of compatibility. We thus conclude that long-term exposure to the left-right reversed audition optimizes the auditory-motor coordination based on the new rule, where the transient unstable situation leads subsequent modulation of early auditory processing.

The vestibular system has a critical role in sensing body tilt, gravity direction, and acceleration input, and receives information from other sensory systems such as the visual system. Because of the technical difficulty in designing a neuroimaging experiment for testing the vestibular system in humans, however, little is known about the interaction between vestibular and visual information. Here, we tested the visual-vestibular interaction using EEG with two apparatuses: a virtual reality head-mounted display and an inversion table. Videos of falling scenes in either upward or downward direction were randomly displayed by the VR device, while a participant's body tilt was alternately manipulated for every eight video plays in either an upright or inverted manner by the inversion table. Therefore, four combinational conditions were established with regard to the orientation of retinal images and the direction of gravity. Event-related potential analysis revealed that continuous attenuation of visual activity was observed from 100-150 ms after the start timing of falling in the inverted body condition as compared with the upright condition, irrespective of the visual falling direction. Moreover, Granger causality analysis showed feedback connection from the temporoparietal vestibular area to the visual area.These findings indicate that visual activity is suppressed by unusual vestibular information and that visual-vestibular interaction begins at a relatively early stage of visual processing.

Transcranial direct current stimulation (tDCS) is a non-invasive neuromodulation technique. There exists literature on tDCS effects that can modulate brain cognitive functions (e.g., mental arithmetic skills) in the posterior parietal cortex (PPC) and dorsolateral prefrontal cortex (DLPFC). In this study, in the hope to find other stimulation effects, we investigate tDCS-induced brain activity with magnetoencephalography (MEG). Fifteen healthy subjects participated in this experiment. All subjects conducted complex mental multiplication task on three different days under different stimulation conditions (anode, cathode, and sham). The tDCS (Starstim) was applied with current of 1.5mA for 25 minutes. Active and reference electrodes were placed on left PPC and right DLPFC, respectively. 152-channel whole-head MEG (KRISS) was recorded before and after tDCS. We observed significant differences in event-related (de)synchronization (ERD/ERS) of alpha power in the post-tDCS conditions. In the left temporo-parietal area, alpha ERD increased in anodal-tDCS than sham. Furthermore, mild increase of alpha ERS in the right centro-parietal area was observed in cathodal-tDCS compared to sham. The results show that anodal-tDCS over the left PPC causes alpha ERD in the left temporo-parietal area, which is the primary feature of complex problem solving. Alpha ERS (associated with cortical idle) appears in the right centro-parietal area by cathodal-tDCS. Based on such observations, it is believed that the alpha ERD on the left temporo-parietal area may be an evidence of tDCS-induced brain activity changes during the task.

*This work was supported by the NRF of Korea (2016R1A2B4010897) and GRI grant funded by the GIST in 2017.

Dry electrodes can improve the viability of electroencephalography (EEG) acquisitions, contributing to an increased use in established and emerging fields of application for EEG. We recently developed and successfully validated a novel pol-ymer-based flexible dry multipin electrode. Dry electrodes rely on a direct, stable contact to the scalp and therefore pose specific requirements for the used cap system. Our investigations emphasize the need for adaptive caps and electrode de-signs ensuring easy, comfortable application and reliable adduction. We propose a novel modular, adaptable cap system for rapid EEG, implementing the specific requirements of dry electrodes.

Our proposed cap system is based on two independent components: a headband and a butterfly-shaped central cap mod-ule. The headband enables adaption to the individual head circumference, integrating frontal, temporal and occipital elec-trodes. The central module enables adaptation to sagittal and coronal width as well as height of the head, carrying central and parietal electrodes. The modules are based on a semi-flexible, washable textile. 64 electrodes are arranged in a quasi-equidistant layout to allow application of state-of-the art methods for artifact removal and source localization.

We present our assessment of electrode and cap requirements, the cap design and preliminary results of the concept val-idation. For our proposed electrode shape, we’ve identified an optimal adduction force between 2-3 N. Adapted pin height at frontal and frontal-temporal areas contribute to improved wearing comfort. The cap system enables rapid application with preparation times below 10 min. The separate modules enable adaptation of the cap to different head shapes.

Social norms are crucial to any society’s functioning, but social norm compliance is characterized by heterogeneous types of behavior. Here, we examined neural traits to uncover the sources of behavioral types in compliance with the fairness norm. Participants played a distribution game in which they decided how much money to share with an anonymous partner in two conditions; i) no consequences for an unfair offer, ii) possible sanctions for an unfair offer. Cluster analyses revealed distinct types of people: voluntary compliers, who follow the fairness norm of an even split in both conditions, sanction-based compliers, who follow the fairness norm only in the possible-sanctions condition but share little in the no-consequences condition, and non-compliers, who never follow the fairness norm and share little in both conditions. Source-localization analyses of resting-state EEG revealed that voluntary compliers are characterized by lower baseline delta activity in the right TPJ, compared to the two other types. Sanction-based compliers are characterized by higher baseline beta3 activity in the DLPFC, compared to non-compliers. These findings are the first demonstration of the sources of three heterogeneous types in social norm compliance. Discussion focuses on the potential roles of the TPJ and DLPFC.

Fatigue may cause a decrease in mental and physical performance capacity, which is a serious safety risk for the drivers in the transportation system. How to visualize and detect the potential implicit risk already draws common concern. Here, we proposed a new spatio-temporal interaction approach for analyzing electroencephalpgram (EEG) signals during driver fatigue to maintain the reliability of fatigue-related information in different dimensions. The methodology began with construction of functional brain network to integrate functional interactions between different brain regions. With clustering on the brain network, the essential part of the spatial information was retained on fewer network’s nodes to reduce the fluctuations in spatial dimension. Subsequently a nonlinear parameter, wavelet entropy, was computed within a sliding window from the selected nodes to extract the temporal features. Finally, the drivers’ spatio-temporal matrices were created to find the fatigue patterns. The experimental results demonstrated that rhythmic alpha activity spread from the frontal node to parietal node and presented a gradient distribution feature (across frontal, central, and parietal regions) in space domain, though its energy fluctuation occurred with the accumulation of fatigue. In time domain, the features from the selected spatial loci of EEG had a certain synchronized decreasing trend, which revealed brain activity complexity reducing. In addition, most inflection points (81.25%) gained from the temporal features can match the points marked by the subjects, which showed the method’s potential value for the analysis of fatigue mechanism and the application of fatigue detection.

Artefactual latency jitter in ERP study: a simulation study

Gamma band oscillation during Sternberg Task

The Dual Mechanisms of Control theory offers to divide cognitive control in two main strategies: proactive and reactive control (Braver et al., 2007). Proactive control is defined as the ability to anticipate and maintain in working memory information necessary to succeed a task, while reactive control is representative of inhibitory mechanisms triggered by an external cue. The AX-CPT context processing task has become a popular paradigm to examine the use of proactive and reactive control strategies. However, the neurophysiological correlates, such as Event-Related Potentials (ERP’s), of cognitive control strategies in the AX-CPT task remain understudied.

This explorative study investigated the ERPs of 100 young and healthy participants confronted to an AX-CPT task. Each subject filled multiple questionnaires assessing personality trait (impulsivity), clinical traits (anxiety, depression, addiction) and eventual medical and psychiatric history along with a neuropsychological testing assessing working memory, sustained attention and inhibition.

Preliminary results on the respective influences of clinical and cognitive variables on both performance and ERPs parameters of proactive and reactive control will be presented.

Sleep spindles are hallmark of electrophysiological activity in the non-rapid eye movement (NREM) sleep stage 2. Numerous studies about spindles have been reported so far and several functional roles of spindles have been identified. However, sleep spindle identification has been commonly performed by expert’s visual inspection and it is quite time consuming. Thus, various automatic spindle detectors have been proposed, however, it is hard to expect stable performance of detectors because of subject-varying characteristics of electroencephalography (EEG). Here, we applied a popular automatic spindle detector to a sleep spindle database to verify a necessity of subject-specific optimization. The spindle detector is based on a constant threshold scheme, which selects the threshold from a distribution of the root mean square (RMS) values of the NREM sleep EEG epochs. We considered window size and step of epochs as optimization parameters as well as a percentile of RMS distribution for threshold selection. We selected four subjects (among eight) who have reasonable number of spindles in the database. For two subjects, significant differences in detecting performance between the detector with typical parameter and the subject-optimized detector in two subjects were found (for example, F1-score changed from 0.32±0.02 to 0.36±0.03 with p-value<0.000001). With these preliminary results, subject-specific optimization scheme for automatic spindle detectors may be compelling. Further investigation on real-time spindle detector for sleep modulation study is under way.

*This work was supported by IITP grant funded by Korea government (No. 2017-0-00451), GRI grant funded by the GIST in 2017, and NRF of Korea (2016R1A2B4010897).

A deqi is known as a phenomenon that occurs by acupuncture stimulation, and play an important role in the acupuncture treatments. However, there is no certain evidence on the mechanism and expression of deqi. In our previous studies, we find out that temperature and touch sensations are related to deqi sensation. Here, in this study, we focused on correlation between individual sensitivity of those sensations through quantitative biomarkers via analyzing electroencephalogram (EEG).

The number of subjects of clinical trials was 30, half male and half female. 64 channels of EEG are measured simultaneously while measuring the sensitivities of cold, hot, touch, and deqi. The analyses of the data are performed in three way: 1) correlation between sensitivity of sensations; 2) changes of ratio of sensorimotor rhythm wave (SMR; 13~15Hz frequency band), which is known as related feature of EEG for sensations; 3) coherence analysis among frontal, temporal, occipital, and parietal lobes.

As results, correlation between temperature sensitivity, especially the cold, and deqi sensitivities, is appeared in all three analyses. There were high correlation between temperature and deqi sensitivities. Same patterns in relative power of sensorimotor rhythm wave and connectivity trends of SMR coherence at part of channels in parietal lobe are appeared in groups of cold sensitive and deqi sensitive.

This study aims to check the impact of acupuncture manipulation caused by breathing on the peripheral nervous system, using a somatosensory evoked potential (SEP) measurement method in EEG study. The subjects were 7 healthy men and women aged 19-35. For the purposes of the experiment, which was conducted using deep respiration with a two-week washout period, the subjects were divided into three groups, i.e. those subject to Bre-acupuncture manipulation (BAM), those subject to acupuncture manipulation (AM) without regard to respiration, and those subject to only deep breathing (DB), and were subjected to acupuncture manipulation of acupuncture point LI4 ten times. SEP measurement was carried out for 5 minutes, applying 2 HZ-pulse electrical stimulation to the thumb and the little finger of the hand acupunctured with 64-channel EEG, alternately. The verbal rating scaling system (VRS) was used in each experiment before and after acupuncture manipulation. The measurement results show evoked potential amplitude in the range of 40-60ms in the order of BAM<AM<DB, and that the difference in VRS before and after acupuncture stimulation was found in the order of BAM>AM>DB. This shows that acupuncture manipulation increases the pain threshold in the nervous system and that it is more effective when performed simultaneously with deep respiration-based control of the autonomic nervous system. The results of this study indicate that acupuncture-based treatment can produce a higher effect by using a patient’s breathing.

FMRI studies have shown that large-scale functional networks are inherently active in the brain at rest. Several distinct resting-state networks (RSNs) have been attributed to different functional states, and shown to be non-stationary in time, but partitioned into stable epochs (Zalesky, 2014). Periods of stable activity have also been robustly described in EEG recordings at rest, albeit on a faster temporal scale (~100ms). They appear recurrently in a reduced number of quasi-stable electrical field topographies, called microstates (Lehmann, 1980). A recent EEG-fMRI study has identified a relation between EEG microstates and fMRI RSNs – namely auditory, visual, salience and attention networks (Britz, 2010). The next step towards a more direct demonstration that different microstates reflect different functional (mental) states would be to show that well-defined cognitive tasks specifically modulate certain states, both in EEG and fMRI. In this pilot study, we examine whether distinct cognitive tasks can differentially manipulate specific EEG microstates and fMRI RSNs. A group of healthy subjects underwent high-density EEG and 7T-fMRI in three distinct paradigms: eyes-closed rest (6min), autobiographical memory retrieval induced by images of personal past episodes (15min), and mental arithmetics (serial subtraction, 15min). Consistently across subjects, EEG microstate analysis revealed task-specific alterations of microstates C and D, previously associated with the salience and attention RSNs, respectively. These results support the possibility of modulating specific electric and hemodynamic RSNs by appropriate cognitive tasks, corroborating their correspondence to specific mental states. Resting-state analysis of the fMRI data will be further performed to confirm this association.

Auditory mismatch negativity (MMN) is believed to be an electrophysiological marker of pre-attentive cognitive processing and it´s deficit is suggested as schizophrenia endophenotype. MMN is usually registered in paradigms with tones or sounds but paradigms using more complex and natural auditory conditions i.e. modified speech or deviant phonemes, are less common.

In our study EEG data were collected from 17 healthy volunteers and 17 schizophrenia patients. 300 five-to-seven-word sentences in Czech language were presented to subjects by headphones. A phonetic "error" (replacement of a diphone in a word by a diphone from another word) appeared randomly in 150 sentences. Subject´s attention was distracted by watching a mute video.

We observed MMN approximately 200ms after "error stimulus" with it´s maximum in fronto-central electrodes in both groups. Schizophrenia patients showed reduced MMN compared to healthy volunteers.

There are two main results of this study. First, we validated that MMN can be reliably elicited in a paradigm based on natural conditions that the human speech is. Second, we showed that this method is sensitive to the pre-attentive cognitive processing deficit in schizophrenia.

This work is supported by Ministry of Health of the Czech Republic, grant nr. 15-29900A, ED2.1.00/03.0078, LO1611/NPU I, MICR VI20172020056, MH CZ - DRO (NIMH-CZ, 00023752) and PROGRES Q35.

An important process underlying meditation and its benefits involves the regulation of attention. Although the two main meditation categories – open-monitoring meditation (OMM) and focused-attention meditation (FAM) – are associated with different benefits and attentional processes, direct comparisons between the attentional neural mechanism of FAM and OMM are rare. This study uses magnetoencephalography (MEG) recordings in 12 expert meditators to compare FAM and OMM by assessing (i) source spectral power, (ii) seed-based functional connectivity of key regions in attention, (including anterior cingulate cortex, dorsolateral prefrontal cortex and the thalamus) and (iii) graph theory metrics that describe brain-wide efficiency of information processing. We reconstructed the source space using minimum norm estimate and computed spectral power and functional connectivity in multiple frequency bands (delta, theta, alpha, beta, gamma) using a custom-designed python-based MEG analysis pipeline (NeuroPycon). The results reveal unique patterns of neural processes specific to FAM or OMM. Among other things, compared to FAM, OMM appears to be characterized by enhanced small-world network properties. By contrast, FAM exhibits greater functional connectivity between the anterior cingulate cortex and frontal regions. These findings shed light onto the mechanisms that potentially mediate the different behavioral and attentional capacities associated with each of the two meditation techniques. Our results are discussed in the context of previous behavioral and fMRI studies on meditation and attention.

Here we present pilot data where we administered tACS (20Hz and 140Hz) and tRNS over the primary motor cortex of healthy children/adolescents and adults to observe possible influences on the corticospinal excitability. Based on previous studies (Minhas et al, 2012; Moliadze et al, 2015) we hypothesize that the effects of tACS will differ in pediatric population compared to adults.

Methods: 20Hz and 140Hz tACS, tRNS and sham stimulation with 1 mA were applied for 10 minutes on the left M1HAND (mean/SD age: children and adolescents 13,1 ± 2,56; adults 25,23 ± 3,47) in a randomized order. Electrical stimulation was delivered by a battery driven stimulator through conductive-rubber electrodes. MEP were measured by TMS before and after the stimulation (0, 30 and 60 minutes).

Results/Conclusion: In all subjects electrical stimulation was well tolerated.

In both groups 1mA tRNS as well as 140Hz tACS resulted in a significant increase of MEP amplitudes compared with baseline recordings and sham stimulation. However, the increment at tRNS starts with a delay of 30 minutes post stimulation in children.

Interestingly, 20Hz tACS leads to a slightly continuing increase of MEP in adults starting 30 minutes after stimulation which is contrary to the results of other studies (Cappon et al, 2016; Wach et al, 2013). In children 20Hz tACS seems to have an inhibitory tendency after one hour.

Based on our preliminary results the electrical stimulation protocols have to be optimized according to age by planning studies in pediatric population.

Anxiety is a common emotion, but when experienced beyond a certain threshold, it may symptomatically compromise cognitive and functional performance. The LIFECOURSE ERC project establishes a multilevel developmental framework utilizing EEG in this context to examine the influence of stress on diverse physiological, emotional, and behavioural outcomes among adolescents. In the reported pilot investigation, we examine the potential interplay between anxiety and cognitive neurodynamics defined by EEG.

University students between the ages of 22 and 29 years (n=32) underwent anxiety assessment defined by emotive and semantic feedback in a designed psychological questionnaire. EEG was acquired, using a 32-channel wet electrode cap, over the duration of several neurocognitive tasks: object naming, mental rotation, antonym generation, and spatial frequency assessment. A Matlab Graphical User Interface (GUI) was developed order to perform statistical assessment of event-related potential (ERP) and power spectral analyses. The GUI was designed to handle both EEG (*.cnt) and spectral files (*.asc). The program likewise allowed for comparative assessment within a single subject, between two different subjects, or against the entire iteratively-updated cohort. Differences in task-related regional hemispheric alpha asymmetry patterns were found between subjects with highest and lowest anxiety levels. This was most evident in left temporal and occipital electrodes. Likewise, changes in central midline delta, along with frontal, central, and occipital theta frequencies were identified. Finally, in the beta band, differences were primarily evidenced in frontal, right occipital, and central midline areas. Altogether, these changes suggest differences in language, motor, and somatosensory cortical activity.

Preclincial work suggests that sleep promotes global downscaling of overall synaptic strength (homeostatic plasticity) and the consolidation of long-term potentiation (LTP) of task-specific synapses (associative plasticity). Here we use electroencephalography (EEG) and transcranial magnetic stimulation (TMS) as non-invasive indices of homeostatic and associative synaptic plasticity in healthy humans before and after brief periods of daytime sleep and wakefulness (repeated measures sleep laboratory study, 14 healthy participants, 5 females, 9 males, 18–30 years). We demonstrate indices of decreased overall synaptic strength (wake EEG theta activity) and increased input-specific synaptic strength (paired associative stimulation (PAS) induced changes in motor-evoked potentials) after sleep compared to wakefulness. The increase in input-specific synaptic strength was positively correlated with EEG slow wave activity (1-4 Hz) over the respective motorcortical area (M1). Our study supports the notion that slow wave sleep orchestrates homeostatic and associative synaptic plasticity, believed to be the neural basis for adaptive behavior, in humans.