Generalized optimal spatial filtering using a kernel approach with application to EEG classification

Common spatial patterns (CSP) has been widely used for finding the linear spatial filters which are able to extract the discriminative brain activities between two different mental tasks. However, the CSP is difficult to capture the nonlinearly clustered structure from the non-stationary EEG signals. To relax the presumption of strictly linear patterns in the CSP, in this paper, a generalized CSP (GCSP) based on generalized singular value decomposition (GSVD) and kernel method is proposed. Our method is able to find the nonlinear spatial filters which are formulated in the feature space defined by a nonlinear mapping through kernel functions. Furthermore, in order to overcome the overfitting problem, the regularized GCSP is developed by adding the regularized parameters. The experimental results demonstrate that our method is an effective nonlinear spatial filtering method.     

Translation of EEG spatial filters from resting to motor imagery using independent component analysis

Electroencephalogram (EEG)-based brain-computer interfaces (BCIs) often use spatial filters to improve signal-to-noise ratio of task-related EEG activities. To obtain robust spatial filters, large amounts of labeled data, which are often expensive and labor-intensive to obtain, need to be collected in a training procedure before online BCI control. Several studies have recently developed zero-training methods using a session-to-session scenario in order to alleviate this problem. To our knowledge, a state-to-state translation, which applies spatial filters derived from one state to another, has never been reported. This study proposes a state-to-state, zero-training method to construct spatial filters for extracting EEG changes induced by motor imagery. Independent component analysis (ICA) was separately applied to the multi-channel EEG in the resting and the motor imagery states to obtain motor-related spatial filters. The resultant spatial filters were then applied to single-trial EEG to differentiate left- and right-hand imagery movements. On a motor imagery dataset collected from nine subjects, comparable classification accuracies were obtained by using ICA-based spatial filters derived from the two states (motor imagery: 87.0%, resting: 85.9%), which were both significantly higher than the accuracy achieved by using monopolar scalp EEG data (80.4%). The proposed method considerably increases the practicality of BCI systems in real-world environments because it is less sensitive to electrode misalignment across different sessions or days and does not require annotated pilot data to derive spatial filters.     

Improving N1 classification by grouping EEG trials with phases of pre-stimulus EEG oscillations

A reactive brain-computer interface using electroencephalography (EEG) relies on the classification of evoked ERP responses. As the trial-to-trial variation is evitable in EEG signals, it is a challenge to capture the consistent classification features distribution. Clustering EEG trials with similar features and utilizing a specific classifier adjusted to each cluster can improve EEG classification. In this paper, instead of measuring the similarity of ERP features, the brain states during image stimuli presentation that evoked N1 responses were used to group EEG trials. The correlation between momentary phases of pre-stimulus EEG oscillations and N1 amplitudes was analyzed. The results demonstrated that the phases of time–frequency points about 5.3 Hz and 0.3 s before the stimulus onset have significant effect on the ERP classification accuracy. Our findings revealed that N1 components in ERP fluctuated with momentary phases of EEG. We also further studied the influence of pre-stimulus momentary phases on classification of N1 features. Results showed that linear classifiers demonstrated outstanding classification performance when training and testing trials have close momentary phases. Therefore, this gave us a new direction to improve EEG classification by grouping EEG trials with similar pre-stimulus phases and using each to train unit classifiers respectively.     

EEG classification of driver mental states by deep learning

Driver fatigue is attracting more and more attention, as it is the main cause of traffic accidents, which bring great harm to society and families. This paper proposes to use deep convolutional neural networks, and deep residual learning, to predict the mental states of drivers from electroencephalography (EEG) signals. Accordingly we have developed two mental state classification models called EEG-Conv and EEG-Conv-R. Tested on intra- and inter-subject, our results show that both models outperform the traditional LSTM- and SVM-based classifiers. Our major findings include (1) Both EEG-Conv and EEG-Conv-R yield very good classification performance for mental state prediction; (2) EEG-Conv-R is more suitable for inter-subject mental state prediction; (3) EEG-Conv-R converges more quickly than EEG-Conv. In summary, our proposed classifiers have better predictive power and are promising for application in practical brain-computer interaction .     

Analysis and classification of speech imagery EEG for BCI

Electroencephalogram (EEG) is generally used in brain–computer interface (BCI), including motor imagery, mental task, steady-state evoked potentials (SSEPs) and P300. In order to complement existing motor-based control paradigms, this paper proposed a novel imagery mode: speech imagery. Chinese characters are monosyllabic and one Chinese character can express one meaning. Thus, eight Chinese subjects were required to read two Chinese characters in mind in this experiment. There were different shapes, pronunciations and meanings between two Chinese characters. Feature vectors of EEG signals were extracted by common spatial patterns (CSP), and then these vectors were classified by support vector machine (SVM). The accuracy between two characters was not superior. However, it was still effective to distinguish whether subjects were reading one character in mind, and the accuracies were between 73.65% and 95.76%. The results were better than vowel speech imagery, and they were suitable for asynchronous BCI. BCI systems will be also extended from motor imagery to combine motor imagery and speech imagery in the future.     

Single molecule analysis of the actomyosin motor

Progress in imaging techniques and nano-manipulation of single molecules has been remarkable. These techniques have allowed the accurate determination of myosin-head-induced displacements and of how the mechanical cycles of the actomyosin motor are coupled to ATP hydrolysis. This has been achieved by measuring mechanical and chemical events of actomyosin directly at the single molecule level. Recent studies have made detailed measurements of myosin step size and mechanochemical coupling. The results of these studies suggest a new model for the mechanism of motion underlying actomyosin motors, which differs from the currently accepted lever-arm swinging model.     

States of productive and reproductive imagination reflected in the local synchronization of alpha and theta EEG frequencies

Healthy subjects (n = 83) performed a task involving reproductive imagination (remembrances): they were asked to remember and to have an imaginary walk along a well-known road (a real walk, the reproductive imagination state, RIS). Then the subjects performed a task involving productive imagination: they were asked to imagine a city that does not actually exist and to have an imaginary walk through it (a fictitious walk, productive imagination state, PIS). The reference values were measured at rest with the eyes open (REO). Monopolar EEGs were recorded from 19 areas of the scalp surface (10–20 system). An increase in EEG power in the α₂ frequency range during RIS and even more pronounced increase during PIS as compared with REO were considered to reflect the internalization of attention. We also observed multidirectional dynamics of EEG power in the θ and α1 ranges during PIS and RIS as compared with REO, which suggests the dominance of free associative manipulation of visual images in PIS in contrast to RIS, where algorithmically ordered operations with visual images stored in the memory were dominant.     

Relevant EEG features for the classification of spontaneous motor-related tasks

 There is a growing interest in the use of physiological signals for communication and operation of devices for the severely motor disabled as well as for healthy people. A few groups around the world have developed brain-computer interfaces (BCIs) that rely upon the recognition of motor-related tasks (i.e., imagination of movements) from on-line EEG signals. In this paper we seek to find and analyze the set of relevant EEG features that best differentiate spontaneous motor-related mental tasks from each other. This study empirically demonstrates the benefits of heuristic feature selection methods for EEG-based classification of mental tasks. In particular, it is shown that the classifier performance improves for all the considered subjects with only a small proportion of features. Thus, the use of just those relevant features increases the efficiency of the brain interfaces and, most importantly, enables a greater level of adaptation of the personal BCI to the individual user. Received: 15 January 2001 / Accepted in revised form: 19 July 2001     

Single molecule energetics of F1-ATPase motor

Motor proteins are essential in life processes because they convert the free energy of ATP hydrolysis to mechanical work. However, the fundamental question on how they work when different amounts of free energy are released after ATP hydrolysis remains unanswered. To answer this question, it is essential to clarify how the stepping motion of a motor protein reflects the concentrations of ATP, ADP, and P_i in its individual actions at a single molecule level. The F₁ portion of ATP synthase, also called F₁-ATPase, is a rotary molecular motor in which the central γ-subunit rotates against the α₃β₃ cylinder. The motor exhibits clear step motion at low ATP concentrations. The rotary action of this motor is processive and generates a high torque. These features are ideal for exploring the relationship between free energy input and mechanical work output, but there is a serious problem in that this motor is severely inhibited by ADP. In this study, we overcame this problem of ADP inhibition by introducing several mutations while retaining high enzymatic activity. Using a probe of attached beads, stepping rotation against viscous load was examined at a wide range of free energy values by changing the ADP concentration. The results showed that the apparent work of each individual step motion was not affected by the free energy of ATP hydrolysis, but the frequency of each individual step motion depended on the free energy. This is the first study that examined the stepping motion of a molecular motor at a single molecule level with simultaneous systematic control of ΔG_ATP. The results imply that microscopically defined work at a single molecule level cannot be directly compared with macroscopically defined free energy input.     

Selection of relevant features for EEG signal classification of schizophrenic patients

In this paper, EEG signals of 20 schizophrenic patients and 20 age-matched control participants are analyzed with the objective of determining the more informative channels and finally distinguishing the two groups. For each case, 22 channels of EEG were recorded. A two-stage feature selection algorithm is designed, such that, the more informative channels are first selected to enhance the discriminative information. Two methods, bidirectional search and plus-L minus-R (LRS) techniques are employed to select these informative channels. The interesting point is that most of selected channels are located in the temporal lobes (containing the limbic system) that confirm the neuro-phychological differences in these areas between the schizophrenic and normal participants. After channel selection, genetic algorithm (GA) is employed to select the best features from the selected channels. In this case, in addition to elimination of the less informative channels, the redundant and less discriminant features are also eliminated. A computationally fast algorithm with excellent classification results is obtained. Implementation of this efficient approach involves several features including autoregressive (AR) model parameters, band power, fractal dimension and wavelet energy. To test the performance of the final subset of features, classifiers including linear discriminant analysis (LDA) and support vector machine (SVM) are employed to classify the reduced feature set of the two groups. Using the bidirectional search for channel selection, a classification accuracy of 84.62% and 99.38% is obtained for LDA and SVM, respectively. Using the LRS technique for channel selection, a classification accuracy of 88.23% and 99.54% is also obtained for LDA and SVM, respectively. Finally, the results are compared and contrasted with two well-known methods namely, the single-stage feature selection (evolutionary feature selection) and principal component analysis (PCA)-based feature selection. The results show improved accuracy of classification in relatively low computational time with the two-stage feature selection.     

Automated diagnosis of epileptic EEG using entropies

Epilepsy is a neurological disorder characterized by the presence of recurring seizures. Like many other neurological disorders, epilepsy can be assessed by the electroencephalogram (EEG). The EEG signal is highly non-linear and non-stationary, and hence, it is difficult to characterize and interpret it. However, it is a well-established clinical technique with low associated costs. In this work, we propose a methodology for the automatic detection of normal, pre-ictal, and ictal conditions from recorded EEG signals. Four entropy features namely Approximate Entropy (ApEn), Sample Entropy (SampEn), Phase Entropy 1 (S1), and Phase Entropy 2 (S2) were extracted from the collected EEG signals. These features were fed to seven different classifiers: Fuzzy Sugeno Classifier (FSC), Support Vector Machine (SVM), K-Nearest Neighbour (KNN), Probabilistic Neural Network (PNN), Decision Tree (DT), Gaussian Mixture Model (GMM), and Naive Bayes Classifier (NBC). Our results show that the Fuzzy classifier was able to differentiate the three classes with a high accuracy of 98.1%. Overall, compared to previous techniques, our proposed strategy is more suitable for diagnosis of epilepsy with higher accuracy.     

Variance and discriminatory analyses for the classification of wakefulness-dependent EEG patterns]

1. To describe 8 alertness-dependent EEG activity patterns to be observed during wakefulness with eyes closed, mean values and standard deviations of the frequencies 4, ..., 13, 14 to 17, 18 to 22 and 23 to 40 c/sec, of the percentage activity in the theta, alpha and beta bands as well as of the theta and alpha amplitudes were calculated. 2. On the basis of 18 variables significant differences between 7 activity patterns were ascertained by means of univariate and multivariate analyses of variance. 3. In the course of stepwise reduction of variables within the framework of a linear discriminatory analysis, an optimal set of variables was determined for the separation of the patterns comprising the following variables: mean value and standard deviation of the alpha amplitudes and mean value of the theta amplitudes as well as the percent quantity of the frequencies 3, ..., 7, 10, 14 to 17 and 23 to 40 c/sec. 4. By means of a linear regression analysis it was shown that the EEG scoring system can be reflected on an interval scale. In connection with results on the reliability of the methods used the results were discussed.     

Hidden Markov models used for the offline classification of EEG data.

Hidden Markov models (HMM) are introduced for the offline classification of single-trail EEG data in a brain-computer-interface (BCI). The HMMs are used to classify Hjorth parameters calculated from bipolar EEG data, recorded during the imagination of a left or right hand movement. The effects of different types of HMMs on the recognition rate are discussed. Furthermore a comparison of the results achieved with the linear discriminant (LD) and the HMM, is presented.     

Rhythmic components of the EEG of 6-month-old children

Spectral EEG characteristics were studied in 12 six-month healthy children in a state of attention, attracted by visual stimuli and also at lowering of the level of visual afferentation and in drowsiness. 12 parts of EEG records free from artefacts were used, with analysis epoch 5 s and discretion frequency 100 counts/s. Three independent rhythmic EEG components have been revealed in alertness state, corresponding by criteria of functional reactivity and topographic localization to theta, alpha and mu EEG rhythms of the adult man. It is suggested that formation of dominating EEG rhythm with age occurs due to changes in dominant relations of the key rhythmic successions with independent genesis.     

Form and classification of motor endings in mammalian muscle spindles

The presynaptic features of 234 motor endings supplied to cat hindlimb muscle spindles have been studied in teased, silver preparations, and the postsynaptic features of a further 27 endings have been studied in serial, 1 micron thick, transverse sections. In the presynaptic study motor endings received by the three types of intrafusal muscle fibre were compared with the endings supplied to spindles by the various functional categories of motor axon. Three forms of motor ending were found that had significantly different presynaptic features. These forms correspond closely to those previously identified in the literature as p1 (beta), p2 (dynamic gamma) and trail (static gamma). The results of the postsynaptic study showed that the degree of indentation of the intrafusal muscle fibres by motor axon terminals increases with greater distance from the primary ending, irrespective of muscle-fibre type. We conclude that the postsynaptic form of intrafusal motor endings is determined by distance from primary ending and muscle-fibre type. It is not determined by type of motor axon, and cannot be correlated with presynaptic form so as to produce a unified classification of intrafusal motor endings.     

Spectral-correlative characteristics of the EEG reactions preceding conditioned reflex motor reactions in dogs

The work is a logical continuation of previous studies (analysis of the background electrical activity in the band 1-100 Hz in interstimulus intervals in the process of lever pressing alimentary conditioning in dogs) and it is dedicated to correlation-spectral analysis of prestimulus periods and EEG-reactions to conditioned stimuli, previous to conditioned lever pressing. Visually the EEG reactions present discharges of high-frequency (40-100 Hz) synchronized activity preceding for 40-300 ms the beginning of the changes in EMG of the "working" limb. It is shown that EEG reactions are characterized (in comparison with the background activity) by a higher energetic level and a greater expression of the high coherence (I greater than 0.75) and also by greater phase shifts, in counterbalance to the domination of little phase shifts in the background activity. It is assumed that the patterns of EEG reactions may participate in trigger mechanisms either eliciting conditioned motor reactions (to positive conditioned stimuli) or preventing them (to inhibitory conditioned stimuli).     

Using eigenstructure decompositions of time-varying autoregressions in common spatial patterns-based EEG signal classification

Brain–computer interfaces based on common spatial patterns (CSP) depend on the operational frequency bands of the events to be discriminated. This problem has been addressed through sub-band decompositions of the electroencephalographic signals using filter banks, then the performance relies on the number of filters that are stacked and the criteria to select their bandwidths. Here, we propose an alternative approach based on an eigenstructure decomposition of the signals’ time-varying autoregressions (TVAR). The eigen-based decomposition of the TVAR allows for subject-specific estimation of the principal time-varying frequencies, then such principal eigencomponents can be used in the traditional CSP-based classification. We show through a series of numerical experiments that the proposed classification scheme can achieve a performance which is comparable with the one obtained through the filter bank-based approach. However, our method does not rely on a preliminary selection of a frequency band, yet good performance is achieved under realistic conditions (such as reduced number of sensors and small amount of training data) independently of the time interval selected.