Complex Network Analysis Of Motor Imagery And P300 EEG Signals

Brain-computer interface(BCI)is a multidisciplinary technology which enables users to interact with the environment without relying on neural pathways and muscles.BCI systems have been extensively applied in fields of assisting the disabled and the old,rehabilitation and military.The study of BCI technology not only improves the efficiency of human-computer interaction,but also helps to understand the mechanism of brain activity during high-level cognitive tasks.Due to the fact that the human brain is a complex non-linear system,mathematical models are usually awkward to analyze its dynamic characteristics.Therefore,how to effectively classify EEG signals and how to fuse multi-channel EEG signals to reveal the cognitive mechanism of human brain are challenging problems still remain to be solved.Experiments are designed and conducted to record electroencephalograph(EEG)signals during left and right hand movement imagery tasks based on BCI systems.Aiming at improving the classification accuracy of motor imagery signals,common spatial pattern(CSP)method is combined with convolutional neural network(CNN)in order to further extract effective features.The origin EEG signals are spatial filtered and then formed into input image of CNN.The input image is gradually transformed into a more advanced feature representation through convolution and pooling operations.Results demonstrate that our CSP-CNN model has a better classification performance compared with traditional methods,the average classification accuracy is increased about 7%.Aiming at probing into brain activities during left and right hand motor imagery,a method of constructing functional brain network based on wavelet time-frequency analysis is proposed in this dissertation.More specifically,the energy sequence of each channel is firstly extracted via continuous wavelet transform method,then brain network is constructed by treating the channels of scalp EEG as nodes and determining edges in terms of the 2-norm distance between energy sequences of each channel.The functional connectivity of derived brain networks could be interpreted via network measures.Results demonstrate that the betweenness centralities of nodes located at contralateral sensorimotor regions are higher,which means that they are more likely to be activated as hub nodes to control information flows during motor imagery tasks.In order to explore the changes of cognitive process caused by fatigue symptom,P300 signals recording experiments are firstly designed and conducted.Then functional brain networks are constructed based on wavelet multi-resolution technique.Small-world indices of brain networks under normal and fatigue state are also calculated.Results demonstrate that the brain network reconfiguration in response to the cognitive task in fatigue status is reflected as the increase of the small-worldness,which indicates that the network is highly segregated and integrated simultaneously under fatigue.