Research On Recognition Of Motion Intentions Of Lower Limbs Adjacent Joints Based On Hybrid Paradigm

Motion imagery(MI)does not need actual limb movement,but is the brain’s mental thinking of different limb movements..An MI based brain computer interface(BCI)can directly convert users’ subjective motor intentions into control instructions,which not only provides a new communication medium for patients who lack actual limb movement,but also brings hope for the rehabilitation treatment of stroke patients.However,current research on MI-BCI mainly focuses on the upper limbs,and there are still fewer studies on the identification of MI at different sites in the lower limb’s adjacent joints,because the lower limb’s adjacent joint MI features have overlapping distribution in the cerebral cortex and low spatial recognition rate,resulting in insufficient identification accuracy.In order to improve the recognition rate of MI of adjacent joints of lower limbs,this paper designs to combine the steady-state somatosensory evoked potentials(SSSEP)evoked by electrical stimulation with the event-related desynchronization features(ERD)associated with MI into a mixed(hybrid,HY)paradigm to explore the identification of different MI tasks in the lower extremity neighboring joints.And the lower limb online MI-BCI is constructed to realize the on-line recognition of lower limb MI.In this paper,the experimental paradigm of somatosensory stimulation parameter screening is designed to collect to collect EEG signals from subjects at 10 different frequencies of electrical stimulation at the popliteal fossa and medial malleolus of the right leg,respectively,and use fast fourier transform(FFT)to analyze the spectral characteristics of the EEG signal in all subjects.The frequency and location that could induce distinct sssep characteristics are chosen as the optimal stimulation frequency and optimal stimulation location for electrical stimulation.The spectral characteristics of the subjects are better than those of the right leg popliteal fossa at the medial aspect of the right ankle,33 Hz being the optimal stimulation frequency.Second,we design the HY paradigm and MI paradigm to collect EEG signals from 15 subjects who perform the MI task of " foot stretching " and " knee bending " at a frequency of 33 Hz from the medial aspect of the right ankle,and comprehensive time-frequency mapping and brain topography analyses find that the characteristic ERD activation in the HY paradigm is greater,more extensive,and characterized by SSSEP than that in the MI paradigm.Filter bank common spatial pattern(FBCSP)is taken for feature extraction on EEG signals under both paradigms,and the average classification accuracy using support vector machine(SVM)and BP neural network under HY paradigm is 73.72% and 80.85%,which improves 8.08% and 11.48%compared to MI paradigm.The result confirms the feasibility of identifying different MI tasks in lower limb adjacent joints in HY paradigm.Finally,in this paper,a lower extremity online MI-BCI system is constructed under the HY paradigm using FBCSP for online feature extraction,BP neural network for online identification,and the identification results were fed back in real time.The average accuracy of online identification of 10 subjects is 79.62% and that of the highest is 92.50%,which verified the effectiveness of on-line recognition of lower limb adjacent joint MI task and promote the practical process of online MI-BCI of lower limb adjacent joints.