A Generalizable Brain-Computer Interface (BCI) Using Machine Learning for Feature Discovery

This work describes a generalized method for classifying motor-related neural signals for a brain-computer interface (BCI), based on a stochastic machine learning method. The method differs from the various feature extraction and selection techniques employed in many other BCI systems. The classifier does not use extensive a-priori information, resulting in reduced reliance on highly specific domain knowledge. Instead of pre-defining features, the time-domain signal is input to a population of multi-layer perceptrons (MLPs) in order to perform a stochastic search for the best structure. The results showed that the average performance of the new algorithm outperformed other published methods using the Berlin BCI IV (2008) competition dataset and was comparable to the best results in the Berlin BCI II (2002–3) competition dataset. The new method was also applied to electroencephalography (EEG) data recorded from five subjects undertaking a hand squeeze task and demonstrated high levels of accuracy with a mean classification accuracy of 78.9% after five-fold cross-validation. Our new approach has been shown to give accurate results across different motor tasks and signal types as well as between subjects.     

Nonnegative tensor factorization for continuous EEG classification

In this paper we present a method for continuous EEG classification, where we employ nonnegative tensor factorization (NTF) to determine discriminative spectral features and use the Viterbi algorithm to continuously classify multiple mental tasks. This is an extension of our previous work on the use of nonnegative matrix factorization (NMF) for EEG classification. Numerical experiments with two data sets in BCI competition, confirm the useful behavior of the method for continuous EEG classification.     

Linear correlation between fractal dimension of EEG signal and handgrip force

Fractal dimension (FD) has been proved useful in quantifying the complexity of dynamical signals in biology and medicine. In this study, we measured FDs of human electroencephalographic (EEG) signals at different levels of handgrip forces. EEG signals were recorded from five major motor-related cortical areas in eight normal healthy subjects. FDs were calculated using three different methods. The three physiological periods of handgrip (command preparation, movement and holding periods) were analyzed and compared. The results showed that FDs of the EEG signals during the movement and holding periods increased linearly with handgrip force, whereas FD during the preparation period had no correlation with force. The results also demonstrated that one method (Katz’s) gave greater changes in FD, and thus, had more power in capturing the dynamic changes in the signal. The linear increase of FD, together with results from other EEG and neuroimaging studies, suggest that under normal conditions the brain recruits motor neurons at a linear progress when increasing the force.     

Classification of Motor Imagery EEG Based on Time-Domain and Frequency-Domain Dual-Stream Convolutional Neural Network

Background and objectiveAn important task of the brain-computer interface (BCI) of motor imagery is to extract effective time-domain features, frequency-domain features or time-frequency domain features from the raw electroencephalogram (EEG) signals for classification of motor imagery. However, choosing an appropriate method to combine time domain and frequency domain features to improve the performance of motor imagery recognition is still a research hotspot.MethodsIn order to fully extract and utilize the time-domain and frequency-domain features of EEG in classification tasks, this paper proposed a novel dual-stream convolutional neural network (DCNN), which can use time domain signal and frequency domain signal as the inputs, and the extracted time-domain features and frequency-domain features are fused by linear weighting for classification training. Furthermore, the weight can be learned by the DCNN automatically.ResultsThe experiments based on BCI competition II dataset III and BCI competition IV dataset 2a showed that the model proposed by this study has better performance than other conventional methods. The model used time-frequency signal as the inputs had better performance than the model only used time-domain signals or frequency-domain signals. The accuracy of classification was improved for each subject compared with the models only used one signals as the inputs.ConclusionsFurther analysis shown that the fusion weight of different subject is specifically, adjusting the weight coefficient automatically is helpful to improve the classification accuracy.     

不同思维作业电信号的高准确度模式分类

本文采用独立分量(ICA)分析对不同思维作业的脑电(EEG)信号进行预处理,再用自回归(AR)参数模型提取EEG信号特征,最后利用BP网络完成对特征样本集的训练和分类。实验结果表明,所采用的方法提高了脑电思维模式作业的准确度,对两种到五种不同思维作业未经训练的数据的平均分类准确度达到79％以上,超过现有文献报道的结果。     

Hidden pattern discovery on event related potential EEG signals

EEG signals are important to capture brain disorders. They are useful for analyzing the cognitive activity of the brain and diagnosing types of seizure and potential mental health problems. The Event Related Potential can be measured through the EEG signal. However, it is always difficult to interpret due to its low amplitude and sensitivity to changes of the mental activity. In this paper, we propose a novel approach to incrementally detect the pattern of this kind of EEG signal. This approach successfully summarizes the whole stream of the EEG signal by finding the correlations across the electrodes and discriminates the signals corresponding to various tasks into different patterns. It is also able to detect the transition period between different EEG signals and identify the electrodes which contribute the most to these signals. The experimental results show that the proposed method allows the significant meaning of the EEG signal to be obtained from the extracted pattern.     

Individually Adapted Imagery Improves Brain-Computer Interface Performance in End-Users with Disability

Brain-computer interfaces (BCIs) translate oscillatory electroencephalogram (EEG) patterns into action. Different mental activities modulate spontaneous EEG rhythms in various ways. Non-stationarity and inherent variability of EEG signals, however, make reliable recognition of modulated EEG patterns challenging. Able-bodied individuals who use a BCI for the first time achieve - on average - binary classification performance of about 75%. Performance in users with central nervous system (CNS) tissue damage is typically lower. User training generally enhances reliability of EEG pattern generation and thus also robustness of pattern recognition. In this study, we investigated the impact of mental tasks on binary classification performance in BCI users with central nervous system (CNS) tissue damage such as persons with stroke or spinal cord injury (SCI). Motor imagery (MI), that is the kinesthetic imagination of movement (e.g. squeezing a rubber ball with the right hand), is the "gold standard" and mainly used to modulate EEG patterns. Based on our recent results in able-bodied users, we hypothesized that pair-wise combination of "brain-teaser" (e.g. mental subtraction and mental word association) and "dynamic imagery" (e.g. hand and feet MI) tasks significantly increases classification performance of induced EEG patterns in the selected end-user group. Within-day (How stable is the classification within a day?) and between-day (How well does a model trained on day one perform on unseen data of day two?) analysis of variability of mental task pair classification in nine individuals confirmed the hypothesis. We found that the use of the classical MI task pair hand vs. feed leads to significantly lower classification accuracy - in average up to 15% less - in most users with stroke or SCI. User-specific selection of task pairs was again essential to enhance performance. We expect that the gained evidence will significantly contribute to make imagery-based BCI technology become accessible to a larger population of users including individuals with special needs due to CNS damage.     

Application of recurrence quantification analysis for the automated identification of epileptic EEG signals

Epilepsy is a common neurological disorder that is characterized by the recurrence of seizures. Electroencephalogram (EEG) signals are widely used to diagnose seizures. Because of the non-linear and dynamic nature of the EEG signals, it is difficult to effectively decipher the subtle changes in these signals by visual inspection and by using linear techniques. Therefore, non-linear methods are being researched to analyze the EEG signals. In this work, we use the recorded EEG signals in Recurrence Plots (RP), and extract Recurrence Quantification Analysis (RQA) parameters from the RP in order to classify the EEG signals into normal, ictal, and interictal classes. Recurrence Plot (RP) is a graph that shows all the times at which a state of the dynamical system recurs. Studies have reported significantly different RQA parameters for the three classes. However, more studies are needed to develop classifiers that use these promising features and present good classification accuracy in differentiating the three types of EEG segments. Therefore, in this work, we have used ten RQA parameters to quantify the important features in the EEG signals.These features were fed to seven different classifiers: Support vector machine (SVM), Gaussian Mixture Model (GMM), Fuzzy Sugeno Classifier, K-Nearest Neighbor (KNN), Naive Bayes Classifier (NBC), Decision Tree (DT), and Radial Basis Probabilistic Neural Network (RBPNN). Our results show that the SVM classifier was able to identify the EEG class with an average efficiency of 95.6%, sensitivity and specificity of 98.9% and 97.8%, respectively.     

Extracting optimal tempo-spatial features using local discriminant bases and common spatial patterns for brain computer interfacing

Brain computer interfaces (BCI) provide a new approach to human computer communication, where the control is realised via performing mental tasks such as motor imagery (MI). In this study, we investigate a novel method to automatically segment electroencephalographic (EEG) data within a trial and extract features accordingly in order to improve the performance of MI data classification techniques. A new local discriminant bases (LDB) algorithm using common spatial patterns (CSP) projection as transform function is proposed for automatic trial segmentation. CSP is also used for feature extraction following trial segmentation. This new technique also allows to obtain a more accurate picture of the most relevant temporal–spatial points in the EEG during the MI. The results are compared with other standard temporal segmentation techniques such as sliding window and LDB based on the local cosine transform (LCT).     

Online Classification of Multiple Motor Imagery Tasks Using Filter Bank Based Maximum-a-Posteriori Common Spatial Pattern Filters

ObjectiveThe main objective of this paper is to propose a novel technique, called filter bank maximum a-posteriori common spatial pattern (FB-MAP-CSP) algorithm, for online classification of multiple motor imagery activities using electroencephalography (EEG) signals. The proposed technique addresses the overfitting issue of CSP in addition to utilizing the spectral information of EEG signals inside the framework of filter banks while extending it to more than two conditions.Materials and methodsThe classification of motor imagery signals is based upon the detection of event-related de-synchronization (ERD) phenomena in the μ and β rhythms of EEG signals. Accordingly, two modifications in the existing MAP-CSP technique are presented: (i) The (pre-processed) EEG signals are spectrally filtered by a bank of filters lying in the μ and β brainwave frequency range, (ii) the framework of MAP-CSP is extended to deal with multiple (more than two) motor imagery tasks classification and the spatial filters thus obtained are calculated for each sub-band, separately. Subsequently, the most imperative features over all sub-bands are selected and un-regularized linear discriminant analysis is employed for classification of multiple motor imagery tasks.ResultsPublicly available dataset (BCI Competition IV Dataset I) is used to validate the proposed method i.e. FB-MAP-CSP. The results show that the proposed method yields superior classification results, in addition to be computationally more efficient in the case of online implementation, as compared to the conventional CSP based techniques and its variants for multiclass motor imagery classification.ConclusionThe proposed FB-MAP-CSP algorithm is found to be a potential / superior method for classifying multi-condition motor imagery EEG signals in comparison to FBCSP based techniques.     

Automatic detection of epileptic EEG signals using higher order cumulant features

The unpredictability of the occurrence of epileptic seizures makes it difficult to detect and treat this condition effectively. An automatic system that characterizes epileptic activities in EEG signals would allow patients or the people near them to take appropriate precautions, would allow clinicians to better manage the condition, and could provide more insight into these phenomena thereby revealing important clinical information. Various methods have been proposed to detect epileptic activity in EEG recordings. Because of the nonlinear and dynamic nature of EEG signals, the use of nonlinear Higher Order Spectra (HOS) features is a seemingly promising approach. This paper presents the methodology employed to extract HOS features (specifically, cumulants) from normal, interictal, and epileptic EEG segments and to use significant features in classifiers for the detection of these three classes. In this work, 300 sets of EEG data belonging to the three classes were used for feature extraction and classifier development and evaluation. The results show that the HOS based measures have unique ranges for the different classes with high confidence level (p-value < 0.0001). On evaluating several classifiers with the significant features, it was observed that the Support Vector Machine (SVM) presented a high detection accuracy of 98.5% thereby establishing the possibility of effective EEG segment classification using the proposed technique.     

Classifying Response Correctness across Different Task Sets: A Machine Learning Approach

Erroneous behavior usually elicits a distinct pattern in neural waveforms. In particular, inspection of the concurrent recorded electroencephalograms (EEG) typically reveals a negative potential at fronto-central electrodes shortly following a response error (Ne or ERN) as well as an error-awareness-related positivity (Pe). Seemingly, the brain signal contains information about the occurrence of an error. Assuming a general error evaluation system, the question arises whether this information can be utilized in order to classify behavioral performance within or even across different cognitive tasks. In the present study, a machine learning approach was employed to investigate the outlined issue. Ne as well as Pe were extracted from the single-trial EEG signals of participants conducting a flanker and a mental rotation task and subjected to a machine learning classification scheme (via a support vector machine, SVM). Overall, individual performance in the flanker task was classified more accurately, with accuracy rates of above 85%. Most importantly, it was even feasible to classify responses across both tasks. In particular, an SVM trained on the flanker task could identify erroneous behavior with almost 70% accuracy in the EEG data recorded during the rotation task, and vice versa. Summed up, we replicate that the response-related EEG signal can be used to identify erroneous behavior within a particular task. Going beyond this, it was possible to classify response types across functionally different tasks. Therefore, the outlined methodological approach appears promising with respect to future applications.     

Electroencephalogram bands modulated by vigilance states in an anuran species: a factor analytic approach

Dramatic changes in neocortical electroencephalogram (EEG) rhythms are associated with the sleep–waking cycle in mammals. Although amphibians are thought to lack a neocortical homologue, changes in rest–activity states occur in these species. In the present study, EEG signals were recorded from the surface of the cerebral hemispheres and midbrain on both sides of the brain in an anuran species, Babina daunchina, using electrodes contacting the meninges in order to measure changes in mean EEG power across behavioral states. Functionally relevant frequency bands were identified using factor analysis. The results indicate that: (1) EEG power was concentrated in four frequency bands during the awake or active state and in three frequency bands during rest; (2) EEG bands in frogs differed substantially from humans, especially in the fast frequency band; (3) bursts similar to mammalian sleep spindles, which occur in non-rapid eye movement mammalian sleep, were observed when frogs were at rest suggesting sleep spindle-like EEG activity appeared prior to the evolution of mammals.     

Characteristics of electroencephalographic responses induced by a pleasant and an unpleasant odor

More than sensory stimuli, odorous stimuli were employed to facilitate the evocation of emotional responses in the present study. The odor-stimulated emotion was evaluated by investigating specific features of encephalographic (EEG) responses produced thereof. In this study, the concentrations of the same odor were altered; viz., the changes in odor-induced emotional level were compared with the concurrently monitored EEG response features. In addition, we performed the mental task to evoke the arousal state of the brain and investigated the resemblance of response characteristics of the resting state to the post-mental task resting state. Subjects having no abnormalities in the sense of smell included 12 male undergraduate and graduate students (age range: 22-26 years). Experiment I involved 2 types of odors that induced favorable odorous stimuli (pleasant induction); test-solutions were either diluted 150 (easily perceptible odorous sensation) or 500 (slightly perceptible odorous stimuli) times. Experiment II had 2 types of odors that evoked unfavorable odorous stimuli (unpleasant induction), and test-solutions with dilution rates similar to those of pleasant induction were prepared. Odorless distilled water was used as the control in both experiments. From results of rating the odorous stimuli of our compounds used, the candidates were respectively found to be appropriate in inducing the pleasant and unpleasant smell sensations. The analyses of EEG responses on inducing pleasant and unpleasant smell sensations revealed that the EEG activities of the left frontal region were enhanced. This finding may establish the hypothesis of a relationship prevailing between the positive approach-related emotion evoked by the visual sensation and the left hemisphere (Davidson, 1992; Tomarken et al., 1989). In other words, it can be interpreted that the negative withdrawal-related emotion may be associated with activities of the right hemisphere. However, this hypothesis may not be applicable to the unpleasant odors, as the unpleasant emotions are activated by the unpleasant odors not only in the bilateral frontal regions but also over an extensive area of the brain. As such, the pleasant emotions are evoked in the left frontal brain region while the unpleasant emotions are incited in the bilateral frontal and extensive regions in the brain with the odorous stimuli. Moreover, intrinsic EEG activities in response to the pleasant and unpleasant inputs were not observed after performing the mental tasks. In other words, EEG responses reflecting central nervous system activities elevated by loading of the mental tasks as a result of exposure to the pleasant and unpleasant odors may not apparently be observed.     

Multimodal information improves the rapid detection of mental fatigue

One of the major challenge in the detection of mental states is improving the accuracy of brain activity-based detectors with additional information from extracranial signals. We assessed the suitability, for real-time mental fatigue detection, of EEG, EOG and ECG measurements, taken separately or together. Thirteen subjects performed six blocks of switching tasks. For each participant, the block with the lowest error rate from the first two blocks and the block with the highest error rate from the last three blocks were discriminated with a machine learning algorithm (support vector machine). The classification scores obtained with ECG or EOG were greater than would be expected by chance (>50%) for time windows of at least 8 s. EEG was the best single mode of detection, with classification scores ranging from 80 ± 3% with a 4 s time window to 94 ± 2% with a 30 s time window. The addition of ECG and EOG features to EEG features significantly increased classification scores for short time windows (e.g., to 86 ± 3% with a 4 s time window, p < 0.001). For short time windows (up to 12 s), ECG significantly increased the discriminatory power of EEG, whereas EOG did not. These results demonstrate that mental state detection on the basis of extracerebral measurements is feasible and that a combination of EEG and ECG is particularly appropriate for the rapid detection of mental fatigue.     

Automated detection of schizophrenia using optimal wavelet-based [... formula ...] norm features extracted from single-channel EEG

Schizophrenia (SZ) is a mental disorder, which affects the ability of human thinking, memory, and way of living. Manual screening of SZ patients is tedious, laborious and prone to human errors. Hence, we developed a computer-aided diagnosis (CAD) system to diagnose SZ patients accurately using single-channel electroencephalogram (EEG) signals. The EEG signals are nonlinear and non-stationary. Hence, we have used wavelet-based features to capture the hidden non-stationary nature present in the signal. First, the EEG signals are subjected to the the wavelet decomposition through six iterations, which yields seven sub-bands. The l1 norm is computed for each sub-band. The extracted norm features are disseminated to various classification algorithms. We have obtained the highest accuracy of 99.21% and 97.2% using K-nearest neighbor classifiers with ten-fold and leave-one-subject-out cross-validations. The developed single-channel EEG wavelet-based CAD model can help the clinicians to confirm the outcome of their manual screening and obtain an accurate diagnosis.     

Relevance of nonlinear lumped-parameter models in the analysis of depth-EEG epileptic signals

In the field of epilepsy, the analysis of stereoelectroencephalographic (SEEG, intra-cerebral recording) signals with signal processing methods can help to better identify the epileptogenic zone, the area of the brain responsible for triggering seizures, and to better understand its organization. In order to evaluate these methods and to physiologically interpret the results they provide, we developed a model able to produce EEG signals from “organized” networks of neural populations. Starting from a neurophysiologically relevant model initially proposed by Lopes Da Silva et al. [Lopes da Silva FH, Hoek A, Smith H, Zetterberg LH (1974) Kybernetic 15: 27–37] and recently re-designed by Jansen et al. [Jansen BH, Zouridakis G, Brandt ME (1993) Biol Cybern 68: 275–283] the present study demonstrates that this model can be extended to generate spontaneous EEG signals from multiple coupled neural populations. Model parameters related to excitation, inhibition and coupling are then altered to produce epileptiform EEG signals. Results show that the qualitative behavior of the model is realistic; simulated signals resemble those recorded from different brain structures for both interictal and ictal activities. Possible exploitation of simulations in signal processing is illustrated through one example; statistical couplings between both simulated signals and real SEEG signals are estimated using nonlinear regression. Results are compared and show that, through the model, real SEEG signals can be interpreted with the aid of signal processing methods. Received: 3 January 2000 / Accepted: 24 March 2000     

Towards a novel monitor of intraoperative awareness: selecting paradigm settings for a movement-based brain-computer interface

During 0.1-0.2% of operations with general anesthesia, patients become aware during surgery. Unfortunately, pharmacologically paralyzed patients cannot seek attention by moving. Their attempted movements may however induce detectable EEG changes over the motor cortex. Here, methods from the area of movement-based brain-computer interfacing are proposed as a novel direction in anesthesia monitoring. Optimal settings for development of such a paradigm are studied to allow for a clinically feasible system. A classifier was trained on recorded EEG data of ten healthy non-anesthetized participants executing 3-second movement tasks. Extensive analysis was performed on this data to obtain an optimal EEG channel set and optimal features for use in a movement detection paradigm. EEG during movement could be distinguished from EEG during non-movement with very high accuracy. After a short calibration session, an average classification rate of 92% was obtained using nine EEG channels over the motor cortex, combined movement and post-movement signals, a frequency resolution of 4 Hz and a frequency range of 8-24 Hz. Using Monte Carlo simulation and a simple decision making paradigm, this translated into a probability of 99% of true positive movement detection within the first two and a half minutes after movement onset. A very low mean false positive rate of <0.01% was obtained. The current results corroborate the feasibility of detecting movement-related EEG signals, bearing in mind the clinical demands for use during surgery. Based on these results further clinical testing can be initiated.     

EEG beta and gamma powers when comparing certain psychophysiological states with normal and weakened electromyogram of facial muscles

With the advancement of contemporary techniques for studies of high-frequency electroencephalograms (EEGs), possible contamination of the EEG with the electromyogram (EMG) of pericranial muscles has raised more and more concern. The aim of the present study was to demonstrate if certain EEG correlates of mental activities can be revealed in a high-frequency scalp EEG in spite of EMG contamination. Nineteen healthy women who performed similar test tasks before and after cosmetic injections of Dysport in various facial regions for reduction of the activity of facial muscles took part in the study. Inductions of emotional states with different valences, memory storing, and extraction of verbal information were used in the test tasks. The default state of rest was examined as well. During performance of the tasks, parallel registrations of the EEG from the scalp surface (19 channels) and EMG from several facial muscles (6 channels) were carried out. Changes in the spectral power in β₂ and low γ frequency bands (18–40 Hz) in EEG- and EMG-derivations after Dysport injections were analyzed. Changes in the spectral power in the same bands in pairwise comparisons for the test tasks before and after Dysport injections were also analyzed separately. It was demonstrated that Dysport injections lead to reduction of the EMG power in areas of the injections and to reduction of EEG power in the frontal and temporal derivations. However, the EEG-correlates revealed when comparing different test tasks remained qualitatively invariable as for after and before Disport injections. These facts confirm that EMG makes a noticeable contribution to the electric activity registered from the scalp in the frequency ranges greater than 18 Hz. At the same time, one can see that at least in certain experimental situations the influence of EMG does not make impossible identification of EEG-correlates of mental activity with EEG registration from the head surface at least in the β₂ and low γ frequency bands (18–40 Hz).     

A Supplementary System for a Brain-Machine Interface Based on Jaw Artifacts for the Bidimensional Control of a Robotic Arm

Non-invasive Brain-Machine Interfaces (BMIs) are being used more and more these days to design systems focused on helping people with motor disabilities. Spontaneous BMIs translate user''s brain signals into commands to control devices. On these systems, by and large, 2 different mental tasks can be detected with enough accuracy. However, a large training time is required and the system needs to be adjusted on each session. This paper presents a supplementary system that employs BMI sensors, allowing the use of 2 systems (the BMI system and the supplementary system) with the same data acquisition device. This supplementary system is designed to control a robotic arm in two dimensions using electromyographical (EMG) signals extracted from the electroencephalographical (EEG) recordings. These signals are voluntarily produced by users clenching their jaws. EEG signals (with EMG contributions) were registered and analyzed to obtain the electrodes and the range of frequencies which provide the best classification results for 5 different clenching tasks. A training stage, based on the 2-dimensional control of a cursor, was designed and used by the volunteers to get used to this control. Afterwards, the control was extrapolated to a robotic arm in a 2-dimensional workspace. Although the training performed by volunteers requires 70 minutes, the final results suggest that in a shorter period of time (45 min), users should be able to control the robotic arm in 2 dimensions with their jaws. The designed system is compared with a similar 2-dimensional system based on spontaneous BMIs, and our system shows faster and more accurate performance. This is due to the nature of the control signals. Brain potentials are much more difficult to control than the electromyographical signals produced by jaw clenches. Additionally, the presented system also shows an improvement in the results compared with an electrooculographic system in a similar environment.