| Brain computer interface(BCI)has achieved successful control of assistive devices,e.g.neuroprosthesis or robotic arm.Previous research based on grasp movements has shown limited success in precise and natural control.Natural human-computer interaction is the core of the integration of human and robot.The decoding of natural hand movement based on EEG is of great significance to the rehabilitation of hand function of patients with motor impairment and the control of rehabilitation equipment.In this study,we explored the possibilities of decoding movement types and kinematic information using movement related cortical potentials(MRCPs)for three reach-and-grasp actions: pinch,palmar and precesion disk rotaion grasp.These actions all involve two different levels of speeds and forces.The research focus of this paper can be summarized into two aspects.On the one hand,the hand movement types under each of four different kinematics conditions were decoded.On the other hand,we decoded different movement kinematics for each of three actions.The core work of this study includes:(1)we proposed the experimental research framework,built the experimental system based on the acquisition requirements of signal acquisition for natural hand movements,and developed the upper computer system to realize the communication interaction of software and hardware platform as well as visual and auditory cues for users.(2)The EEG experimental paradigm for motor-related tasks was designed,the subjects were recruited to carry out the EEG experiment,and the collected EEG data were preprocessed offline.Two classification models,contraction linear discriminant classifier(s LDA)and minimum distance to riemannian mean(MDRM),were constructed for classification research.(3)We analyzed the action behaviors of the subjects,and unified the key time points during the action execution.The MRCPs and the EEG topographic map of key brain areas were analyzed physiologically under different task conditions.(4)The decoding results based on s LDA show that the average peak accuracy of two classification and three classification could reach 83.44% and 73.83% respectively;In the case of different kinematics of classification,the average peak accuracy of two classification and four classification could reach 82.9% and 58.2% respectively.In addition,the feasibility of decoding hand motion during hand retraction was also proved.(5)At the beginning of grasping,the decoding results based on MDRM show that its decoding performance is significantly higher than the probability chance level,and the multi-classification accuracy of MDRM is significantly improved compared with s LDA.The highest result of three classification of action types is 81.76%,and the highest result of four classification of motor parameters is 75.28%.The research on natural movement decoding based on kinematics information is of great significance to the application of BCI system.Our findings provide new ideas for the fine and natural control of g neuroprosthesis or other rehabilitation devices,which will greatly improve the acceptance of users with motor impairment and contribute to the future development of BCI systems. |
