| Brain-computer interfaces(BCIs)can provide a means of communicating information from the brain to the outside world,accomplishing direct interaction with the central nervous system and peripheral devices.Motor imagery-based brain-computer interfaces(MI-BCI)are driven by the mental activity of the subject and the EEG signals are easily detectable in individuals such as healthy people and patients with neuromuscular diseases.MI-BCI currently plays an important role for brain-controlled robots,such as rehabilitation robots,nursing bed robots and unmanned aerial vehicles.Most MI-BCIs,mainly provide discrete logical control commands to peripherals,such as simple directional control intentions,which is achieved by recognising the type of limb involved in the imagined movement;however,complex and precise control commands such as force,position or velocity are not available to meet the demand for flexible motion control.Although a small number of studies based on brain-computer interfaces for motion parameters have compensated for this limitation to some extent,however,there are problems such as overall classification accuracy is not yet satisfactory,few online studies and less obvious recognition features.In this paper,the brain-computer functional network is constructed for the motorparametric imagination brain-computer interface;the topology of the brain network during motor-parametric imagination is studied,and the characteristic parameters of the brain functional network are calculated for the analysis and recognition of the motorparametric imagination task.The main work is as follows.(1)Based on the theory of complex networks,phase lag index brain functional networks were constructed for actual and imagined grip movements at four different frequency bands,Delta,Theta,Alpha and Beta.The differences in the topology of the brain networks when performing three different grip tasks of motor imagery were analysed;it was found that the topology of the brain functional networks differed for different grip value tasks,and the strength of connectivity between the nodes in the networks differed with some variability;moreover,the topology of the networks differed under different filtering frequency bands.Based on graph theory analysis,clustering coefficients,global efficiency,local efficiency,feature path length and node degree were calculated.The results show that the network attribute values of the same grip value size are different under different frequency bands;indicating that the brain is not consistent in the activity of EEG rhythm waves when processing motor processes;the network attribute values of different grip tasks are also different under the same bands;indicating that the brain has different mechanisms to process different motor task processes under the same bands.Based on the results of the graph-theoretic analysis,statistical analyses of the network attribute features were conducted.The results of the statistical analysis showed that the main significant differences in the characteristics of the network parameters when performing the actual and imagined grip tasks were the clustering coefficients and local efficiencies in the Delta and Alpha bands;and there were also significant differences in the local efficiencies in the Beta band when performing the imagined grip task.Multiple comparisons were also corrected for the network parameters between the two two grip tasks.(2)The reliability of five network parameter features in different frequency bands,the reliability of all network parameter features in different frequency bands,and the reliability of all network parameter features in different time periods in different frequency bands were analysed based on phase lag exponential networks to calculate network attribute features and to analyse the reliability of the features.The degree of reliability is evaluated by the calculated ICC values,which show that the ICC values are all different on different frequency bands,which may imply that the networks are organised differently on different frequency bands and have different functional roles,and that the network parameter features constructed based on the PLI method are reliable.Based on the network parameter features,different feature vectors are constructed,namely a combination of all network attribute features,a combination of network attribute features and adjacency matrix,and a combination of network attribute features and adjacency matrix incorporating other traditional method features;a single decoding study is conducted using a support vector machine for three types of actual or imaginary grip force(4 kg force,10 kg force,16 kg force)tasks.The results show that the recognition rate of the three types of actual or imaginary grip force by fusing the network attribute features with the adjacency matrix and traditional method features as recognition vectors achieves 83.31% ± 6.27%,indicating that the combined features have good separability effects.(3)Dynamic brain network analysis was used to study the change pattern of brain activity during motor parameter imagination,and the dynamic brain function network was constructed based on the weighted phase lag index method.The EEG time series were segmented into identical subseries according to certain non-overlapping windows;by time series,the network was constructed and features were calculated,and timevarying analysis was performed.Next,all subseries were filtered using the phasesynchronization property;the filtered valid data segments were identified and analyzed.The results of the dynamic network construction show that the average recognition rate increases to 74% ± 0.05% after phase synchronization screening;and the highest recognition rate reaches 81.25% by two-by-two decoding of the three grip motion parameter information on a single frequency band.The method based on brain network studies motor parameter imagination,hoping to increase the instruction set of the motor imagery brain-computer interface(MI-BCI),providing some inspiration for the realization of more advanced brain-controlled robots;to a certain extent,it can meet the needs of movement disorder rehabilitation. |
