Decoding Forelimb Movements By Neural Spikes In Rats

Brain computer interface(BCI)is a direct connection between the brain and external devices.The BCI for motor function reconstruction can decode motor behavior by analysing neural signals in motor cortex.Then it is expected to achieve the recovery of motor function through mechanical prosthesis,neural electrical stimulation regulation,etc,which is of great significance for patients with motor disorders.The decoding of motor behavior based on neural signal is the most important part irn the BCI system for motor function reconstruction.Neural signals for motion decoding include Electroencephalography(EEG),local field potential(LFP),neuron action potential,etc.Among these signals,the neuron action potential records the firing of a single neuron and contains abundant motion-related information.Therefore,neuronal action potential is expected to achieve accurate decoding of motor behavior.However,how to realize high precision decoding of the motion with high degrees of freedom based on the neuron action potential is still a difficult problem,which needs further research.The purpose of this thesis is to realize the motion decoding based on neuron action potential.Different from the traditional method of decoding motion parameters(such as position,speed,etc.),this thesis studies the mapping between neuronal action potential and the signal from muscles.By accurately decoding muscle emission,it is expected to achieve more refined and natural motion control.Specifically,this thesis took the neuron action potential in the primary motor cortex of rats as the object,and decoded the motor process of upper limb extension and grasp of rats from the perspectives of electromyogram(EMG)regression and motion classification.The main contents and contributions of this thesis are as follows:1.Establish the decoding datasets of neuron action potential in rats.We recorded the neuron action potential and EMG in rats synchronously.Then,based on the tasks of EEG regression and action classification,we designed and normalized the regression dataset and classification dataset.In addition,the optimal selection of parameters in preprocessing of corresponding datasets is analyzed and obtained.2.Establish the mapping model between neuron action potential and EMG.We realized the regression of action potential to EMG base on LSTM,and the correlation coefficient was about 0.65.Meanwhile,classification models for the recognition of forlimb motion which were based on principal component analysis(PCA)and machine learning was built,and the recognition accuracy was about 95%.3.On the basis of the above models and methods,the online decoding system of neuron action potential is established,which realized real-time motion decoding.In conclusion,this thesis has realized the decoding of the neuron signal in motor cortex to the signal of upper limb muscles,and built a online decoding system which realized the accurate decoding of neural signals to EMG and is worthy of further study.