A Study Exploring Neural Signal Analyzing And Limb Control Mechanism Towards Brain Machine Interface

Seeking effective approaches to help disabled people reconstruct impaired motor functions is a meaningful work in both personal and social aspects.By implanting microelectrode arrays into the cerebral cortex of the users,implanted brain machine interface can extract motor control commands from the brain directly,and construct a pathway for connecting brain and equipment like computers or prosthetics when bypassing the injured spinal cord or peripheral nerves.By enabling the users control the outer equipment directly by thoughts,brain machine interface can fulfill the impaired motor functions effectively,and has been in the spotlight for years.The development of implanted brain machine interface depends not only on the innovative technologies and methods in neural signal processing and decoding,but also the deeper understanding of the work mechanisms of cortical neurons about limb motor control.This dissertation worked on both of these two sides by designing monkey behavioral experimental paradigms and in vivo neural signal acquisition protocols.On the same time,this dissertation investigated the activity of cortical neurons which devote to upper limb or lower limb motor control,and studied the possible methods for decoding motor intentions,limb electromyography and moving trajectories.Firstly,this dissertation designed a unique monkey upper limb reach-to-grasp experimental paradigm.By implanting microelectrode arrays into 3 motor control related brain areas(motor cortex,sensory cortex,and posterior parietal cortex),a large amount of data reflecting the activity features of cortical neurons directly were acquired.By leading the monkey to grasp target objects with different shapes and at different spatial locations,the relationship between the cortical neurons' activity and the two dominating factors of the movement(target locations and target shapes)was investigated.With the calculation of mutual information,such relationships were quantified,and the characteristics of the cortical neurons were elucidated.Based on the monkey reach-to-grasp experiments and corresponding in vivo multi-channel signal recordings,the problems of recognizing the monkeys' motor intentions about reaching direction and grasping gestures from cortical spike trains were investigated.An extreme learning machine based neural signal classification algorithm was employed for discriminating the neural signal patterns during the reaction time intervals when the monkey was still in preparation for the movements in the next step.For testing the online and real time performance of the movement intention detecting algorithms,a Simulink based mechanical upper limb model was constructed,and can be controlled precisely by the neural decoding results.Meanwhile,a real mechanical prosthetics control system was also constructed,and the postures of the mechanical prosthetics can also be controlled flexibly under the neural decoding based brain machine interface control framework.Since lower limb motor control was insufficiently studied at present,this dissertation designed a visually guided standing and squatting for monkey experimental paradigms and corresponding in vivo neural signal recording protocols.By using acute recordings exploring the motor cortical neurons in a wide range,a number of neural signals characterizing the lower limb voluntary motor control were collected,and the neurons which were activated during different movement stages were found.The direct innervating relationship between motor cortical neurons and lower limb voluntary movements was also validated based on the experimental data.Meanwhile,by performing the spike triggered average analysis on cortical spike trains and lower limb electromyography,the cortical neurons corresponding to each lower limb muscle innervation was categorized,and the spatial distribution density patterns of these neurons were also plotted,which were in fact spatial maps from motor cortex to lower limb muscles.Based on the lower limb standing and squatting experiments,this dissertation investigated predicting lower limb muscular activity and moving trajectories form cortical spike trains.A Kalman filter neural decoding algorithm was designed under the framework of recursive Bayesian estimation framework.The relationships between cortical spike trains and lower limb electromyography were investigated,and showing large variety and nonlinearity,which led the proposal of nonlinear neural tuning models based on artificial neural networks.An unscented Kalman filter was then designed for predicting the lower limb electromyography and moving trajectories for cortical spike trains,and the decoding accuracy was satisfactory.