Research On Decoding And Correlation Analysis Of Motor-related Neural Signals

Brain-computer interface provides a way to communicate and control between brain and equipment by neural signal decoding and command transfer.Motor-related brain signal originates from brain avtivities during self-paced motion control.It can be used to decode the characteristic neural signal when the brain performs the limbs control,therefore enabling directly control of equipments without spinal cord and peripheral nerve pathways,which is very meaningful for motor-assisted and neurological rehabilitation.According to the way of data acquisition,motor-related brain signals can be divided into electroencephalograms(EEG),which are acquired on the scalp surface,and electrocorticograms(ECoG),local field potentials(LFPs)and single neuron spikes(Spikes),which are acquired by intracerebral microelectrodes.Among them,the motor imagery EEG signal has some advantages,such as high time domain resolution,simple and non-invasive acquisition method,etc.It has been successfully applied to the motor rehabilitation of the disabled,robot/drone control,etc.However it is limited by the low signal to noise ratio(SNR)and difficulty in accuracy decoding.Neural signals acquired by implanted microelectrodes directly reflect the origin neural activity of the brain motor control,and are better in accurate decoding,but limited by signal acquisition.During motor control,the neural signals from multiple levels such as the scalp and cortex are essentially originated from the activities of neurons in brain regions related to the motor cortex.If the correlation between these neural signals was considered when we decoded these neural signals,it would be helpful and meaningful for better analysis of motor-related neural signals.So,this research focuses on the limited decoding capability and unclear mechanisms in motor imagery EEG analysis.In order to solve the above difficulties,we performed decoding and correlation analysis of multi-level motor related neural signals,such as Spike,LFP,ECoG,and EEG.In the aspect of motor imagery EEG signal analysis,we made effort to optimize training procedure and to develop a new decoding algorithms for better feature extraction and decoding capabilities.And then we analysed the feature of functional brain network in the different frequency components of EEG signals.Besides,by using rhesus monkeys as the model,we performed the association analysis between Spike,LFP and ECoG signals to reveal the intrinsic neural mechanism of field potential signal generation.Finally,by using rats as the model,we analyzed the correlation between Spike and EEG signals in the motor cortex under optogenetic stimulation,and the intrinsic neural origins of different frequency components of the EEG signals were discussed.The main contents and results of this thesis are summarized as follows:Chapter 1: Analysis of motor imagery EEG based on subject-specific training optimization and L1 norm sparse regularization.To improve the performance of decoding of motor imagery EEG signal,we optimized the subject training procedure of motor imagery through the guidance of different visual image cues,and analysed the characteristics of EEG rhythm of motor imagery under different visual guiding conditions.We improved the feature extraction procedure by EEG channel optimization based on L1 norm regularization(COL),which reduced the complexity of the classification algorithm and enhanced the generalization.The results show that the subject's motor imagery after proper visual guidance demonstrate the bigger event-related desynchronization(ERD)feature of the ? rhythm,which was increased by approximately 35% in ERD intensity.On the two open source data set for BCI competition III and IV,the classification accuracy results of the COL algorithm were 86.13% and 85.86%,respectively.Moreover it had good generalization performance.Classification accuracy under the small sample of training set was obviously better than the performance of other reported algorithms for feature optimization.This work provided efficient algorithm architecture for accurate decoding of EEG signals.Chapter 2: Analysis of EEG signals and related mechanisms based on network analysisThe ability of brain functional integration was not completely considered in the analysis of motor imagery EEG signals.So,the cross-covariance and phase lag index(PLI)were utilized to construct static brain network and dynamic brain network on the public data sets of motor imagery EEG.And the characteristics of the brain network were analyzed.It was found that the changes of the brain network connection characteristics in different motor imagery states were significant in the ? and ? bands.The local network characteristics of the left and right hemispheres and the global network properties of the brain demonstrated the distinct differences between different motor imagery classes.The connection between different time period showed significant dynamic changes.These results provided a useful reference for a deeper understanding of the brain mechanisms of motor imagery and for better features exaction.Chapter 3: Correlation analysis of neural signals related to motor control in macaque monkey motor cortex.In order to enrich the study of association between field potentials and single unit activities,the rhesus monkeys were used as the model.Multiple neural signals were acquired by implanted electrodes in the paradigm of center-out behavior,and the neural signals were analyzed by covariance correlation to reveal the correlation between single neuron spike signals,LFP and ECoG.It was found that local field potentials and Spikes during motor tasks had good modulation relations with the motor behavior.Besides the alpha and gamma bands in the LFPs,the beta and gamma band components in the ECoG had significant correlation relationship with Spikes,indicating that these frequency band components in the field potential might reflect the deep features of motor-related neuron activities and were important for accurately analyzing the brain's motor control process.These results would help to better understand the intrinsic neural correlates,to explore the distinctive features of motor-related neural signals at different levels,and thereby may be used to establish a decoding model which is more compatible with characteristic neural activities.Chapter 4: Correlation analysis of Spike and EEG in motor cortex based on optogenetic modulation,and development of a novel flexible optoelectrode.For the correlation analysis of motor-related neuronal signals,rats were used as models.We used optogenetic methods to influence target motor cortical neuron activity,and then analysed the correlation between the Spike and EEG signals of motor cortex.It was found that the optogenetic modulation effectively changed the firing pattern of motor cortical neurons,and also had some influence on the time-frequency characteristics of EEG signals.It was found by further correlation analysis that the activation of excitatory neurons in the motor cortex by light stimulation would enhance the correlation between ?-band components in EEG signals and Spikes,indicating that the beta band in the EEG signal may be closely related to the activation of excitatory neurons in the motor cortex.In addition,a new type of flexible optoelectrode was developed,which had the flexible polydimethylsiloxane(PDMS)as the waveguide substrate and a recording site with composite conductive gel.The optoelectrode had excellent electrical and mechanical properties,and might provide a better research tool for optogenetic experiments.In summary,this study established a technical system for the decoding and correlation analysis of motor-related neural signals,achieved good classification accuracy and generalization performance results in public data sets of motor imagery based on the L1 norm regularization.Preliminary results revealed the characteristics of functional brain network of different EEG frequency bands when perform motor imagery,and found significant features of functional connectivity in EEG frequency band during the motor imagery,deepening the understanding of the brain's mechanism to perform motor imagery from the perspective of system integration.Besides,the time-frequency correlation among the spike,ECoG and EEG in the brain motor cortex was analyzed,and it was found that there was a significant correlation between spike and a specific component of EEG under the light stimulation control.This study would be helpful to theoretically analysis of the mechanism under motor related neural signals and construct effective decoding models.