A brain functional network feature extraction method based on directed transfer function and graph theory for MI-BCI decoding tasks.

Background: The development of Brain-Computer Interface (BCI) technology has brought tremendous potential to various fields. In recent years, prominent research has focused on enhancing the accuracy of BCI decoding algorithms by effectively utilizing meaningful features extracted from electroencephalographic (EEG) signals. Objective: This paper proposes a method for extracting brain functional network features based on directed transfer function (DTF) and graph theory. The method incorporates the extracted brain network features with common spatial pattern (CSP) to enhance the performance of motor imagery (MI) classification task. Methods: The signals from each electrode of the EEG, utilizing a total of 32 channels, are used as input signals for the network nodes. In this study, 26 healthy participants were recruited to provide EEG data. The brain functional network is constructed in Alpha and Beta bands using the DTF method. The node degree (ND), clustering coefficient (CC), and global efficiency (GE) of the brain functional network are obtained using graph theory. The DTF network features and graph theory are combined with the traditional signal processing method, the CSP algorithm. The redundant network features are filtered out using the Lasso method, and finally, the fused features are classified using a support vector machine (SVM), culminating in a novel approach we have termed CDGL. Results: For Beta frequency band, with 8 electrodes, the proposed CDGL method achieved an accuracy of 89.13%, a sensitivity of 90.15%, and a specificity of 88.10%, which are 14.10, 16.69, and 11.50% percentage higher than the traditional CSP method (75.03, 73.46, and 76.60%), respectively. Furthermore, the results obtained with 8 channels were superior to those with 4 channels (82.31, 83.35, and 81.74%), and the result for the Beta frequency band were better than those for the Alpha frequency band (87.42, 87.48, and 87.36%). Similar results were also obtained on two public datasets, where the CDGL algorithm's performance was found to be optimal. Conclusion: The feature fusion of DTF network and graph theory features enhanced CSP algorithm's performance in MI task classification. Increasing the number of channels allows for more EEG signal feature information, enhancing the model's sensitivity and discriminative ability toward specific activities in brain regions. It should be noted that the functional brain network features in the Beta band exhibit superior performance improvement for the algorithm compared to those in the Alpha band.

A fused multi-subfrequency bands and CBAM SSVEP-BCI classification method based on convolutional neural network.

For the brain-computer interface (BCI) system based on steady-state visual evoked potential (SSVEP), it is difficult to obtain satisfactory classification performance for short-time window SSVEP signals by traditional methods. In this paper, a fused multi-subfrequency bands and convolutional block attention module (CBAM) classification method based on convolutional neural network (CBAM-CNN) is proposed for discerning SSVEP-BCI tasks. This method extracts multi-subfrequency bands SSVEP signals as the initial input of the network model, and then carries out feature fusion on all feature inputs. In addition, CBAM is embedded in both parts of the initial input and feature fusion for adaptive feature refinement. To verify the effectiveness of the proposed method, this study uses the datasets of Inner Mongolia University of Technology (IMUT) and Tsinghua University (THU) to evaluate the performance of the proposed method. The experimental results show that the highest accuracy of CBAM-CNN reaches 0.9813 percentage point (pp). Within 0.1-2 s time window, the accuracy of CBAM-CNN is 0.0201-0.5388 (pp) higher than that of CNN, CCA-CWT-SVM, CCA-SVM, CCA-GNB, FBCCA, and CCA. Especially in the short-time window range of 0.1-1 s, the performance advantage of CBAM-CNN is more significant. The maximum information transmission rate (ITR) of CBAM-CNN is 503.87 bit/min, which is 227.53 bit/min-503.41 bit/min higher than the above six EEG decoding methods. The study further results show that CBAM-CNN has potential application value in SSVEP decoding.

A Fused Multidimensional EEG Classification Method Based on an Extreme Tree Feature Selection.

When a brain-computer interface (BCI) is designed, high classification accuracy is difficult to obtain for motor imagery (MI) electroencephalogram (EEG) signals in view of their relatively low signal-to-noise ratio. In this paper, a fused multidimensional classification method based on extreme tree feature selection (FMCM-ETFS) is proposed for discerning motor imagery EEG tasks. First, the EEG signal was filtered by a Butterworth filter for preprocessing. Second, C3, C4, and CZ channels were selected to extract time-frequency domain and spatial domain features using autoregressive (AR), common spatial pattern (CSP), and discrete wavelet transform (DWT). The extracted features were fused for a further feature elimination. Then, the features were selected using three feature selection methods: recursive feature elimination (RFE), principal component analysis method (PCA), and extreme trees (ET). The selected feature vectors were classified using support vector machines (SVM). Finally, a total of twelve subjects' EEG data from Inner Mongolia University of Technology (IMUT data), the 2nd BCI competition in 2003, and the 4th BCI competition in 2008 were employed to show the effectiveness of this proposed FMCM-ETFS method. The results show that the classification accuracy using the multidimensional fused feature extraction (AR + CSP + DWT) is 3%-20% higher than those using the aforementioned three single feature extractions (AR, CSP, and DWT). Extreme trees (ET), which is a sort of tree-based model method, outperforms RFE and PCA by 1%-9% in term of classification accuracies, when these three methods were applied to the procedure of feature extraction, respectively.

A Feature Extraction Algorithm of Brain Network of Motor Imagination Based on a Directed Transfer Function.

Aiming at the feature extraction of left- and right-hand movement imagination EEG signals, this paper proposes a multichannel correlation analysis method and employs the Directed Transfer Function (DTF) to identify the connectivity between different channels of EEG signals, construct a brain network, and extract the characteristics of the network information flow. Since the network information flow identified by DTF can also reflect indirect connectivity of the EEG signal networks, the newly extracted DTF features are incorporated into the traditional AR model parameter features and extend the scope of feature sets. Classifications are carried out through the Support Vector Machine (SVM). The classification results show the enlarged feature set can significantly improve the classification accuracy of the left- and right-hand motor imagery EEG signals compared to the traditional AR feature set. Finally, the EEG signals of 2 channels, 10 channels, and 32 channels were selected for comparing their different effects of classifications. The classification results showed that the multichannel analysis method was more effective. Compared with the parameter features of the traditional AR model, the network information flow features extracted by the DTF method also achieve a higher classification effect, which verifies the effectiveness of the multichannel correlation analysis method.

A classification algorithm of an SSVEP brain-Computer interface based on CCA fusion wavelet coefficients.

BACKGROUND: In the study of brain-computer interfaces (BCIs) based on steady-state visual evoked potentials (SSVEPs), how to improve the classification accuracies of BCIs has always been the focus of researchers. Canonical correlation analysis (CCA) is widely used in BCI systems of SSVEPs because of its rapidity and scalability. However, the classical CCA algorithm always encounters the difficulty of low accuracy in a short time. NEW METHOD: For targetless stimuli, this paper proposes a fusion algorithm (CCA-CWT-SVM) that is combined with CCA, a continuous wavelet transform, and a support vector machine (SVM) to improve the low classification accuracies when a single feature extraction method is used. RESULTS: This fusion algorithm achieves high accuracies and information transfer rates (ITRs) in the SSVEP paradigm with few targets. COMPARISON WITH EXISTING METHODS AND CONCLUSIONS: Through the study of 400 groups of experimental data from 10 subjects, the results show that CCA-CWT-SVM has a classification accuracy of 91.76% within 2 s and an ITR of 48.92 bits/min, which are 10.88% and 13.18 bits/min higher than those of the standard CCA. Compared with a mainstream EEG decoding algorithm, filter bank canonical correlation analysis (FBCCA), the classification accuracy and ITR of the CCA-CWT-SVM algorithm also improved (4.45% and 5.69 bit/min, respectively). Using a dataset from Tsinghua University (THU), we also showed that the fusion algorithm is better than the classical algorithms. The CCA-CWT-SVM algorithm obtained an 89.1% accuracy and a 39.91 bit/min ITR in a time window of 2 s. The results were significantly improved compared with those of CCA and the FBCCA (CCA: 79.44% and 28.23 bits/min, FBCCA: 84.03% and 33.4 bits/min). Hence, this work provides an experimental basis for designing an SSVEP-based BCI system with a high task classification accuracy in some crucial biomedical applications.

Identifying the pulsed neuron networks' structures by a nonlinear Granger causality method.

BACKGROUND: It is a crucial task of brain science researches to explore functional connective maps of Biological Neural Networks (BNN). The maps help to deeply study the dominant relationship between the structures of the BNNs and their network functions. RESULTS: In this study, the ideas of linear Granger causality modeling and causality identification are extended to those of nonlinear Granger causality modeling and network structure identification. We employed Radial Basis Functions to fit the nonlinear multivariate dynamical responses of BNNs with neuronal pulse firing. By introducing the contributions from presynaptic neurons and detecting whether the predictions for postsynaptic neurons' pulse firing signals are improved or not, we can reveal the information flows distribution of BNNs. Thus, the functional connections from presynaptic neurons can be identified from the obtained network information flows. To verify the effectiveness of the proposed method, the Nonlinear Granger Causality Identification Method (NGCIM) is applied to the network structure discovery processes of Spiking Neural Networks (SNN). SNN is a simulation model based on an Integrate-and-Fire mechanism. By network simulations, the multi-channel neuronal pulse sequence data of the SNNs can be used to reversely identify the synaptic connections and strengths of the SNNs. CONCLUSIONS: The identification results show: for 2-6 nodes small-scale neural networks, 20 nodes medium-scale neural networks, and 100 nodes large-scale neural networks, the identification accuracy of NGCIM with the Gaussian kernel function was 100%, 99.64%, 98.64%, 98.37%, 98.31%, 84.87% and 80.56%, respectively. The identification accuracies were significantly higher than those of a traditional Linear Granger Causality Identification Method with the same network sizes. Thus, with an accumulation of the data obtained by the existing measurement methods, such as Electroencephalography, functional Magnetic Resonance Imaging, and Multi-Electrode Array, the NGCIM can be a promising network modeling method to infer the functional connective maps of BNNs.

Discerning Functional Connections in the Pulsed Neural Networks with the Dynamic Bayesian Network Structure Search Method Based on a Genetic Algorithm.

It is important to explore potential structural characteristics of biological networks and regulatory mechanisms of network behaviors at the system level. In this study, a dynamic Bayesian network structure search method (DBNSSM) based on a genetic algorithm is employed to infer and locate functional connections in pulsed neural networks (PNNs) as typical artificial neural networks. In the process of network structure searching, a minimum description length score is calculated for each candidate network structure. The score indicates two characteristics of the network structure: (1) the likelihood based on network dynamic response data and (2) the complexity. Both should be considered together on selecting network structures. The DBNSSM is applied to analyze time-series data from PNNs, thereby discerns functional connections showing network structures collectively. It is feasible to analyze multichannel electrophysiological data of biological neural networks using the DBNSSM.

Identification of feedback loops in neural networks based on multi-step Granger causality.

MOTIVATION: Feedback circuits are crucial network motifs, ubiquitously found in many intra- and inter-cellular regulatory networks, and also act as basic building blocks for inducing synchronized bursting behaviors in neural network dynamics. Therefore, the system-level identification of feedback circuits using time-series measurements is critical to understand the underlying regulatory mechanism of synchronized bursting behaviors. RESULTS: Multi-Step Granger Causality Method (MSGCM) was developed to identify feedback loops embedded in biological networks using time-series experimental measurements. Based on multivariate time-series analysis, MSGCM used a modified Wald test to infer the existence of multi-step Granger causality between a pair of network nodes. A significant bi-directional multi-step Granger causality between two nodes indicated the existence of a feedback loop. This new identification method resolved the drawback of the previous non-causal impulse response component method which was only applicable to networks containing no co-regulatory forward path. MSGCM also significantly improved the ratio of correct identification of feedback loops. In this study, the MSGCM was testified using synthetic pulsed neural network models and also in vitro cultured rat neural networks using multi-electrode array. As a result, we found a large number of feedback loops in the in vitro cultured neural networks with apparent synchronized oscillation, indicating a close relationship between synchronized oscillatory bursting behavior and underlying feedback loops. The MSGCM is an efficient method to investigate feedback loops embedded in in vitro cultured neural networks. The identified feedback loop motifs are considered as an important design principle responsible for the synchronized bursting behavior in neural networks.

An optimally evolved connective ratio of neural networks that maximizes the occurrence of synchronized bursting behavior.

BACKGROUND: Synchronized bursting activity (SBA) is a remarkable dynamical behavior in both ex vivo and in vivo neural networks. Investigations of the underlying structural characteristics associated with SBA are crucial to understanding the system-level regulatory mechanism of neural network behaviors. RESULTS: In this study, artificial pulsed neural networks were established using spike response models to capture fundamental dynamics of large scale ex vivo cortical networks. Network simulations with synaptic parameter perturbations showed the following two findings. (i) In a network with an excitatory ratio (ER) of 80-90%, its connective ratio (CR) was within a range of 10-30% when the occurrence of SBA reached the highest expectation. This result was consistent with the experimental observation in ex vivo neuronal networks, which were reported to possess a matured inhibitory synaptic ratio of 10-20% and a CR of 10-30%. (ii) No SBA occurred when a network does not contain any all-positive-interaction feedback loop (APFL) motif. In a neural network containing APFLs, the number of APFLs presented an optimal range corresponding to the maximal occurrence of SBA, which was very similar to the optimal CR. CONCLUSIONS: In a neural network, the evolutionarily selected CR (10-30%) optimizes the occurrence of SBA, and APFL serves a pivotal network motif required to maximize the occurrence of SBA.

Identification of feedback loops embedded in cellular circuits by investigating non-causal impulse response components.

Feedback circuits are crucial dynamic motifs which occur in many biomolecular regulatory networks. They play a pivotal role in the regulation and control of many important cellular processes such as gene transcription, signal transduction, and metabolism. In this study, we develop a novel computationally efficient method to identify feedback loops embedded in intracellular networks, which uses only time-series experimental data and requires no knowledge of the network structure. In the proposed approach, a non-parametric system identification technique, as well as a spectral factor analysis, is applied to derive a graphical criterion based on non-causal components of the system's impulse response. The appearance of non-causal components in the impulse response sequences arising from stochastic output perturbations is shown to imply the presence of underlying feedback connections within a linear network. In order to extend the approach to nonlinear networks, we linearize the intracellular networks about an equilibrium point, and then choose the magnitude of the output perturbations sufficiently small so that the resulting time-series responses remain close to the chosen equilibrium point. In this way, the impulse response sequences of the linearized system can be used to determine the presence or absence of feedback loops in the corresponding nonlinear network. The proposed method utilizes the time profile data from intracellular perturbation experiments and only requires the perturbability of output nodes. Most importantly, the method does not require any a priori knowledge of the system structure. For these reasons, the proposed approach is very well suited to identifying feedback loops in large-scale biomolecular networks. The effectiveness of the proposed method is illustrated via two examples: a synthetic network model with a negative feedback loop and a nonlinear caspase function model of apoptosis with a positive feedback loop.

Systematic analysis of synchronized oscillatory neuronal networks reveals an enrichment for coupled direct and indirect feedback motifs.

MOTIVATION: Synchronized bursting behavior is a remarkable phenomenon in neural dynamics. So, identification of the underlying functional structure is crucial to understand its regulatory mechanism at a system level. On the other hand, we noted that feedback loops (FBLs) are commonly used basic building blocks in engineering circuit design, especially for synchronization, and they have also been considered as important regulatory network motifs in systems biology. From these motivations, we have investigated the relationship between synchronized bursting behavior and feedback motifs in neural networks. RESULTS: Through extensive simulations of synthetic spike oscillation models, we found that a particular structure of FBLs, coupled direct and indirect positive feedback loops (PFLs), can induce robust synchronized bursting behaviors. To further investigate this, we have developed a novel FBL identification method based on sampled time-series data and applied it to synchronized spiking records measured from cultured neural networks of rat by using multi-electrode array. As a result, we have identified coupled direct and indirect PFLs. CONCLUSION: We therefore conclude that coupled direct and indirect PFLs might be an important design principle that causes the synchronized bursting behavior in neuronal networks although an extrapolation of this result to in vivo brain dynamics still remains an unanswered question.