Finger Movement Decoding With Human ECoG Signals

During phases of movements intention,plan,and execution,even during the observation of others moving,the neurons in the brain generate corresponding neural signals.This thesis put efforts on decoding human finger movement from human ECoG.In this thesis:We summarized the structure of the brain,the generation of neural signal,and the recording of ECoG signal.We qualitatively compared the collection area of EEG and ECoG and the difference between them.We introduced finger movement experiments paradigm.We applied traditional feature extraction methods Principal Component Analysis(PCA),Common Spatial Pattern(CSP)in finger movement decoding.For PCA,we did both the channel-way PCA to extract the spatial information and frequency-way PCA to extract frequency spectrum information independently,for CSP we just used the channel-way method.For these three methods,the best detection results was got by frequency-way PCA features with an accuracy of 88%,and the best classification accuracy is 51% got by CSP features.We came up a rhythm energy analysis method that including steps of channel selection and decoding.We used a signed energy distance between movement and rest to select the channel.With multiple rhythms energy taken into consideration as a combined feature,we got an detection accuracy of 96% and a classification accuracy of 72%.We studied the relationship between signals from different channels and the information flow between channels during finger movements.We used mutual information(MI)to measure the functional connectivity,and transfer entropy(TE)to study the effective connectivity.The functional connectivity matrix got by MI showed the similar distribution pattern with the connectivity matrix got by covariance,correlation coefficient and phase lock value(PLV).MI gave results in bits and directly reflect the information between any two channels.The effective matrices got by TE and the time lags in cross correlation,the phase difference also share the similar pattern.They reflected the signals flow from S1 to M1 during finger movements.This signal flow was further proofed by the fact that the time of satisfied detection accuracy achieved by S1 channels was early than that got by M1 channels.