Search

Brain-computer interfaces based on motor imagery have played important roles in motor rehabilitation, brain function regulation, disease monitoring, etc. However, due to the low signal-to-noise ratio and spatial resolution of the EEG, their decoding performance still needs to be further improved. In this paper, we propose a multi-scale fusion convolutional neural network model (MSFCNNet) for four classification tasks involving motor imagery EEG signals. Based on EEGNet, we add an attention module and a two-dimensional dilated convolution layer to construct networks of different scales and carry out network fusion. The attention module utilizes the multi-head self-attention mechanism to highlight the most valuable features. The two-dimensional dilated convolution layer recognizes features effectively and increases the receptive field of the model. In addition, the multi-scale network fusion strategy further improves the decoding performance of the model. Our proposed model is tested on the BCI Competition IV-2a dataset and the High-Gamma dataset. Under subject-dependent and subject-independent models, the classification accuracy of the proposed model on the BCI Competition IV-2a dataset reaches 87.16% and 71.03%, respectively. And in the High-Gamma dataset, the classification accuracy of the proposed model reached 94.43% in the subject-dependent model. Compared with the filter bank common spatial pattern (FBCSP) algorithm and other deep network models, our model achieves a certain improvement in classification accuracy. The experimental results show that MSFCNNet has better decoding performance and stronger robustness. • Propose a multi-scale fusion convolutional neural network for EEG classification. • Apply the multi-scale model feature fusion strategy for model training. • Apply the multi-head attention mechanism to improve decoding performance. • Achieve 87.16% EEG classification accuracy in subject-dependent mode. • Achieve 71.03% EEG classification accuracy in subject-independent mode. [ABSTRACT FROM AUTHOR]

• A novel coarse-to-fine classification approach is proposed to decode MI tasks of the same upper limb. • A adaptive spatial–temporal graph convolutional network with residuals approach is proposed to extract spatial–temporal features. • Various classification models are applied to different classification stages, and each model is specifically designed according to the feature differences among categories. • The proposed approach obtains better results than classical approaches. In the field of Brain Computer Interface (BCI) technology, Motor Imagery (MI) plays an important role as a paradigm. One of the primary focuses of this research area lies in exploring the MI of various upper limbs. Decoding MI signals from distinct joints within the same limb poses more intricate challenges in comparison to decoding MI signals originating from different upper limbs. In order to explore more efficient decoding methods, we propose a novel coarse-to-fine classification approach to investigate categorical decoding across three tasks, namely 'rest', 'hand', and 'elbow'. This approach consists of two classification stages performed from coarse to fine. In the coarse classification stage, in order to capture the features of both resting state and the moving state in the temporal domain, the EEGNet network with temporal domain convolution is used to extract temporal domain features and classify the original samples into categories of 'rest' and 'move' ('hand', 'elbow'). In the fine classification stage, the samples of 'move' category are segmented by time to form a graph sequence. Then, an adaptive spatial–temporal graph convolutional network with residuals is utilized to extract both temporal and spatial domains' features from the graph sequence. The proposed algorithm has been validated experimentally on the MI-2 dataset and compared with contemporary methods. Its classification performance is quantified by the average accuracy which achieves a value of 72.21%. Extensive experimental results indicate that the novel coarse-to-fine classification approach is superior to the single classification approach. [ABSTRACT FROM AUTHOR]

[Display omitted] • We introduced a deep transfer learning network with two classifiers based on sample selection for motor imagery brain-computer interface. • By selecting similar samples and using two adversarial classifiers, an objective function was designed to minimize domain differences while maximizing classification accuracy. • The results indicated that the classification performance of the proposed method was significantly higher than other comparison methods. • The experimental results on two public datasets showed that the proposed method performed better than the comparison methods, and can achieve higher classification accuracy. The non-stationary of Motor Imagery (MI) electroencephalogram (EEG) signals makes the traditional machine learning methods ineffective in EEG recognition across time, which limits the practicability of Motor Imagery Brain-computer Interface(MI BCI). Transfer learning makes the classification model of non-stationary MI EEG signals collected and trained at different times universal. In this paper, we proposed a new deep transfer neural network model that can effectively utilize labeled EEG data from previous times, thereby achieving good recognition performance for a small number of labeled EEG signals at the current time. Firstly, a filter bank was used to filter the original motor imagery EEG, and the filtered EEG retained the key features of each domain through the domain specific attention module. Then designed two domain adaptation modules simultaneously. The first domain adaptation module selected similar source domain data, and the second domain adaptation module minimized the difference between the source and target domains. At last, two adversarial classifiers were used to improve the accuracy and robustness of the classification. The results of average classification accuracy, feature visualization, and paired t -test on two public datasets all indicated that the classification performance of the proposed method was significantly higher than other comparison methods. The experimental results on two public datasets showed that the proposed method performed better than the comparison methods when classifying only a small number of labeled MI EEG data, and can achieve higher classification accuracy. This method has practical application value in the rehabilitation of stroke patients. [ABSTRACT FROM AUTHOR]

• Subspace-based method (SM) & Difference subspace method (DSM) are proposed for the motor imagery (MI) events classification in EEG signals. • Principal component analysis (PCA) is used to learn the subspaces in both SM and DSM-based classification. • The efficiency of the proposed methods is evaluated on two benchmark datasets (BCI competition III and IV). • The mean classification accuracy and Cohen's kappa coefficient for BCI competition III and IV datasets were 83.97% & 0.68, and 86.22% & 0.72, respectively. Motor imagery (MI) electroencephalography (EEG) based brain-computer interface (BCI) systems are in great demand for many real-time applications both in the medical and consumer sectors. However, the processing of MI EEG data is challenging, and many of the approaches are complex and suffer from poor classification accuracy due to the unstability of EEG signals. To enhance the classification accuracy, a novel computationally efficient difference subspace method (DSM) of classification is proposed in this study. The proposed method is driven directly by the data without the necessity of extracting any handcrafted features. Subspaces are learned using principal component analysis (PCA). A search strategy based on overlapping criteria is used for selecting the subspace dimensions. The similarity between the two subspaces, measured based on the canonical angles, is used as the criteria to classify MI activity. The proposed method is implemented on BCI competition III and IV datasets, and the performance is evaluated on the basis of classification accuracy (%CA), Cohen's kappa coefficient, F1-score etc., By retention of only the difference components between the two groups of data after removing the common components, the difference subspace method could do dominatingly well in classifying the left and right hand MI classification compared to the subspace method (SM) and many of state-of-the-art techniques. The results show that the DSM achieves an average classification accuracy and Cohen's kappa coefficient for BCI competition III and IV datasets as 83.97% & 0.68 and 86.22% & 0.72, respectively. [ABSTRACT FROM AUTHOR]

• Subspace-based method (SM) & Difference subspace method (DSM) are proposed for the motor imagery (MI) events classification in EEG signals. • Principal component analysis (PCA) is used to learn the subspaces in both SM and DSM-based classification. • The efficiency of the proposed methods is evaluated on two benchmark datasets (BCI competition III and IV). • The mean classification accuracy and Cohen's kappa coefficient for BCI competition III and IV datasets were 83.97% & 0.68, and 86.22% & 0.72, respectively. Motor imagery (MI) electroencephalography (EEG) based brain-computer interface (BCI) systems are in great demand for many real-time applications both in the medical and consumer sectors. However, the processing of MI EEG data is challenging, and many of the approaches are complex and suffer from poor classification accuracy due to the unstability of EEG signals. To enhance the classification accuracy, a novel computationally efficient difference subspace method (DSM) of classification is proposed in this study. The proposed method is driven directly by the data without the necessity of extracting any handcrafted features. Subspaces are learned using principal component analysis (PCA). A search strategy based on overlapping criteria is used for selecting the subspace dimensions. The similarity between the two subspaces, measured based on the canonical angles, is used as the criteria to classify MI activity. The proposed method is implemented on BCI competition III and IV datasets, and the performance is evaluated on the basis of classification accuracy (%CA), Cohen's kappa coefficient, F1-score etc., By retention of only the difference components between the two groups of data after removing the common components, the difference subspace method could do dominatingly well in classifying the left and right hand MI classification compared to the subspace method (SM) and many of state-of-the-art techniques. The results show that the DSM achieves an average classification accuracy and Cohen's kappa coefficient for BCI competition III and IV datasets as 83.97% & 0.68 and 86.22% & 0.72, respectively. [ABSTRACT FROM AUTHOR]

• A signal prediction-based method is proposed for motor imagery classification. • The proposed method uses EEG recorded from only a small number of electrodes with fixed locations as the input. • The proposed method can accurately predict full-channel EEG from the recorded few-channel EEG. • The proposed method is comparable to or even superior to the traditional method using full-channel EEG directly. Motor imagery (MI)-based brain-computer interface (BCI) usually uses EEG signals recorded from numerous scalp electrodes for accurate MI classification, which requires long preparation time and high cost, hindering its practical application. A novel method based on signal prediction was proposed in this study to achieve high MI classification accuracy by using EEG recorded from only a small number of electrodes. The signal prediction model was built by using regression technique to estimate the full-channel EEG signals (e.g., recorded from 19, 22, and 29 electrodes) from the few-channel EEG signals on the central region, and then the predicted full-channel EEG signals were used for MI feature extraction and classification. The proposed methods based on multiple linear regression, ridge regression and extreme learning machine regression were compared with the traditional methods that directly used recorded few-channel and full-channel EEG signals, respectively, on three public EEG datasets. The proposed method significantly outperformed the traditional method that directly used few-channel EEG, and was comparable to or even superior to the traditional method that directly used full-channel EEG. The proposed method can accurately predict full-channel EEG and achieve high classification accuracy with only a small number of electrodes with fixed locations, which has great potential in building MI-based BCIs that meet the requirements of practical applications. [ABSTRACT FROM AUTHOR]

In this study, three popular signal processing techniques (Empirical Mode Decomposition, Discrete Wavelet Transform, and Wavelet Packet Decomposition) were investigated for the decomposition of Electroencephalography (EEG) Signals in Brain Computer Interface (BCI) system for a classification task. Publicly available BCI competition III dataset IVa, a multichannel 2-class motor-imagery dataset, was used for this purpose. Multiscale Principal Component Analysis method was applied for the purpose of noise removal. In addition, different sets of features were formed to examine the effect of a particular group of features. The parameter selection process for signal decomposition methods was thoroughly explained as well. Our results show that the combination of Multiscale Principal Component Analysis de-noising and higher order statistics features extracted from wavelet packet decomposition sub-bands resulted in highest average classification accuracy of 92.8%. Our study is one among very few that provides a comprehensive comparison between signal decomposition methods in combination with higher order statistics in classification of BCI signals. In addition, we stressed the importance of higher frequency ranges in improving the classification task for EEG signals in Brain Computer Interface Systems. Obtained results indicate that the proposed model has the potential to obtain a reliable classification of motor imagery EEG signals, and can thus be used as a practical system for controlling a wheelchair. It can also further enhance the current rehabilitation therapies where appropriate feedback is delivered once the individual executes the correct movement. In that way, motor rehabilitation outcomes may improve over time. [ABSTRACT FROM AUTHOR]

• The proposed lightweight CNN based on parallel structure improves the MI EEG decoding performance of the existing shallow networks. • The experimental results revealed that it has excellent decoding performance on benchmark dataset compared with mainstream wide or deep and hybrid networks. • The proposed model maintains a good balance between performance and complexity, obtains a satisfactory decoding performance with a low resource cost. • It is friendly for BCI hardware loaded with the algorithms because of low computer resource. Deep learning is an important pathway for investigation of motor imagery signal classification. Nevertheless, maintaining a good compromise between performance and computational cost has been a major challenge in developing deep models for decoding motor imagery EEG. In this paper, a novel shallow double-branch convolutional neural network (DSCNN) is proposed for four-class motor imagery classification. The proposed CNN adopts parallel extraction of two branches to improve classification accuracy. Meanwhile, in order to constrain the depth of the whole network, the left branch only contained two single temporal and spatial convolutional layers to extract common EEG features. Similarly, the right branch first introduced 1D convolution to exploit the channel dependency and temporal features across multiple time-scales, secondly a depth-wise separable convolutional layer was applied for optimizing EEG signal series. Then the feature representation for final classification was obtained by merging intermediate features extracted from the two branches. Also, the DSCNN is an end-to-end decoder, as it employs the raw EEG data as inputs and does not require additional complex preprocessing. The proposed model was evaluated on public benchmark BCI competition IV dataset 2a and achieved in terms of accuracy is 85% and kappa value is 0.79. Compared with other state-of-the-art algorithms, the experiment results reveal that the DSCNN has higher decoding accuracy and robustness, as well as a 10% improvement in accuracy than the single general shallow model. Furthermore, as a lightweight architecture, the DSCNN relies on a lower computational power than similar mainstream models, which is more in line with the requirements of low delay and real-time performance in practical BCI applications. [ABSTRACT FROM AUTHOR]

• An efficient multi-scale CNN(MS-CNN) model has been proposed with intrinsic feature integration for motor imagery EEG subject classification in brain-machine interfaces. • Muti-scale convolution block (MSCB) has been designed which can extract the distinguishable features of several non-overlapping canonical frequency bands of EEG signal from multiple scales, and enhance the overall model performance. • It is found that, proposed model records mean average accuracy of 93.74% which is 9.4% average increase (up to 18.1% in subject specific case) compared to the state-of-the art models in BCI competition IV2b dataset. Electroencephalogram (EEG) based motor imagery (MI) classification is an important aspect in brain-machine interfaces (BMIs) which bridges between neural system and computer devices decoding brain signals into recognizable machine commands. However, the MI classification task is challenging due to inherent complex properties, inter-subject variability, and low signal-to-noise ratio (SNR) of EEG signals. To overcome the above-mentioned issues, the current work proposes an efficient multi-scale convolutional neural network (MS-CNN) which can extract the distinguishable features of several non-overlapping canonical frequency bands of EEG signals from multiple scales for MI-BCI classification. In the framework, discriminant user-specific features have been extracted and integrated to improve the accuracy and performance of the CNN classifier. Additionally, different data augmentation methods have been implemented to further improve the accuracy and robustness of the model. The model achieves an average classification accuracy of 93.74 % and Cohen's kappa-coefficient of 0.92 on the BCI competition IV2b dataset outperforming several baseline and current state-of-the-art EEG-based MI classification models. The proposed algorithm effectively addresses the shortcoming of existing CNN-based EEG-MI classification models and significantly improves the classification accuracy. The current framework can provide a stimulus for designing efficient and robust real-time human-robot interaction. [ABSTRACT FROM AUTHOR]

• An optimized Gaussian Mixture Model (GMM) clustering technique is proposed for clustering EEG. • The technique exhibits low sensitivity to outliers within clusters. • The outliers handling mechanism obtained by the proposed technique is applicable to other clustering algorithms. • Both the simulation and real EEG data applications revealed that our proposed technique was less influenced by the outliers compared with traditional GMM. The Gaussian mixture model (GMM) is utilized to illustrate the possibility of applying probabilistic models to data clustering and provide an efficient method for processing EEG signals. However, the existence of outliers in EEG will reduce the robustness of GMM and affect the clustering results. In this paper, an optimized GMM clustering technique that exhibits low sensitivity with respect to outliers within clusters has been proposed, which eliminates deviations caused by outliers. Experimental research is conducted to verify the effectiveness of the proposed methods. The results are supported by statistical inference and characteristic curves. The proposed model outperforms traditional methods by achieving the accuracy of 84.4%, 77.2%, 81.6%, and 88.3% on the BCI Competition IV Dataset 1. Furthermore, we combined this improved method with the state-of-the-art clustering methods, the experiments on public datasets show a comparable improvement in accuracy. This paper provides an optimized GMM clustering technique that exhibits low sensitivity to outliers, which may promote the development of BCI applications. [ABSTRACT FROM AUTHOR]

The application of brain-computer interface (BCI) has always been limited by low decoding accuracy due to excessive noise in electroencephalogram (EEG) signals. The traditional methods employ some representative features while losing too much information. Deep learning methods have achieved good results, but subject to the insufficient ability of extracting discriminative features from EEG. In this paper, we propose a novel decoding framework that effectively uses spatial and temporal information by time-contained spatial filtering and spatial–temporal analysis network (TSF-STAN) for EEG multi-classification tasks. Firstly, the TSF with the joint one-versus-rest (Joint-OVR) strategy is given to map the signal to a new space, where each category can be more easily distinguished, while preserving the time-domain characteristics. Next, the STAN is designed to extract discriminative spatial and temporal features further with convolutional layers, and then perform classification. Detailed experiments have been carried out to verify the effectiveness of our framework on BCI competition IV-2a and IV-2b datasets of motor imagery (MI) EEG. The results show that our method has outperformed recent outstanding algorithms, with the average accuracy of 83.0% on IV-2a and 88.0% on IV-2b. Spatial and temporal information is well used to obtain better performance, which has a good potential of EEG decoding for application of BCI. [ABSTRACT FROM AUTHOR]

Deep learning has been used popularly and successfully in state of art researches to classify different types of images. However, so far, the applications of deep learning methods for the electroencephalography (EEG) motor imagery classification is very limited. In this study, a pre-processing algorithm is proposed for the EEG signals representation. Then, a parallel convolutional neural network (PCNN) architecture is proposed to classify motor imagery signals. For the raw EEG signals representation, a new form of the images is created to combine spatial filtering and frequency bands extracting together. By feeding the represented images into the PCNN, it stacks three unique sub-models together aiming to optimize the performance of classification. The average accuracy of the proposed method achieves 83.0 ± 3.4% on BCI Competition IV dataset 2b, which outperforms the compared methods at least 5.2%. The average Kappa value of the proposed method achieves 0.659 ± 0.067 on dataset 2b, providing at least 20.5% improvement with respect to the compared algorithms. The results show that the proposed method performs better in EEG motor imagery signals classification. [ABSTRACT FROM AUTHOR]

• The combination of LAE and CSP performs better than either method. • LAE-CSP achieves a mean accuracy of 78.52 % over 28 subjects using 10 training trials. • CSP, FBRCSP, EA-CSP and LAE were compared with LAE-CSP using 4 different datasets. The common spatial pattern (CSP) and its variants are popular in the EEG-based motor imagery BCIs. However, this method has some drawbacks, especially when a few labeled samples are available. The local activities estimation (LAE) method works well with small training sets, but it is more sensitive to the position of the electrodes. Here, we suggest a combination of the LAE and CSP, namely LAE-CSP, which performs better than both methods. In LAE-CSP, the EEG signal passes through regularized CSP and LAE spatial filters after band-pass filtering and then the data dimension are reduced based on the physiological data. Afterwards, the features are extracted using fast Fourier transform (FFT). In this work, CSP, FBRCSP, EA-CSP, LAE and LAE-CSP, methods were evaluated and compared. Three sets of motor imagery data from BCI competition III and IV along with Cho et al. dataset, including EEG signals from 28 subjects were used in this study. For each dataset, the training set were selected in 21 different sizes. LAE-CSP performs better than all tested methods. Particularly it has good performance using only ten labeled samples per class, with an average accuracy of almost 80 %. LAE-CSP reliably enhances the performance of the motor imagery-based brain-computer interfaces. The LAE-CSP takes advantage of the LAE and CSP and compensates for the drawbacks of both methods. [ABSTRACT FROM AUTHOR]

• A BCI system is proposed for neuro-aid system based on TSD based feature extraction. • With the extracted features, QDA classifier is adopted to generate control signals for regulating BLDC motor. • The rotation of BLDC motor is governed by the QDA classifier output for different classes of MI data. • The proposed scheme has been validated on 1 kW motor drive system. Brain-computer interface (BCI) systems are used to establish communication pathways between the brain signals and a peripheral device. One such pathway can be fashioned between these signals and an electric drive to be further used for any neuro-aid application. Brain signals, which in this study are the electroencephalogram (EEG) signals from a specified focal area, get synchronized and desynchronized upon the occurrence of different motor imagery (MI). Some specific features of the neuronal brain activities occurring during MI can be extracted to get the relevant information from them. This paper proposes an algorithm based on time-varying signed distance (TSD) to obtain well-separated EEG data for the process of feature extraction. Post TSD application, power spectral density (PSD) based features were extracted, followed by using the measure of cross-correlation to select the most dominant PSD based features. A quadratic discriminant analysis (QDA) is adopted to classify the test data and to generate further control signals for regulating the motion of brushless DC (BLDC) motor. The rotation of the motor is governed by the different classes of MI data. For the experimental dataset, the accuracy of the classifier was found to be 83.3 %, 71.6 %, and 78.3 %, respectively, for three different subjects. The case corroborates the efficacy of the proposed algorithm for practical applications. Experimental validation of the proposed system is presented on a 1 kW brushless DC (BLDC) motor drive system. [ABSTRACT FROM AUTHOR]

It is still a big challenge to extract effective features from raw electroencephalogram (EEG) signals and then to improve classification accuracy of motor imagery (MI) applications on brain–computer interface (BCI). Traditionally, features are extracted from time, frequency, or time–frequency domains for MI pattern recognition achieved by classifiers. However, the features from a single domain can only provide limited information useful for final classification, thus may lead to unsatisfactory performance. Also, the features from different domains may contain different and complementary information for MI pattern classification. Therefore, it is necessary to fuse them to enhance pattern classification capability. To this end, a two-level feature extraction approach based on sparse representation (SR) for MI EEG signals is proposed in this paper, which mainly consists of multi-domain feature extraction and sparse feature fusion. In the proposed method, multi-domain features, including Hjorth, the power spectrum estimation via maximum entropy, and time–frequency energy, are first extracted as the initial feature space. Then sparse representation is used to fuse extracted multi-domain features to obtain low-dimensional informative features with better discriminative ability. Finally, these transformed low-dimensional features are fed into a classifier to identify different MI patterns. The proposed method is evaluated using the public competition datasets (BCI2008), and achieved the average accuracy of over 79%. The results indicate that compared with existing methods and single domain-based feature extraction methods, the proposed method achieved better classification performance. • A two-level multi-domain feature extraction approach for motor imagery is proposed. • The proposed method with three different classifiers obtained superior performance. • It outperformed single-domain and multi-domain feature extraction methods. • It has great potentials in practical BCI applications with high accuracy. [ABSTRACT FROM AUTHOR]