Multi-mode Analysis Of Eeg And Application In BCI

The 21st century has been considered the"brain science era", while exploring and re-vealing the mysteries of the brain has become the most significant challenge of natural sci-ence. Electroencephalogram(EEG) is the recording of electrical activity produced by thefiring of neurons within the brain, therefore it contains plenty of brain state information.EEG based brain computer interface(BCI) system, which provides a new directly informa-tion interaction and communication channels between brain and computer, has become animportant means to understand and enhance brain function. The BCI research has drawn at-tention of scientists in brain science, rehabilitation engineering, biomedical engineering andintelligent information processing. The performance of BCI system depends on the qualityof EEG feature representation and the accuracy of pattern classification of the recorded sin-gle trial EEG. However, due to the non-stability and low signal to noise ratio(SNR) of EEGsignals, it is very difficult to extract reliable features from EEG signals. Therefore, therestill exist many difficulties and challenges in the development and application of BCI. In thispaper, we mainly focus on some key techniques in BCI research, including the scheme forsingle trial EEG classification, feature extraction algorithms and online BCI systems, andemphasize particularly on multi-mode analysis of EEG signals.The main contributions and innovations of this paper have been listed as below:1. Non-negative multi-way factorization (NMWF) based scheme for single trial EEGclassification:In this paper, based on the non-negative multi-way factorization al-gorithm, a scheme is proposed for class discrimination of single trial EEG data. Itcan analyze high-way EEG data, and extract the temporal, spatial, spectral project pat-terns and features for classification. Experiment results present that the tensor basedscheme is efficient for EEG data analysis and classification by the multi-way featureextraction.2. General tensor discriminant analysis (GTDA) based scheme for single trial EEG clas-sification: Popular classification algorithms usually highly depend on the prior neu-rophysiologic knowledge for noise removal. In this paper, a GTDA based scheme is proposed for single trial EEG classification,which could extract multi-linear discrim-inative subspace. Computer simulations confirm the effectiveness and the robustnessof the proposed tensor scheme in extracting classification features and identifying dis-criminative properties in the case of lacking prior neurophysiologic knowledge.3. Regularized tensor discriminant analysis (RTDA) based scheme for single trial EEGclassification in BCI: RTDA algorithm is proposed for EEG signals analysis. By multi-way discriminative analysis and regularization terms incorporating reasonable assump-tions about EEG signals, RTDA overcomes the difficulties in extracting class featuresfrom the EEG signal with its low SNR and high data dimensionality. Furthermore,the scheme could identify the most important channels for classification, and then beapplied to channel selection in BCI. It is has been proved that the scheme is effectivefor EEG single trail classification, and especially the number of used channels can begreatly reduced with very little loss in performance.4. EEG phase information multi-mode analysis:In neuroscience, phase in EEG is dis-covered to contain abundant information about the neural electric activity. In thispaper, we define phase interval value (PIV) to explore the phase information of EEGfrom a new perspective and propose a computational model based on ordered Paral-lel Factors (PARAFAC) algorithm to extract features from multi-way PIV for singletrial EEG classification. Computer simulations demonstrate that phase information iseffective for EEG classification, and the related distribution information of features intemporal, spatial, spectral modes can be obtained from phase.5. On-line BCI system platform design and implementation: In this paper, a ?exible andextendible platform for online BCI system is designed and developed, which can pro-vide a BCI research platform for experiment development and algorithm verification.Multi-mode EEG characteristics can be visualized in the platform to reveal and under-stand the dynamic features of EEG in specific mental task, and then be applied to buildthe relationship among mental tasks, features representation, and control decisions.The platform provides command translation interface, therefore we can establish theconnection between mental tasks and system control commands ?exibly. Using thisinterface, we design interactive training system prototype, and realize two self-pacedsynchronous BCI applications based on multi-class motor imagery tasks, the onlinegame control in virtual environment and remote car driving in real environment. Goodperformance can be achieved after interactive training. In conclusion, this paper investigates EEG patterns and dynamic features during specificmental tasks, and proposes multi-mode analysis method to extract temporal, spatial, spectraldiscriminative patterns from complicated EEG signals, which is important to improve rateof information transmission and practicability of BCI. Furthermore, this paper designs inter-active training system prototype, and realizes self-paced synchronous BCI systems in virtualand real environments. These work provide theory basis and technology prototype for EEGsignals analysis and BCI development.