Ventrointermediate thalamic stimulation improves motor learning in humans.

Ventrointermediate thalamic stimulation (VIM-DBS) modulates oscillatory activity in a cortical network including primary motor cortex, premotor cortex, and parietal cortex. Here we show that, beyond the beneficial effects of VIM-DBS on motor execution, this form of invasive stimulation facilitates production of sequential finger movements that follow a repeated sequence. These results highlight the role of thalamo-cortical activity in motor learning.

A toolbox for decoding BCI commands based on event-related potentials.

Commands in brain-computer interface (BCI) applications often rely on the decoding of event-related potentials (ERP). For instance, the P300 potential is frequently used as a marker of attention to an oddball event. Error-related potentials and the N2pc signal are further examples of ERPs used for BCI control. One challenge in decoding brain activity from the electroencephalogram (EEG) is the selection of the most suitable channels and appropriate features for a particular classification approach. Here we introduce a toolbox that enables ERP-based decoding using the full set of channels, while automatically extracting informative components from relevant channels. The strength of our approach is that it handles sequences of stimuli that encode multiple items using binary classification, such as target vs. nontarget events typically used in ERP-based spellers. We demonstrate examples of application scenarios and evaluate the performance of four openly available datasets: a P300-based matrix speller, a P300-based rapid serial visual presentation (RSVP) speller, a binary BCI based on the N2pc, and a dataset capturing error potentials. We show that our approach achieves performances comparable to those in the original papers, with the advantage that only conventional preprocessing is required by the user, while channel weighting and decoding algorithms are internally performed. Thus, we provide a tool to reliably decode ERPs for BCI use with minimal programming requirements.

Seeing and extrapolating motion trajectories share common informative activation patterns in primary visual cortex.

The natural environment is dynamic and moving objects become constantly occluded, engaging the brain in a challenging completion process to estimate where and when the object might reappear. Although motion extrapolation is critical in daily life-imagine crossing the street while an approaching car is occluded by a larger standing vehicle-its neural underpinnings are still not well understood. While the engagement of low-level visual cortex during dynamic occlusion has been postulated, most of the previous group-level fMRI-studies failed to find evidence for an involvement of low-level visual areas during occlusion. In this fMRI-study, we therefore used individually defined retinotopic maps and multivariate pattern analysis to characterize the neural basis of visible and occluded changes in motion direction in humans. To this end, participants learned velocity-direction change pairings (slow motion-upwards; fast motion-downwards or vice versa) during a training phase without occlusion and judged the change in stimulus direction, based on its velocity, during a following test phase with occlusion. We find that occluded motion direction can be predicted from the activity patterns during visible motion within low-level visual areas, supporting the notion of a mental representation of motion trajectory in these regions during occlusion.

Suppression of Motor Sequence Learning and Execution Through Anodal Cerebellar Transcranial Electrical Stimulation.

Cerebellum (CB) and primary motor cortex (M1) have been associated with motor learning, with different putative roles. Modulation of task performance through application of transcranial direct current stimulation (TDCS) to brain structures provides causal evidence for their engagement in the task. Studies evaluating and comparing TDCS to these structures have provided conflicting results, however, likely due to varying paradigms and stimulation parameters. Here we applied TDCS to CB and M1 within the same experimental design, to enable direct comparison of their roles in motor sequence learning. We examined the effects of anodal TDCS during motor sequence learning in 60 healthy participants, randomly allocated to CB-TDCS, M1-TDCS, or Sham stimulation groups during a serial reaction time task. Key to the design was an equal number of repeated and random sequences. Reaction times (RTs) to implicitly learned and random sequences were compared between groups using ANOVAs and post hoc t-tests. A speed-accuracy trade-off was excluded by analogous analysis of accuracy scores. An interaction was observed between whether responses were to learned or random sequences and the stimulation group. Post hoc analyses revealed a preferential slowing of RTs to implicitly learned sequences in the group receiving CB-TDCS. Our findings provide evidence that CB function can be modulated through transcranial application of a weak electrical current, that the CB and M1 cortex perform separable functions in the task, and that the CB plays a specific role in motor sequence learning during implicit motor sequence learning.

Repetitive Anodal TDCS to the Frontal Cortex Increases the P300 during Working Memory Processing.

Transcranial direct current stimulation (TDCS) is a technique with which neuronal activity, and therefore potentially behavior, is modulated by applying weak electrical currents to the scalp. Application of TDCS to enhance working memory (WM) has shown promising but also contradictory results, and little emphasis has been placed on repeated stimulation protocols, in which effects are expected to be increased. We aimed to characterize potential behavioral and electrophysiological changes induced by TDCS during WM training and evaluate whether repetitive anodal TDCS has a greater modulatory impact on the processes underpinning WM than single-session stimulation. We examined the effects of single-session and repetitive anodal TDCS to the dorsolateral prefrontal cortex (DLPFC), targeting the frontal-parietal network, during a WM task in 20 healthy participants. TDCS had no significant impact on behavioral measures, including reaction time and accuracy. Analyzing the electrophysiological response, the P300 amplitude significantly increased following repetitive anodal TDCS, however, positively correlating with task performance. P300 changes were identified over the parietal cortex, which is known to engage with the frontal cortex during WM processing. These findings support the hypothesis that repetitive anodal TDCS modulates electrophysiological processes underlying WM.

Neurological Outpatients Prefer EEG Home-Monitoring over Inpatient Monitoring-An Analysis Based on the UTAUT Model.

Home monitoring examinations offer diagnostic and economic advantages compared to inpatient monitoring. In addition, these technical solutions support the preservation of health care in rural areas in the absence of local care providers. The acceptance of patients is crucial for the implementation of home monitoring concepts. The present research assesses the preference for a health service that is to be introduced, namely an EEG home-monitoring of neurological outpatients-using a mobile, dry-electrode EEG (electroencephalography) system-in comparison to the traditional long-time EEG examination in a hospital. Results of a representative study for Germany (n = 421) reveal a preference for home monitoring. Importantly, this preference is partially driven by a video explaining the home monitoring system. We subsequently analyzed factors that influence the behavioral intention (BI) to use the new EEG system, drawing on an extended Unified Theory of Acceptance and Use of Technology (UTAUT) model. The strongest positive predictor of BI is the belief that EEG home-monitoring will improve health quality, while computer anxiety and effort expectancy represent the strongest barriers. Furthermore, we find the UTAUT model's behavioral intention construct to predict the patients' decision for or against home monitoring more strongly than any other patient's characteristic such as gender, health condition, or age, underlying the model's usefulness.

Feasibility assessment of patient-controlled EEG home-monitoring: More results from the HOMEONE study.

OBJECTIVE: The feasibility phase of the HOME (Home-Monitoing and Education) project aims to show the practical feasibility of Electroencephalography (EEG)home-monitoring using a patient-controlled mobile system. Its objective is to assess the potential diagnostic and therapeutic yields of home-monitoring compared to conventional healthcare. METHODS: 16 office-based practitioners chose 97 patients and recorded standard 20-minute EEGs using conventional recorders. After training, the same patients used a patient-controlled mobile dry electrode EEG system in their home environment. The practitioners in charge and two additional raters assessed all recordings. We conducted inter-rater and intra-rater comparisons between the diagnostic findings. RESULTS: 89 patients successfully conducted home-monitoring recordings. The intra-rater comparison results for the diagnostic findings of the conventional recordings and the patient-made recordings show a fair Cohen's kappa value (0.21). Additionally, we documented a change of patient management in 9 cases. CONCLUSIONS: The feasibility of EEG home-monitoring using a patient-controlled device is confirmed. The yield of EEG home-monitoring comprises information that can influence patient management. SIGNIFICANCE: Patient-controlled EEG home-monitoring is feasible as part of routine care for neurological outpatients as its technical efficacy and practical feasibility are shown and significantly positive effects on patient management are evidenced.

Anticipating social incentives recruits alpha-beta oscillations in the human substantia nigra and invigorates behavior across the life span.

Anticipating social and non-social incentives recruits shared brain structures and promotes behavior. However, little is known about possible age-related behavioral changes, and how the human substantia nigra (SN) signals positive and negative social information. Therefore, we recorded intracranial electroencephalography (iEEG) from the SN of Parkinson's Disease (PD) patients (n = 12, intraoperative, OFF medication) in combination with a social incentive delay task including photos of neutral, positive or negative human gestures and mimics as feedback. We also tested a group of non-operated PD patients (n = 24, ON and OFF medication), and a sample of healthy young (n = 51) and older (n = 52) adults with behavioral readouts only. Behaviorally, the anticipation of both positive and negative social feedback equally accelerated response times in contrast to neutral social feedback in healthy young and older adults. Although this effect was not significant in the group of operated PD patients - most likely due to the small sample size - iEEG recordings in their SN showed a significant increase in alpha-beta power (9-20 Hz) from 300 to 600 ms after cue onset again for both positive and negative cues. Finally, in non-operated PD patients, the behavioral effect was not modulated by medication status (ON vs OFF medication) suggesting that other processes than dopaminergic neuromodulation play a role in driving invigoration by social incentives. Together, our findings provide novel and direct evidence for a role of the SN in processing positive and negative social information via specific oscillatory mechanisms in the alpha-beta range, and they suggest that anticipating social value in simple cue-outcome associations is intact in healthy aging and PD.

Multiscale criticality measures as general-purpose gauges of proper brain function.

The brain is universally regarded as a system for processing information. If so, any behavioral or cognitive dysfunction should lend itself to depiction in terms of information processing deficiencies. Information is characterized by recursive, hierarchical complexity. The brain accommodates this complexity by a hierarchy of large/slow and small/fast spatiotemporal loops of activity. Thus, successful information processing hinges upon tightly regulating the spatiotemporal makeup of activity, to optimally match the underlying multiscale delay structure of such hierarchical networks. Reduced capacity for information processing will then be expressed as deviance from this requisite multiscale character of spatiotemporal activity. This deviance is captured by a general family of multiscale criticality measures (MsCr). MsCr measures reflect the behavior of conventional criticality measures (such as the branching parameter) across temporal scale. We applied MsCr to MEG and EEG data in several telling degraded information processing scenarios. Consistently with our previous modeling work, MsCr measures systematically varied with information processing capacity: MsCr fingerprints showed deviance in the four states of compromised information processing examined in this study, disorders of consciousness, mild cognitive impairment, schizophrenia and even during pre-ictal activity. MsCr measures might thus be able to serve as general gauges of information processing capacity and, therefore, as normative measures of brain health.

Mind-wandering Is Accompanied by Both Local Sleep and Enhanced Processes of Spatial Attention Allocation.

Mind-wandering (MW) is a subjective, cognitive phenomenon, in which thoughts move away from the task toward an internal train of thoughts, possibly during phases of neuronal sleep-like activity (local sleep, LS). MW decreases cortical processing of external stimuli and is assumed to decouple attention from the external world. Here, we directly tested how indicators of LS, cortical processing, and attentional selection change in a pop-out visual search task during phases of MW. Participants' brain activity was recorded using magnetoencephalography, MW was assessed via self-report using randomly interspersed probes. As expected, the performance decreased under MW. Consistent with the occurrence of LS, MW was accompanied by a decrease in high-frequency activity (HFA, 80-150 Hz) and an increase in slow wave activity (SWA, 1-6 Hz). In contrast, visual attentional selection as indexed by the N2pc component was enhanced during MW with the N2pc amplitude being directly linked to participants' performance. This observation clearly contradicts accounts of attentional decoupling that would predict a decrease in attention-related responses to external stimuli during MW. Together, our results suggest that MW occurs during phases of LS with processes of attentional target selection being upregulated, potentially to compensate for the mental distraction during MW.

The role of the anterior nuclei of the thalamus in human memory processing.

Extensive neuroanatomical connectivity between the anterior thalamic nuclei (ATN) and hippocampus and neocortex renders them well-placed for a role in memory processing, and animal, lesion, and neuroimaging studies support such a notion. The deep location and small size of the ATN have precluded their real-time electrophysiological investigation during human memory processing. However, ATN electrophysiological recordings from patients receiving electrodes implanted for deep brain stimulation for pharmacoresistant focal epilepsy have enabled high temporal resolution study of ATN activity. Theta frequency synchronization of ATN and neocortical oscillations during successful memory encoding, enhanced phase alignment, and coupling between ATN local gamma frequency activity and frontal neocortical and ATN theta oscillations provide evidence of an active role for the ATN in memory encoding, potentially integrating information from widespread neocortical sources. Greater coupling of a broader gamma frequency range with theta oscillations at rest than during memory encoding provides additional support for the hypothesis that the ATN play a role in selecting local, task-relevant high frequency activity associated with particular features of a memory trace.

Frontotemporal Regulation of Subjective Value to Suppress Impulsivity in Intertemporal Choices.

Impulsive decisions arise from preferring smaller but sooner rewards compared with larger but later rewards. How neural activity and attention to choice alternatives contribute to reward decisions during temporal discounting is not clear. Here we probed (1) attention to and (2) neural representation of delay and reward information in humans (both sexes) engaged in choices. We studied behavioral and frequency-specific dynamics supporting impulsive decisions on a fine-grained temporal scale using eye tracking and MEG recordings. In one condition, participants had to decide for themselves but pretended to decide for their best friend in a second prosocial condition, which required perspective taking. Hence, conditions varied in the value for themselves versus that pretending to choose for another person. Stronger impulsivity was reliably found across three independent groups for prosocial decisions. Eye tracking revealed a systematic shift of attention from the delay to the reward information and differences in eye tracking between conditions predicted differences in discounting. High-frequency activity (175-250 Hz) distributed over right frontotemporal sensors correlated with delay and reward information in consecutive temporal intervals for high value decisions for oneself but not the friend. Collectively, the results imply that the high-frequency activity recorded over frontotemporal MEG sensors plays a critical role in choice option integration.SIGNIFICANCE STATEMENT Humans face decisions between sooner smaller rewards and larger later rewards daily. An objective benefit of losing weight over a longer time might be devalued in face of ice cream because they prefer currently available options because of insufficiently considering long-term alternatives. The degree of contribution of neural representation and attention to choice alternatives is not clear. We investigated correlates of such decisions in participants deciding for themselves or pretending to choose for a friend. Behaviorally participants discounted less in self-choices compared with the prosocial condition. Eye movement and MEG recordings revealed how participants represent choice options most evident for options with high subjective value. These results advance our understanding of neural mechanisms underlying decision-making in humans.

Impact of Stimulus Features on the Performance of a Gaze-Independent Brain-Computer Interface Based on Covert Spatial Attention Shifts.

Regaining communication abilities in patients who are unable to speak or move is one of the main goals in decoding brain waves for brain-computer interface (BCI) control. Many BCI approaches designed for communication rely on attention to visual stimuli, commonly applying an oddball paradigm, and require both eye movements and adequate visual acuity. These abilities may, however, be absent in patients who depend on BCI communication. We have therefore developed a response-based communication BCI, which is independent of gaze shifts but utilizes covert shifts of attention to the left or right visual field. We recorded the electroencephalogram (EEG) from 29 channels and coregistered the vertical and horizontal electrooculogram. Data-driven decoding of small attention-based differences between the hemispheres, also known as N2pc, was performed using 14 posterior channels, which are expected to reflect correlates of visual spatial attention. Eighteen healthy participants responded to 120 statements by covertly directing attention to one of two colored symbols (green and red crosses for "yes" and "no," respectively), presented in the user's left and right visual field, respectively, while maintaining central gaze fixation. On average across participants, 88.5% (std: 7.8%) of responses were correctly decoded online. In order to investigate the potential influence of stimulus features on accuracy, we presented the symbols with different visual angles, by altering symbol size and eccentricity. The offline analysis revealed that stimulus features have a minimal impact on the controllability of the BCI. Hence, we show with our novel approach that spatial attention to a colored symbol is a robust method with which to control a BCI, which has the potential to support severely paralyzed people with impaired eye movements and low visual acuity in communicating with their environment.

Early Shift of Attention Is Not Regulated by Mind Wandering in Visual Search.

Unique to humans is the ability to report subjective awareness of a broad repertoire of external and internal events. Even when asked to focus on external information, the human's mind repeatedly wanders to task-unrelated thoughts, which limits reading comprehension or the ability to withhold automated manual responses. This led to the attentional decoupling account of mind wandering (MW). However, manual responses are not an ideal parameter to study attentional decoupling, given that during MW, the online adjustment of manual motor responses is impaired. Hence, whether early attentional mechanisms are indeed downregulated during MW or only motor responses being slowed is not clear. In contrast to manual motor responses, eye movements are considered a sensitive proxy of attentional shifts. Using a simple target detection task, we asked subjects to indicate whether a target was presented within a visual search display by pressing a button while we recorded eye movements and unpredictably asked the subjects to rate their actual level of MW. Generally, manual reaction times increased with MW, both in target absent and present trials. But importantly, even in trials with MW, subjects detected earlier a presented than an absent target. The decoupling account would predict more fixations of the target before pressing the button during MW. However, our results did not corroborate this assumption. Most importantly, subject's time to direct gaze at the target was equally fast in trials with and without MW. Our results corroborate our hypothesis that during MW early, bottom-up driven attentional processes are not decoupled but selectively manual motor responses are slowed.

Decoding the covert shift of spatial attention from electroencephalographic signals permits reliable control of a brain-computer interface.

OBJECTIVE: One of the main goals of brain-computer interfaces (BCI) is to restore communication abilities in patients. BCIs often use event-related potentials (ERPs) like the P300 which signals the presence of a target in a stream of stimuli. The P300 and related approaches, however, are inherently limited, as they require many stimulus presentations to obtain a usable control signal. Many approaches depend on gaze direction to focus the target, which is also not a viable approach in many cases, because eye movements might be impaired in potential users. Here we report on a BCI that avoids both shortcomings by decoding spatial target information, independent of gaze shifts. APPROACH: We present a new method to decode from the electroencephalogram (EEG) covert shifts of attention to one out of four targets simultaneously presented in the left and right visual field. The task is designed to evoke the N2pc component-a hemisphere lateralized response, elicited over the occipital scalp contralateral to the attended target. The decoding approach involves decoding of the N2pc based on data-driven estimation of spatial filters and a correlation measure. MAIN RESULTS: Despite variability of decoding performance across subjects, 22 out of 24 subjects performed well above chance level. Six subjects even exceeded 80% (cross-validated: 89%) correct predictions in a four-class discrimination task. Hence, the single-trial N2pc proves to be a component that allows for reliable BCI control. An offline analysis of the EEG data with respect to their dependence on stimulation time and number of classes demonstrates that the present method is also a workable approach for two-class tasks. SIGNIFICANCE: Our method extends the range of strategies for gaze-independent BCI control. The proposed decoding approach has the potential to be efficient in similar applications intended to decode ERPs.

Self-regulated critical brain dynamics originate from high frequency-band activity in the MEG.

Brain function requires the flexible coordination of billions of neurons across multiple scales. This could be achieved by scale-free, critical dynamics balanced at the edge of order and disorder. Criticality has been demonstrated in several, often reduced neurophysiological model systems. In the intact human brain criticality has yet been only verified for the resting state. A more direct link between the concept of criticality and oscillatory brain physiology, which is strongly related to cognition, is yet missing. In the present study we therefore carried out a frequency-specific analysis of criticality in the MEG, recorded while subjects were in a defined cognitive state through mindfulness meditation. In a two-step approach we assessed whether the macroscopic neural avalanche dynamics is scale-free by evaluating the goodness of a power-law fits of cascade size and duration distributions of MEG deflections in different frequency bands. In a second step we determined the closeness of the power-law exponents to a critical value of -1.5. Power-law fitting was evaluated by permutation testing, fitting of alternative distributions, and cascade shape analysis. Criticality was verified by defined relationships of exponents of cascade size and duration distributions. Behavioral relevance of criticality was tested by correlation of indices of criticality with individual scores of the Mindful Attention Awareness Scale. We found that relevant scale-free near-critical dynamics originated only from broad-band high-frequency (> 100 Hz) MEG activity, which has been associated with action potential firing, and therefore links criticality on the macroscopic level of MEG to critical spike avalanches on a microscopic level. Whereas a scale-free dynamics was found under mindfulness meditation and rest, avalanche dynamics shifted towards a critical point during meditation by reduction of neural noise. Together with our finding that during mindfulness meditation avalanches show differences in topography relative to rest, our results show that self-regulated attention as required during meditation can serve as a control parameter of criticality in scale-free brain dynamics.

Comparison between a wireless dry electrode EEG system with a conventional wired wet electrode EEG system for clinical applications.

Dry electrode electroencephalogram (EEG) recording combined with wireless data transmission offers an alternative tool to conventional wet electrode EEG systems. However, the question remains whether the signal quality of dry electrode recordings is comparable to wet electrode recordings in the clinical context. We recorded the resting state EEG (rsEEG), the visual evoked potentials (VEP) and the visual P300 (P3) from 16 healthy subjects (age range: 26-79 years) and 16 neurological patients who reported subjective memory impairment (age range: 50-83 years). Each subject took part in two recordings on different days, one with 19 dry electrodes and another with 19 wet electrodes. They reported their preferred EEG system. Comparisons of the rsEEG recordings were conducted qualitatively by independent visual evaluation by two neurologists blinded to the EEG system used and quantitatively by spectral analysis of the rsEEG. The P100 visual evoked potential (VEP) and P3 event-related potential (ERP) were compared in terms of latency, amplitude and pre-stimulus noise. The majority of subjects preferred the dry electrode headset. Both neurologists reported that all rsEEG traces were comparable between the wet and dry electrode headsets. Absolute Alpha and Beta power during rest did not statistically differ between the two EEG systems (p > 0.05 in all cases). However, Theta and Delta power was slightly higher with the dry electrodes (p = 0.0004 for Theta and p < 0.0001 for Delta). For ERPs, the mean latencies and amplitudes of the P100 VEP and P3 ERP showed comparable values (p > 0.10 in all cases) with a similar spatial distribution for both wet and dry electrode systems. These results suggest that the signal quality, ease of set-up and portability of the dry electrode EEG headset used in our study comply with the needs of clinical applications.

Deep brain stimulation of the nucleus basalis of Meynert attenuates early EEG components associated with defective sensory gating in patients with Alzheimer disease - a two-case study.

Alzheimer's disease (AD) is associated with deterioration of memory and cognitive function and a degeneration of neurons of the nucleus basalis of Meynert (NBM). The NBM is the major input source of acetylcholine (ACh) to the cortex. The decreasing cholinergic innervation of the cortex due to degeneration of the NBM might be the cause of loss of memory function. NBM-Deep brain stimulation (NBM-DBS) is considered to serve as a potential therapeutic option for patients with AD by supporting residual cholinergic transmission to stabilize oscillatory activity in memory-relevant circuits. However, whether DBS could improve sensory memory functions in patients with AD is not clear. Here, in a passive auditory oddball paradigm, patients with AD (N = 2) listened to repetitive background tones (standard tones) randomly interrupted by frequency deviants in two blocks with NBM-DBS OFF and then NBM-DBS ON, while age-matched healthy controls (N = 6) repeated the experiment twice. The mismatch negativity in NBM-DBS OFF significantly differed from controls in both blocks, but not under NBM-DBS, which was likely due to a pronounced P50 increase overlapping with the N1 in NBM-DBS OFF. This early complex of EEG components recovered under stimulation to a normal level as defined by responses in controls. In this temporal interval, we found in patients with NBM-DBS ON (but not with NBM-DBS OFF) and in controls a strong repetition suppression effect to standard tones - with more attenuated responses to frequently repeated standard tones. This highlights the role of NBM-DBS for sensory gating of familiar auditory information into sensory memory.

Assessment of the technical usability and efficacy of a new portable dry-electrode EEG recorder: First results of the HOMEONE study.

OBJECTIVES: The HOME project is intended to provide evidence of diagnostic and therapeutic yield of a patient-controlled EEG home-monitoring for neurological outpatients. METHODS: This study evaluated the technical and practical usability and efficacy of a new portable dry-electrode EEG recorder in comparison to conventional EEG devices based on technical assessments and inter-rater comparisons of EEG record examinations of office-based practitioners and two experienced neurologists. RESULTS: The technical assessment was based on channel-wise comparisons of band power values derived from power spectra as observed in two recording modalities. Slight yet significant differences were observed only in the Delta-frequency band (1.5-4 Hz). The fraction of automatically detected artifact segments was larger in the new portable recordings than in conventional recordings (20% vs. 11%, median). Overall, 93% of raters' stated diagnostic findings gathered from conventional devices were concordant with stated diagnostic findings gathered from the new portable device. CONCLUSION: The new EEG device was shown to have technical comparability to and a high concordance rate of diagnostic findings with conventional EEG devices. SIGNIFICANCE: The new portable dry-electrode EEG device is suitable to meet the HOME projects' goal of establishing a patient-controlled EEG home-monitoring in the routine care of neurological outpatients. TRIAL REGISTRATION: DRKS DRKS00012685. Registered 09 August 2017, retrospectively registered.

Convolutional neural networks for decoding of covert attention focus and saliency maps for EEG feature visualization.

OBJECTIVE: Convolutional neural networks (CNNs) have proven successful as function approximators and have therefore been used for classification problems including electroencephalography (EEG) signal decoding for brain-computer interfaces (BCI). Artificial neural networks, however, are considered black boxes, because they usually have thousands of parameters, making interpretation of their internal processes challenging. Here we systematically evaluate the use of CNNs for EEG signal decoding and investigate a method for visualizing the CNN model decision process. APPROACH: We developed a CNN model to decode the covert focus of attention from EEG event-related potentials during object selection. We compared the CNN and the commonly used linear discriminant analysis (LDA) classifier performance, applied to datasets with different dimensionality, and analyzed transfer learning capacity. Moreover, we validated the impact of single model components by systematically altering the model. Furthermore, we investigated the use of saliency maps as a tool for visualizing the spatial and temporal features driving the model output. MAIN RESULTS: The CNN model and the LDA classifier achieved comparable accuracy on the lower-dimensional dataset, but CNN exceeded LDA performance significantly on the higher-dimensional dataset (without hypothesis-driven preprocessing), achieving an average decoding accuracy of 90.7% (chance level = 8.3%). Parallel convolutions, tanh or ELU activation functions, and dropout regularization proved valuable for model performance, whereas the sequential convolutions, ReLU activation function, and batch normalization components reduced accuracy or yielded no significant difference. Saliency maps revealed meaningful features, displaying the typical spatial distribution and latency of the P300 component expected during this task. SIGNIFICANCE: Following systematic evaluation, we provide recommendations for when and how to use CNN models in EEG decoding. Moreover, we propose a new approach for investigating the neural correlates of a cognitive task by training CNN models on raw high-dimensional EEG data and utilizing saliency maps for relevant feature extraction.