Research On Steady-state Visual Evoked Potential Of Its Brain Mechanisms And Applications In Brain-computer Interface

Due to the high signal-to-noise ratio and stable spectrum, steady-state visualevoked potential (SSVEP) is widely used in cognitive neuroscience and clinicalresearches. Morever, SSVEP is also used in the brain-computer interface (BCI).SSVEP-based BCI requires little training time and has high information transfer rate(IRT), it becomes an important branch of BCI field. The brain mechanism of SSVEP isnot completely explored, which may hinder the development of SSVEP-based BCI.This dissertation focused on the electroencephalograph (EEG) data analysis, andadopted the graph theoretical network analysis to study the brain mechanism. It alsoutilized the multidimensional information coding and modern signal processingmethod to propose some solutions to some technical problems. The main contents andresults are presented as follows:1. The graph theoretical network analysis was used to study the brain mechanismof SSVEP. Two frequencies (12.5Hz and16.6Hz) were used and eleven subjects wererecruited in the experiment. First, the coherence and sparasity value were used toconstruct the functional brain networks (Network0) of the stimulus frequencies.According to the relationships between the connections and the SSVEP amplitude, theNetwork0was further divided into three subnetworks, i.e. Network1, Network2andNetwork3. Network1was composed of connections in which the connection strengthswere significantly positively correlated with SSVEP amplitude. In Network3, theconnection strengths were significantly negatively correlated with SSVEP amplitude.Network2was composed of the remaining connections for which the strengths werenot significantly correlated with SSVEP responses. After that, the five properities, i.e.clustering coefficients, characteristic path length, global efficiency, local efficiency andmean functional connectivity of the three networks were calculated. Then, therelationships between the SSVEP amplitude and the five properities of the threenetworks were investigated respectively. For all the three networks, the results showedthat the SSVEP amplitude was positively correlated with the clustering coefficients,global efficiency, local efficiency and mean functional connectivity, while negativelycorrelated with characteristic path length. Furthermore, the strengths of theseconnections that significantly correlated with the SSVEP (both positively and negatively) were mainly found to be long-range connections between theparietal-occipital and frontal regions. Using permutation tests, we also found that thedifferences of the subjects’ SSVEP amplitudes were related to the differences of theconnection strengths of the long-range connections between the parietal-occipital andfrontal regions. With the same methods, we analyzed the relationships between thedynamic SSVEP amplitudes and the time-varying nework properties within thesubjects, and got many similar results. These findings indicated that the higher SSVEPamplitude corresponded to the more efficient nework topology. Meanwhile, thesefindings provide new new insights for understanding brain mechanisms of SSVEP.2. With the resting-state EEG data, the coherence and sparasity value were alsoused to construct resting-state brain networks (Network0) of the stimulus frequencies.Then, the three properities, i.e. clustering coefficients, characteristic path length, andmean functional connectivity of the three networks were calculated. With thecorrelation analysis, we found that SSVEP amplitudes of each frequency werenegatively correlated with the mean functional connectivity and clustering coefficient,but positively correlated with characteristic path length. Most of the connections werenegatively correlated with the SSVEP amplitudes, and the main contributors forSSVEP were the connections between the parietal-occipital regions and frontal regions.Furthermore, we found that the BCI performance was related to the averagedresting-state network properities across frequencies, and the classification accuracycould be predicted by three averaged network properities and their combinations couldfurther improved the prediction performance.3. A novel multivariate synchronization index (MSI) for frequency recognitionwas proposed. In this method, the S-estimator was used to estimate the synchronizationbetween multichannel EEG and the reference signals, the latter of which were definedaccording to the stimulus frequency. For the simulation and real data, the proposedmethod showed higher classification accuracy and robustness to noises than the widelyused canonical correlation analysis (CCA) and minimum energy combination (MEC),especially for short data length and a small number of channels. The proposed methodcould potentially enhance the performance of BCI systems.4. By adding the time factor, we proposed to utilize three dimensional information,i.e. temporal information, frequency and spatial information, to code the targets inSSVEP-BCI. In this protocol, the targets were displayed in different positions of the monitor. Each target was coded by one permutation sequence of the stimulusfrequencies cycle by cycle. In each cycle, the frequencies in the permutation sequencewere presented one by one according to the order in the sequence. With coding lengthbeing2, we used two frequencies to realize four targets and showed that new protocolis feasible and efficient with an offline experiment. Based on this new coding protocoland our proposed frequency recognition method, we developed a SSVEP-BCI tocontrol a virtual robot. We got satisfying performance from this system.